AUTONOMOUS WELL CONSTRUCTION SYSTEM WITH INTEGRATED AI

Description

CROSS-REFERENCE TO RELATED APPLICATIONS

This application is a Continuation-in-Part of U.S. patent application Ser. No. 18/813,606, filed Aug. 23, 2024, entitled AUTONOMOUS DRILLING SYSTEM WITH INTEGRATED AI, issuing as U.S. Pat. No. 12,503,937 on Dec. 23, 2025 (Atty. Dkt. No. ADSS01-36136), now U.S. Pat. No. 12,503,937 issued Dec. 23, 2025. This application also claims the benefit of U.S. Provisional Application No. 63/903,682, filed Oct. 22, 2025, entitled AUTOMATED WELL FLOWBACK AND PRODUCTION SYSTEM (Atty. Dkt. No. ADSS01-36110). This application also claims the benefit of U.S. Provisional Application No. 63/927,006, filed Nov. 28, 2025, entitled AUTOMATED WELL DRILL OUT AND TESTING SYSTEM (Atty. Dkt. No. ADSS01-36133). The above-listed U.S. patent application Ser. No. 18/813,606, U.S. Pat. No. 12,503,937, U.S. Provisional Application No. 63/903,682, and U.S. Provisional Application No. 63/927,006 are incorporated by reference herein in their entireties.

TECHNICAL FIELD

The disclosure relates to drilling systems and methods for well construction activities in the oil, gas, and geothermal industries. More particularly, the disclosure pertains to an autonomous drilling system that integrates artificial intelligence, digital twin technology, and advanced sensor systems to optimize drilling operations, enhance well control, and improve overall drilling efficiency and safety.

The disclosure also relates to systems and methods for well construction, stimulation, completion, flowback, and production activities in the oil, gas, and geothermal industries. More particularly, the disclosure relates to automated systems for controlling and optimizing flowback operations including separating sand and solid debris from hydrocarbons and water produced by oil and gas wells after stimulation (i.e., fracking). Still more particularly, the disclosure relates to automated systems for controlling wellbore pressure and well flowrate during separation and removal of sand and debris from the flow stream during post-stimulation (i.e., post-fracking) flowback operations. In some embodiments, the system integrates artificial intelligence, digital twin technology, and advanced sensor systems to optimize separation operations, enhance well control and productivity, measure all fluids, solids, and gas coming from the well, and improve overall efficiency and safety during flowback and subsequent production operations.

The disclosure also relates to automated systems for controlling wellbore pressure and well flowrate during frac-plug drill-out operations. In some embodiments, the system integrates artificial intelligence, digital twin technology, and advanced sensor systems to optimize well control using managed pressure during drill-out to test well production under controlled conditions and measure all fluids, solids, and gas coming from the well, to improve overall efficiency and safety during drill-out and subsequent operations.

BACKGROUND

Drilling operations in the oil, gas, and geothermal industries have traditionally relied on human expertise and manual control systems to manage the complex process of well construction. These operations typically involve a combination of surface equipment, downhole tools, and drilling fluids working in concert to create boreholes that can extend thousands of meters into the earth. Conventional drilling systems utilize various sensors to monitor parameters such as weight on bit, rotary speed, mud flow rate, and downhole pressure, with conventional programming algorithms and/or human operators interpreting this data and making decisions about drilling parameters.

However, the conventional approach to drilling operations faces several challenges. The vast amount of data generated during drilling are not intelligently utilized by conventional programming algorithms and can overwhelm human operators, leading to suboptimal decision-making and increased risk of errors leading to inefficient and sometimes dangerous drilling situations. Additionally, the reliance on human interpretation introduces variability and potential inconsistencies in drilling practices across different wells and operators. Furthermore, the complex interactions between various drilling parameters and subsurface conditions make it difficult for human operators to consistently optimize drilling performance and maintain wellbore stability. These limitations can result in increased non-productive time, higher drilling costs, and potential safety risks.

A well that has been stimulated (i.e., fracked) with a mix of sand and frack fluid will produce significant sand and water for an initial period before producing clean hydrocarbons suitable for sending to a pipeline or production facility. These initially-produced fluids must be scrubbed of most of the returned frack sand and debris so they do not plug or damage downstream equipment. This post-frack or post-completion scrubbing is commonly known as flowback operations.

Flowback operations in the oil, gas, and geothermal industries have traditionally relied on human expertise and manual control systems to manage the complex process of receiving the initial post-fracking or post-completion flow from a well and separating large amounts of sand and other solid debris from the associated hydrocarbons and water in a way that maximizes the production potential of the well. These operations typically involve a combination of pressure and flow control equipment and separation vessels working in concert to separate sand and other solids from hydrocarbons and water flowing from the wellbore. Ideally, the well pressure and flowrate should be controlled to recover excess sand (i.e., proppant) and water from the wellbore and formation without also removing or displacing the sand needed to hold the formation open (for stimulation purposes), however, this balance is difficult to achieve in practice. Further, the produced sand and solids trapped in the separation vessels must be periodically removed (i.e., dumped) while maintaining a balance between dumping too little sand and clogging the vessel (i.e., “sanding out”) versus dumping too much sand and allowing hydrocarbons to exit via the sand port of the vessel (i.e., “blowing through”), either of which condition may disrupt the flowback operation. Conventional flowback systems are known that utilize various sensors to monitor parameters such as surface pressure and well flow rate with conventional programming algorithms and/or human operators interpreting this data and making decisions about flowback parameters.

However, the conventional approach to flowback operations faces several challenges. Often, only a fraction of the vast amount of data generated during flowback is captured. And even the data that is captured is not intelligently utilized by conventional programming algorithms and can overwhelm human operators, leading to suboptimal decision making and increased risk of errors leading to inefficient and sometimes dangerous flowback situations. Additionally, the reliance on human interpretation introduces variability and potential inconsistencies in flowback practices across different wells and operators. Furthermore, the complex interactions between various flowback parameters and subsurface conditions make it difficult for human operators to consistently optimize flowback performance to maximize desired cleanup and production without unintentionally sweeping needed proppant from the formation or otherwise damaging the reservoir. Still further, well control events are not recorded for future analysis. Yet further, recordable HSE/safety incidents or near misses are not recorded for future analysis. These limitations can result in increased non-productive time, higher flowback costs, temporary or permanent damage to the productivity of the reservoir, and potential safety risks.

A description of an automated sand dump system for oil and gas wells is disclosed in United States Patent Application Publication US2023/0313660 A1, now issued as U.S. Patent No. U.S. Pat. No. 12,209,491B2, assigned to ADS Services, Inc. U.S. Pat. No. 12,209,491, issued Jan. 28, 2025, is hereby incorporated by reference in its entirety.

A description of an autonomous drilling system with integrated AI is disclosed in U.S. patent application Ser. No. 18/813,606, now issued as U.S. Pat. No. 12,503,937, assigned to ADS Services, Inc. U.S. patent application Ser. No. 18/813,606 and U.S. Pat. No. 12,503,937 are hereby incorporated by reference in their entireties.

A description of an automated well flowback and production system is disclosed in U.S. Provisional Patent Application No. 63/903,682, assigned to ADS Services, Inc. U.S. Provisional Patent Application No. 63/903,682 is hereby incorporated by reference in its entirety.

A description of an automated well drill out and testing system is disclosed in U.S. Provisional Patent Application No. 63/927,006, assigned to ADS Services, Inc. U.S. Provisional Patent Application No. 63/927,006 is hereby incorporated by reference in its entirety

SUMMARY

This summary is provided to introduce a selection of concepts in a simplified form that are further described below in the detailed description. This summary is not intended to identify key features or essential features of the claimed subject matter, nor is it intended to be used as an aid in determining the scope of the claimed subject matter.

In one aspect, the present disclosure relates to an autonomous drilling system for well construction activities, comprising a data acquisition system configured to collect real-time data, the real-time data comprising drilling parameters from a drilling controller, and sensor data from sensors comprising flowrate sensors measuring an amount of fluid flowing through the drilling system, pressure sensors measuring a pressure of the fluid flowing through the drilling system, and cuttings sensors measuring physical characteristics of cuttings exiting the wellbore as a result of the drilling, a digital twin framework comprising a plurality of drilling models configured to simulate drilling operations based on the collected real-time data, and output predictions for the drilling operations based on the simulation, an artificial intelligence (AI) agent module configured to perform an analysis of the autonomous drilling system by aggregating the predictions from the digital twin framework, analyzing the real-time data, and analyzing historical well data, determine well health based on the analysis, and recommend adjustments to the drilling parameters, an integrated rig control system configured to automate drilling operations based on the recommended adjustments to the drilling parameters, and a human-machine interface configured to provide manual control the drilling operations based on the recommended adjustments to the drilling parameters.

In embodiments of this aspect, the disclosure according to any one of the above example embodiments, the flowrate sensors comprise at least one of a high-pressure flow meter measuring the flow of drilling fluid entering the wellbore and a low-pressure flow meter measuring the flow of drilling fluid exiting the wellbore.

In embodiments of this aspect, the disclosure according to any one of the above example embodiments, the low-pressure flow meter can be configured to measure mass flow rate of the drilling fluid in both open loop and closed loop drilling configurations.

In embodiments of this aspect, the disclosure according to any one of the above example embodiments, the sensors further comprise downhole sensors comprising at least one of pressure sensors, temperature sensors, and vibration sensors.

In embodiments of this aspect, the disclosure according to any one of the above example embodiments, the digital twin framework comprises multiple digital twins, each simulating a specific aspect of a construction process of the wellbore.

In embodiments of this aspect, the disclosure according to any one of the above example embodiments, the specific aspects simulated by the digital twins comprise at least one of drill string dynamics, cuttings circulation, wellbore geometry, and geological formations.

In embodiments of this aspect, the disclosure according to any one of the above example embodiments, the AI agent module employs deep learning or reinforcement learning techniques.

In embodiments of this aspect, the disclosure according to any one of the above example embodiments, the cuttings sensors comprise a cutting weight sensor configured to analyze volume of cuttings from the wellbore.

In embodiments of this aspect, the disclosure according to any one of the above example embodiments, the cutting sensors comprise a cutting mass sensor configured to analyze a mass flow of the cuttings from the wellbore.

In embodiments of this aspect, the disclosure according to any one of the above example embodiments, wherein the AI agent module is further configured to use data from the cuttings sensors to assess the volume of the cuttings from the wellbore.

In embodiments of this aspect, the disclosure according to any one of the above example embodiments, the AI agent module is further configured to use data from at least one of a cutting weight sensor and a cutting mass sensor to assess wellbore stability and hole cleaning efficiency.

In embodiments of this aspect, the disclosure according to any one of the above example embodiments, the integrated rig control system can be configured to transition between open loop and closed loop drilling operations based on output from the AI agent module.

In one aspect, the present disclosure relates to a method for autonomous drilling of a wellbore, comprising collecting, by a data acquisition system, real-time data comprising drilling parameters from a drilling controller, and sensor data from sensors comprising flowrate sensors measuring an amount of fluid flowing through a drilling system, pressure sensors measuring a pressure of the fluid flowing through the drilling system, and cuttings sensors measuring physical characteristics of cuttings exiting the wellbore as a result of the drilling, simulating, by a digital twin framework comprising a plurality of drilling models, drilling operations based on the collected real-time data, outputting, by the digital twin framework, predictions for the drilling operations based on the simulation, analyzing, by an artificial intelligence (AI) agent module, the autonomous drilling by aggregating the predictions from the digital twin framework, analyzing the real-time data, and analyzing historical well data, determining, by the AI agent module, well health and recommended adjustments to the drilling parameters, automating drilling operations, by an integrated rig control system, based on the recommended adjustments to the drilling parameters, and providing manual control of the drilling operations, by a human-machine interface, based on the recommended adjustments to the drilling parameters.

In embodiments of this aspect, the disclosure according to any one of the above example embodiments, the flowrate sensors comprise at least one of a high-pressure flow meter measuring the flow of drilling fluid entering the wellbore and a low-pressure flow meter measuring the flow of drilling fluid exiting the wellbore.

In embodiments of this aspect, the disclosure according to any one of the above example embodiments, the low-pressure flow meter measures mass flow rate of the drilling fluid in both open loop and closed loop drilling configurations.

In embodiments of this aspect, the disclosure according to any one of the above example embodiments, the sensors further comprise downhole sensors comprising at least one of pressure sensors, temperature sensors, and vibration sensors.

In embodiments of this aspect, the disclosure according to any one of the above example embodiments, the digital twin framework comprises multiple digital twins, each simulating a specific aspect of a construction process of the wellbore.

In embodiments of this aspect, the disclosure according to any one of the above example embodiments, the specific aspects simulated by the digital twins comprise at least one of drill string dynamics, cuttings circulation, wellbore geometry, and geological formations.

In embodiments of this aspect, the disclosure according to any one of the above example embodiments, the AI agent module employs deep learning or reinforcement learning techniques.

In embodiments of this aspect, the disclosure according to any one of the above example embodiments, the cuttings sensors comprise a cutting weight sensor configured to analyze volume of cuttings from the wellbore.

In embodiments of this aspect, the disclosure according to any one of the above example embodiments, further comprising using data from the cutting weight sensor to assess wellbore stability and hole cleaning efficiency by the AI agent module.

In embodiments of this aspect, the disclosure according to any one of the above example embodiments, further comprising transitioning between open loop and closed loop drilling operations using the integrated rig control system based on output from the AI agent module.

In further aspects, an automated well flowback and production system is disclosed including a wellhead choke, plug/junk catcher, and sand separator. The fluid outlet of the separator is connected to an automated well test manifold for controlling the flow of produced oil, gas and water. The sand outlet of the separator is connected to an automated sand dump choke for controlling the flow of produced sand. A central control unit is operably connected to the sensors and actuated devices of the system for receiving data and transmitting control signals for controlling the pressures and flows of the system during a flowback operation. The system can be configured to incorporate AI agents and digital twin frameworks to utilize the historical data and the data collected from the system to optimize the control parameters for improved flowback performance.

The flowback operation described in this patent can be utilized at various stages of the completion-during each individual plug drill-out which allows the detailed evaluation at each frac stage to understand the production profile, reservoir performance and frac efficiency by evaluating the produced fluids from the well and resultant pressure profiles, doing so in a wide spectrum of bhp profiles from over pressured to underbalanced windows. The ability to directly measure the sand and solids coming from the well during drill-out allows the comparison and evaluation of drill-out methods to better quantify the most efficient methods, i.e., coil, snubbing, hydraulic workover, drilling rig, pressure management equipment and systems requirements, etc.

The unique ability to have all the drilled data and evaluation from the drilling phase and the direct corresponding data for post completion and frac and with the support of reservoir data, digital twins, AI, and cloud computing provides the ability to better understand reserves, frac efficiency, production decline profiles etc.

The systems disclosed herein can record data regarding well control events for future analysis and for use as modeling inputs. Further, the systems disclosed herein can record data regarding HSE/safety incidents or near misses for future analysis and for use as modeling inputs. The disclosed systems' incorporation of AI agents and digital twin frameworks to utilize historical data in addition to the real-time data collected from the system can optimize the control parameters throughout the flowback operation (and related operations) to provide efficient sand management, minimize equipment erosion/wear, maximize equipment uptime, improve cleanup and transition speed, minimize formation damage, minimize environmental discharges and costs, maximalize reservoir and flowback data gathering and learning, and/or maximize initial production at hand over.

In one aspect, an automated well flowback and production system comprises a wellhead choke connected by a first wellhead flowline to a first inlet of a sand separator, the sand separator having a fluid outlet connected to an automated well test manifold for controlling the flow of produced oil, gas and water released from the sand separator. The sand separator has a sand outlet connected to a sand dump line, and an automated sand dump choke attached to the sand dump line for controlling the flow of sand released from the sand separator. An automatic block valve is attached to the sand dump line between the sand outlet and the automated sand dump choke. A central control unit is operably connected to the wellhead choke, the automated well test manifold, the automated sand dump choke, the automatic block valve, a plurality of sensors, and a plurality of actuated devices. The central control unit receives data signals from the wellhead choke, the automated well test manifold, the automated sand dump choke, the automatic block valve, the plurality of sensors, and the plurality of actuated devices, the data signals being indicative of sensed values of parameters in the wellhead flowline, the fluid line, and the sand dump line. The central control unit transmits control signals to the wellhead choke, the automated well test manifold, and the automated sand dump choke, and the automatic block valve for configuring them to control pressures and flows through the wellhead flowline, the fluid line, and the sand dump line during a flowback operation. The system is configured to incorporate AI agents and digital twin frameworks to utilize the data signals collected from the plurality of sensors and the plurality of actuated devices to optimize the control parameters for improved flowback performance.

In one embodiment, the automated well flowback and production system further comprises a first block valve installed on the first wellhead flowline between the wellhead choke and the first inlet of the sand separator. The sand separator further includes a second inlet, and a second wellhead flowline is connected between the wellhead choke and the second inlet. A second block valve is installed on the second wellhead flowline between the wellhead choke and the second inlet of the sand separator. When the first block valve is open and the second block valve is closed, fluid from the wellhead choke flows to only the first inlet, and the sand separator has a first predetermined flow capacity at a predetermined pressure. When the first block valve is closed and the second block valve is open, fluid from the wellhead choke flows to only the second inlet, and the sand separator has a second predetermined flow capacity at the predetermined pressure. When the first block valve is open and the second block valve is open, fluid from the wellhead choke flows to both the first inlet and the second inlet, and the send separator has a third predetermined flow capacity at the predetermined pressure.

In another embodiment of the automated well flowback and production system, the first block valve is an automated valve operably connected to the central control unit and the second block valve is an automated block valves operably connected to the central control unit. The plurality of sensors includes a flow rate sensor detecting the wellhead flow leaving the wellhead choke and sending signals indicative of the wellhead flow to the central control unit. The central control unit, upon receiving the signals indicative of the wellhead flow, selectively operates one or both of the first block valve and the second block valve to configure the sand separator with one of the first predetermined flow capacity, the second predetermined flow capacity, and the third predetermined flow capacity.

In another embodiment of the automated well flowback and production system, the plurality of sensors comprises first solids sensors measuring physical characteristics of solids exiting a wellbore as a result of the flowback operations.

In yet another embodiment of the automated well flowback and production system, the first solids sensors comprise a solids weight sensor configured to analyze a weight of the solids from the wellbore.

In still another embodiment of the automated well flowback and production system, the first solids sensors comprise a solids mass sensor configured to analyze a mass flow of the solids from the wellbore.

In a further embodiment of the automated well flowback and production system, the AI agent module is further configured to use data from the first solids sensors to assess at least one of: the weight of the solids from the wellbore; the volume of the solids from the wellbore; the mass of the solids from the wellbore; and the composition of the solids from the wellbore.

In a yet further embodiment of the automated well flowback and production system, the AI agent module is further configured to use data from the first solids sensors to assess sand recovery and hole cleaning efficiency.

In another embodiment of the automated well flowback and production system, the plurality of sensors comprises second solids sensors measuring physical characteristics of solids exiting the sand separator through the fluid outlet during the flowback operations.

In yet another embodiment of the automated well flowback and production system, the second solids sensors comprise a solids weight sensor configured to analyze a weight of the solids exiting the sand separator through the fluid outlet.

In still another embodiment of the automated well flowback and production system, the second solids sensors comprise a solids mass sensor configured to analyze a mass flow of the solids exiting the sand separator through the fluid outlet.

In a further embodiment of the automated well flowback and production system, the AI agent module is further configured to use data from the second solids sensors to assess at least one of: the weight of the solids exiting the sand separator through the fluid outlet; the volume of the solids exiting the sand separator through the fluid outlet; the mass of the solids exiting the sand separator through the fluid outlet; and the composition of the solids exiting the sand separator through the fluid outlet.

In a yet further embodiment of the automated well flowback and production system, the AI agent module is further configured to use data from the second solids sensors to assess an amount of sand carryover from the sand separator into the fluid outlet and a sand carryover efficiency.

In another embodiment of the automated well flowback and production system, the plurality of sensors comprises third solids sensors measuring physical characteristics of solids exiting the sand separator through the sand outlet during the flowback operations.

In yet another embodiment of the automated well flowback and production system, the third solids sensors comprise a solids weight sensor configured to analyze a weight of the solids exiting the sand separator through the sand outlet.

In still another embodiment of the automated well flowback and production system, the third solids sensors comprise a solids mass sensor configured to analyze a mass flow of the solids exiting the sand separator through the sand outlet.

In a further embodiment of the automated well flowback and production system, the AI agent module is further configured to use data from the third solids sensors to assess at least one of: the weight of the solids exiting the sand separator through the sand outlet; the volume of the solids exiting the sand separator through the sand outlet; the mass of the solids exiting the sand separator through the sand outlet; and the composition of the solids exiting the sand separator through the sand outlet.

In another aspect, an automated drill-out system is disclosed for controlling fluid production operations in an underground borehole having a movable drill string, the borehole initially comprising a plurality of borehole sections separated by successive fluid-tight frac plugs, each borehole section being in fluid communication with a different portion of a reservoir. The system comprises a data acquisition subsystem configured to collect real-time operational data, the real-time operational data comprising: drilling parameters received from a drilling controller; and sensor data received from sensors. The sensors providing sensor data comprise: one or more flowrate sensors configured to measure an amount of fluid flowing through the drill-out system; one or more solids sensors configured to measure an amount of solids flowing through the drill-out system, and one or more pressure sensors configured to measure a pressure of the fluid flowing through the drill-out system. A digital twin framework comprises a plurality of drill-out models configured to simulate drill-out operations based on the real-time operational data and to output data characterizing the simulated drill-out operations. An artificial intelligence (AI) agent module is programmed to: aggregate the data outputs from the digital twin framework; analyze the real-time operational data and historical well data; determine production characteristics of each borehole section as the associated frac plug is drilled out based on the aggregation and analyses; and compute recommended adjustments to drill-out parameters to maintain a managed pressure condition in the borehole as each frac plug is drilled out. An integrated rig control system is operatively coupled to the AI agent module and to the drill-out controller, the integrated rig control system being configured to automate the drill-out operations by implementing the recommended adjustments to the drill-out parameters. A human-machine interface is configured to present the recommended adjustments and associated system state and to enable manual control of the drill-out operations based on the recommended adjustments.

In one embodiment of the automated drill-out system, the drill-out parameters comprise at least one of: pump rate, choke position, mud density, rate of penetration, weight on bit, and drill string rotational speed.

In another embodiment of the automated drill-out system, the data output by the digital twin framework comprise at least one of: simulated pressure response, simulated flow response, simulated frac plug removal progress, simulated influx detection, and fluid composition trends.

In yet another embodiment of the automated drill-out system, the AI agent module is further programmed to compare the output data from the simulations of the digital twins to the real-time operational data, quantify a model discrepancy, and update a selection or weighting of the plurality of drill-out models based on the model discrepancy.

In still another embodiment of the automated drill-out system, the integrated rig control system is configured to transition between automated and manual modes in response to an override command received via the human-machine interface.

In a further embodiment of the automated drill-out system, the data acquisition subsystem further comprises at least one downhole sensor configured to provide pressure or temperature measurements from within the borehole.

In a yet further embodiment of the automated drill-out system, determining the production characteristics of each borehole section comprises estimating, while drilling out the associated frac plug, at least one of: fluid influx rate, gas-oil ratio, water cut, and reservoir pressure.

In a still further embodiment of the automated drill-out system, maintaining the managed pressure condition comprises controlling the drill-out parameters to keep borehole pressure within a specified pressure window relative to pore pressure and fracture gradient.

In another embodiment of the automated drill-out system, the human-machine interface is further configured to display confidence metrics associated with the recommended adjustments and to receive user selections approving, modifying, or rejecting the recommended adjustments.

In yet another embodiment of the automated drill-out system, the digital twin framework comprises physics-based models and data-driven models, and the AI agent module is configured to select between or fuse outputs of the physics-based and data-driven models based on operating conditions inferred from the real-time operational data.

In yet another embodiment of the automated drill-out system, the one or more solids sensors comprises first solids sensors measuring physical characteristics of solids exiting a wellbore as a result of the drill-out operations.

In another embodiment of the automated drill-out system, the first solids sensors comprise a solids weight sensor configured to analyze a weight of the solids from the wellbore.

In yet another embodiment of the automated drill-out system, the first solids sensors comprise a solids mass sensor configured to analyze a mass flow of the solids from the wellbore.

In still another embodiment of the automated drill-out system, the AI agent module is further configured to use data from the first solids sensors to assess at least one of: the weight of the solids from the wellbore; the volume of the solids from the wellbore; the mass of the solids from the wellbore; and the composition of the solids from the wellbore.

In a further embodiment of the automated drill-out system, the AI agent module is further configured to use data from the first solids sensors to determine a fraction of plug fragments in the solids from the wellbore.

In another embodiment of the automated drill-out system, the one or more solids sensors comprises second solids sensors measuring physical characteristics of solids exiting the sand separator through the fluid outlet during the drill-out operations.

In yet another embodiment of the automated drill-out system, the second solids sensors comprise at least one of: a solids weight sensor configured to analyze a weight of the solids exiting the sand separator through the fluid outlet; and a solids mass sensor configured to analyze a mass flow of the solids exiting the sand separator through the fluid outlet.

In still another embodiment of the automated drill-out system, the AI agent module is further configured to use data from the second solids sensors to assess at least one of: the weight of the solids exiting the sand separator through the fluid outlet; the volume of the solids exiting the sand separator through the fluid outlet; the mass of the solids exiting the sand separator through the fluid outlet; and the composition of the solids exiting the sand separator through the fluid outlet.

In a further embodiment of the automated drill-out system, the AI agent module is further configured to: use data from the second solids sensors to assess an amount of sand carryover from the sand separator into the fluid outlet; and when the amount of sand carryover exceeds a predetermined value, to reduce the flow rate through the sand separator.

In another embodiment of the automated drill-out system, the plurality of sensors comprises third solids sensors measuring physical characteristics of solids exiting the sand separator through the sand outlet during the drill-out operations.

In yet another embodiment of the automated drill-out system, the third solids sensors comprise at least one of: a solids weight sensor configured to analyze a weight of the solids exiting the sand separator through the sand outlet; and a solids mass sensor configured to analyze a mass flow of the solids exiting the sand separator through the sand outlet.

In still another embodiment of the automated drill-out system, the AI agent module is further configured to use data from the third solids sensors to assess at least one of: the weight of the solids exiting the sand separator through the sand outlet; the volume of the solids exiting the sand separator through the sand outlet; the mass of the solids exiting the sand separator through the sand outlet; and the composition of the solids exiting the sand separator through the sand outlet.

In a further embodiment of the automated drill-out system, the AI agent module is further configured to use data from the third solids sensors to assess an amount of sand removed by the sand separator and a sand separator removal efficiency.

In another aspect, an autonomous drill-out system for post-stimulation well operations, comprises a data acquisition system configured to collect real-time operational data, the real-time operational data comprising drill-out parameters from a drill-out controller, and sensor data from sensors comprising: flowrate sensors measuring an amount of fluid flowing through the drill-out system, solids sensors measuring an amount of solids flowing through the drill-out system, and pressure sensors measuring a pressure of the fluid flowing through the drill-out system. A digital twin framework comprises a plurality of drill-out models configured to simulate drill-out operations based on the collected real-time operational data, and to output predictions for the drill-out operations based on the simulations. An artificial intelligence (AI) agent module is configured to perform an analysis of the drill-out system by aggregating the predictions from the digital twin framework, analyzing the real-time operational data, and analyzing historical well data, to determine production characteristics of a wellbore section being opened by a frac-plug drill-out, and to compute recommended adjustments to drill-out parameters to maintain a managed pressure condition in the wellbore during the drill-out. An integrated drill-out control system is configured to automate the drill-out operations based on the recommended adjustments to the drill-out parameters. A human-machine interface configured to provide manual control of the drill-out operations based on the recommended adjustments to the drill-out parameters.

In one embodiment of the autonomous drill-out system, the flowrate sensors comprise at least one upstream flow meter measuring the flow of fluid entering the wellbore from the drill-out system and at least one multiphase flow meter measuring the flow of fluid and gas returning from the wellbore into the drill-out system.

In another embodiment of the autonomous drill-out system, the system further comprises a sealing system configured to selectively seal the wellbore to pressurize the wellbore to facilitate a closed loop drill-out configuration, and to unseal the wellbore to unpressurize the wellbore to facilitate an open loop drill-out configuration. The recommended adjustments to the drill-out parameters from the AI agent module to maintain a managed pressure condition during the drill-out comprise adjusting the sealing system between the pressurized closed-loop drill-out configuration having a positive surface backpressure condition and the open-loop drill-out configuration having a zero surface backpressure to maintain a desired bottom hole pressure in the wellbore during drill-out. The at least one multiphase flow meter is configured to measuring the flow of fluid and gas returning from the wellbore into the drill-out system in both the open-loop drill-out configuration when the wellbore is unpressurized and the closed-loop drill-out configuration when the wellbore is pressurized.

In yet another embodiment of the autonomous drill-out system, at least some of the solids sensors are cutting sensors measuring physical characteristics of cuttings exiting the wellbore during drill-out. The AI agent module is further configured to utilize real-time operational data from the cutting sensors to detect plug cutting onset and plug cutting completion events. When detecting one of a plug cutting onset and plug cutting completion event, the AI agent module is configured to recommended adjustments to drill-out parameters to maintain the desired managed pressure condition in the wellbore.

In still another embodiment of the autonomous drill-out system, the adjustments to the drill-out parameters recommended by the AI agent module when detecting one of a plug cutting onset and plug cutting completion events comprise at least one of: switching the drill-out parameters between an open-loop drill-out configuration and a closed-loop drill-out configuration; and initiating a post-frac dynamic flow test of the wellbore section opened by the frac-plug drill-out.

In a further embodiment of the autonomous drill-out system, the digital twin framework comprises multiple digital twins, each simulating a specific aspect of the drill-out process.

In a yet further embodiment of the autonomous drill-out system, the specific drill-out aspects simulated by the multiple digital twins comprise at least two of: wellbore hydraulics and transients; fluids and chemistry; pressure management, rig, and surface processes; bit or bottom hole assembly (BHA) mechanics for plug milling; reservoir-near-wellbore connectivity and fracture interaction; and sand production and separation.

In a still further embodiment of the autonomous drill-out system, the AI agent module is further configured to receive a pre-frac dynamic testing data set corresponding to a first plurality of wellbore formation sections and receive a post-frac dynamic testing data set corresponding to a second plurality of plug drill-out sections, The AI agent module is further configured to determine a transform to associate each group of one-or-more wellbore formation sections with a corresponding group of one-or-more plug drill-out sections based on criteria established by the AI agent module to define a respective associated pair. The AI agent module is still further configured to determine well efficiency values by comparing the respective pre-frac dynamic testing data to the respective post-frac dynamic testing data for the respective one-or-more wellbore formation sections and one-or-more plug drill-out sections of each associated pair.

In another embodiment of the autonomous drill-out system, the post-frac dynamic testing data set corresponding to the second plurality of plug drill-out sections and the pre-frac dynamic testing data set corresponding to the first plurality of formation sections are obtained from the same wellbore.

In yet another embodiment of the autonomous drill-out system, the post-frac dynamic testing data set corresponding to the second plurality of plug drill-out sections is obtained from an actual first wellbore, and the pre-frac dynamic testing data set corresponding to the first plurality of formation sections is obtained from at least one of the following: recorded data from an actual second wellbore different from the first wellbore; synthesized data based on recorded data from a plurality of actual wellbores; synthesized data based on an evaluation of geological data; and synthesized data based on an evaluation of seismic data.

In still another embodiment of the autonomous drill-out system, the criteria used by the AI agent module to associate each group of one-or-more wellbore formation sections with a corresponding group of one-or-more plug drill-out sections comprise at least one of the following: correlating the depth of the respective pre-frac and post-frac sections; correlating the geologic features of the respective pre-frac and post-frac sections; and correlating the seismic features of the respective pre-frac and post-frac sections.

In another aspect, an automated drilling system is disclosed for controlling fluid production operations in an underground borehole containing a movable drill string with a drill bit, the borehole initially comprising a plurality of borehole sections, each respective borehole section being defined by one of a respective fluid-tight frac plug or a respective reservoir well interval, each respective borehole section being in fluid communication with a different portion of a reservoir. The automated drilling the system comprises a data acquisition subsystem configured to collect real-time operational data, the real-time operational data comprising: drilling parameters received from a drilling controller; and sensor data received from sensors. The sensors comprise one or more flowrate sensors configured to measure an amount of fluid flowing through the drilling system; one or more solids sensors configured to measure an amount of solids flowing through the drilling system; and one or more pressure sensors configured to measure a pressure of the fluid flowing through the drilling system. The automated drilling the system further comprises a digital twin framework comprises a plurality of drilling models configured to simulate drilling operations based on the real-time operational data and output data characterizing the simulated drilling operations. The automated drilling the system further comprises an artificial intelligence (AI) agent module programmed to: aggregate the data outputs from the digital twin framework; analyze the real-time operational data and historical well data; determine production characteristics of each borehole section when the borehole section is drilled out based on the data aggregation and analyses; and compute recommended adjustments to drilling parameters to maintain a managed pressure condition in the borehole when each borehole section is drilled out. When the borehole section is defined by a respective frac plug, the respective borehole section is drilled out when the respective frac plug is drilled out by the drill bit, and when the respective borehole section is defined by a respective reservoir well interval, the respective borehole section is drilled out when the respective reservoir well interval is drilled out by the drill bit. The automated drilling the system further comprises an integrated rig control system operatively coupled to the AI agent module and to the drilling controller, the integrated rig control system being configured to automate the drilling operations by implementing the recommended adjustments to the drilling parameters. The automated drilling the system further comprises a human-machine interface configured to present the recommended adjustments and associated system state and to enable manual control of the drilling operations based on the recommended adjustments.

In one embodiment of the automated drilling system, the drilling parameters comprise at least one of: pump rate, choke position, mud density, rate of plug penetration, weight on bit, and drill bit rotational speed.

In another embodiment of the automated drilling system, the data output by the digital twin framework comprise at least one of: simulated pressure response, simulated flow response, simulated frac plug removal progress, simulated influx detection, and fluid composition trends.

In still another embodiment of the automated drilling system, the AI agent module is further programmed to compare the output data from the simulations of the digital twins to the real-time operational data, quantify a model discrepancy, and update a selection or weighting of the plurality of drill-out models based on the model discrepancy.

In yet another embodiment of the automated drilling system, the integrated rig control system is configured to transition between automated and manual modes in response to an override command received via the human-machine interface.

In a further embodiment of the automated drilling system, the data acquisition subsystem further comprises at least one downhole sensor configured to provide pressure or temperature measurements from within the borehole.

In a still further embodiment of the automated drilling system, determining the production characteristics of each borehole section comprises estimating, while drilling out the borehole section, at least one of: fluid influx rate, gas-oil ratio, water cut, and reservoir pressure.

In a yet further embodiment of the automated drilling system, maintaining the managed pressure condition comprises controlling the drilling parameters to keep borehole pressure within a specified pressure window relative to pore pressure and fracture gradient.

In another embodiment of the automated drilling system, the human-machine interface is further configured to display confidence metrics associated with the recommended adjustments and to receive user selections approving, modifying, or rejecting the recommended adjustments.

In still another embodiment of the automated drilling system, the digital twin framework comprises physics-based models and data-driven models, and the AI agent module is configured to select between or fuse outputs of the physics-based and data-driven models based on operating conditions inferred from the real-time operational data.

In yet another embodiment of the automated drilling system, the one or more solids sensors comprises first solids sensors measuring physical characteristics of solids exiting a wellbore as a result of the drill out operations.

In a further embodiment, the AI agent module is further configured to use data from the first solids sensors to assess at least one of: the weight of the solids from the wellbore; the volume of the solids from the wellbore; the mass of the solids from the wellbore; or the composition of the solids from the wellbore.

In another aspect, an autonomous drilling system for post-stimulation well operations comprises a data acquisition system configured to collect real-time operational data, the real-time operational data comprising: drilling parameters from a drilling controller, and sensor data from sensors. The sensors comprise flowrate sensors measuring an amount of fluid flowing through the drilling system, solids sensors measuring an amount of solids flowing through the drilling system, and pressure sensors measuring a pressure of the fluid flowing through the drilling system. The autonomous drilling system further comprises a digital twin framework comprising a plurality of drilling models configured to simulate drilling operations based on the collected real-time operational data, and to output predictions for the drilling operations based on the simulations. The autonomous drilling system further comprises an artificial intelligence (AI) agent module configured to: perform an analysis of the drilling system by aggregating the predictions from the digital twin framework, analyzing the real-time operational data, and analyzing historical well data; determine production characteristics of a wellbore section being opened by a drill bit drilling into the wellbore section, wherein the wellbore section is defined by one of a respective fluid-tight frac plug or a respective reservoir well interval; and compute recommended adjustments to drilling parameters to maintain a managed pressure condition in the wellbore during the drilling into the wellbore section. The autonomous drilling system further comprises an integrated drilling control system configured to automate the drilling operations based on the recommended adjustments to the drilling parameters. The autonomous drilling system further comprises a human-machine interface configured to provide manual control of the drilling operations based on the recommended adjustments to the drilling parameters.

In one embodiment of the autonomous drilling system, the flowrate sensors comprise at least one upstream flow meter measuring the flow of fluid entering the wellbore from the drilling system and at least one multiphase flow meter measuring the flow of fluid and gas returning from the wellbore into the drilling system.

In another embodiment of the autonomous drilling system, at least some of the solids sensors are cutting sensors measuring physical characteristics of cuttings exiting the wellbore during drilling a borehole section defined by a respective frac plug, the AI agent module is further configured to utilize real-time operational data from the cutting sensors to detect plug cutting onset and plug cutting completion events, and when detecting one of a plug cutting onset and plug cutting completion event, the AI agent module is configured to recommended adjustments to drilling parameters to maintain the desired managed pressure condition in the wellbore.

In still another embodiment of the autonomous drilling system, the adjustments to the drilling parameters recommended by the AI agent module when detecting one of a plug cutting onset and plug cutting completion events comprise at least one of adjustment to managed pressure condition in the wellbore, and initiating a post-frac dynamic flow test of the wellbore section opened by the frac-plug drill-out.

In yet another embodiment of the autonomous drilling system, the digital twin framework comprises multiple digital twins, each simulating a specific aspect of the drilling process.

In a further embodiment of the autonomous drilling system, the specific drilling aspects simulated by the multiple digital twins comprise at least two of: wellbore hydraulics and transients; fluids and chemistry; pressure management, rig, and surface processes; bit or bottom hole assembly (BHA) mechanics for plug milling; reservoir-near-wellbore connectivity and fracture interaction; or sand production and separation.

In a still further embodiment of the autonomous drilling system, the AI agent module is further configured to: receive a pre-frac dynamic testing data set corresponding to a first plurality of borehole formation sections; receive a post-frac dynamic testing data set corresponding to a second plurality of wellbore sections; determine a transform to associate each group of one-or-more borehole formation sections with a corresponding group of one-or-more wellbore sections based on criteria established by the AI agent module to define a respective associated pair; and determine well efficiency values by comparing the respective pre-frac dynamic testing data to the respective post-frac dynamic testing data for the respective one-or-more borehole formation sections and one-or-more wellbore sections of each associated pair.

In a still further embodiment of the autonomous drilling system, the post-frac dynamic testing data set corresponding to the second plurality of wellbore sections and the pre-frac dynamic testing data set corresponding to the first plurality of formation borehole sections are obtained from the same wellbore.

In a yet further embodiment of the autonomous drilling system, the criteria used by the AI agent module to associate each group of one-or-more formation borehole sections with a corresponding group of one-or-more wellbore sections comprise at least one of the following: correlating the depth of the respective pre-frac and post-frac sections; correlating the geologic features of the respective pre-frac and post-frac sections; and correlating the seismic features of the respective pre-frac and post-frac sections.

BRIEF DESCRIPTION OF THE DRAWINGS

So that the way the above-recited features of the present disclosure can be understood in detail, a more particular description of the disclosure, briefly summarized above, can be made by reference to example embodiments, some of which are illustrated in the appended drawings. It is to be noted, however, that the appended drawings illustrate only example embodiments of this disclosure and are therefore not to be considered limiting of its scope, for the disclosure can admit to other equally effective example embodiments.

FIG. 1A illustrates a diagram of a drilling rig system, according to aspects of the present disclosure;

FIG. 1B illustrates a high-side Venturi meter for measuring drilling fluid flow rate, according to an embodiment;

FIG. 1C illustrates a low-side Venturi meter for measuring drilling mud flow rate, according to an embodiment;

FIG. 1D illustrates a diagram of a mud cuttings measurement system, according to aspects of the present disclosure;

FIG. 2A illustrates a flowchart for an overall operational process of an autonomous drilling system, according to an embodiment;

FIG. 2B illustrates a flowchart for operator actions in an autonomous drilling system, according to an embodiment;

FIG. 3 illustrates a network diagram for an autonomous drilling system, according to aspects of the present disclosure;

FIG. 4 illustrates a block diagram of a neural network module for an autonomous drilling system, according to an embodiment;

FIG. 5 illustrates a front view of an HMI display for an autonomous drilling system, according to aspects of the present disclosure;

FIG. 6 is a schematic illustration of an automated well flowback and production system in accordance with another aspect;

FIG. 7A is a flowchart for an overall operation process of an automated well flowback and production system in accordance with another aspect;

FIG. 7B is a flowchart for operation actions in an automated well flowback and production system in accordance with another aspect; and

FIG. 8 is a flowchart of a process for automated wellbore annulus pressure relief in accordance with another aspect;

FIG. 9 is a flowchart of a process for automated management one or more sand separators (SKUs) collectively having multiple inlets in accordance with another aspect;

FIG. 10 is a schematic illustration of an automated frac plug drill-out system in accordance with another aspect;

FIGS. 11A and 11B provide a flowchart illustrating a method for automated well evaluation and determination of quantitative measures of well construction performance in accordance with another aspect; and

FIG. 12 is a schematic illustration of an agentic AI automated drill-out system in accordance with another aspect.

DETAILED DESCRIPTION

Various example embodiments of the present disclosure will now be described in detail with reference to the drawings. It should be noted that the relative arrangement of the components and steps, the numerical expressions, and the numerical values set forth in these example embodiments do not limit the scope of the present disclosure unless it is specifically stated otherwise. The following description of at least one example embodiment is merely illustrative in nature and is in no way intended to limit the disclosure, its application, or its uses. Techniques, methods, and apparatus as known by one of ordinary skill in the relevant art cannot be discussed in detail but are intended to be part of the specification where appropriate. In the examples illustrated and discussed herein, any specific values should be interpreted to be illustrative and non-limiting. Thus, other example embodiments can have different values. Notice that similar reference numerals and letters refer to similar items in the following figures, and thus once an item is defined in one figure, it is possible that it need not be further discussed for the following figures. Below, the example embodiments will be described with reference to the accompanying figures.

The following description sets forth examples of the present disclosure. It should be recognized, however, that such description is not intended as a limitation on the scope of the present disclosure. Rather, the description also encompasses combinations and modifications to those examples described herein.

The present disclosure provides an autonomous drilling system designed to optimize well construction operations in the oil, gas, and geothermal industries. This system integrates a data acquisition system, a digital twin framework, an artificial intelligence (AI) agent module, an integrated rig control system, and an optional human-machine interface (HMI).

In some aspects, the autonomous drilling system may include a version of the cloud infrastructure located on-site at the well or rig, enabling local processing and storage capabilities. Additionally, the system may be configured to connect to off-site infrastructure, such as cloud services or other remote systems, to support off-site learning, updates, and modifications to digital twin models and well plans. This hybrid architecture allows for remote work and collaboration while ensuring that the system can operate independently without an internet connection. The ability to function autonomously, without relying on external connectivity, is beneficial for the system's commercial viability and scalability, particularly in remote or challenging environments where internet access may be limited or unreliable.

The data acquisition system can collect real-time data from surface and downhole sensors, including but not limited to flow meters and cutting weight sensors. This data can then be preprocessed to clean, filter, and format it for compatibility with the system.

In one example, the digital twin framework can create sophisticated virtual models of the wellbore and drilling operations. These models can be continuously updated with real-time data, allowing for accurate simulations of various components and scenarios. In some aspects, the system may also collect external data inputs from the rig and external service companies. For instance, the system may incorporate directional drilling data from specialized service providers. This integration of external data sources may enhance the accuracy and comprehensiveness of the digital twin models, potentially enabling more precise simulations and predictions for the drilling operations.

The autonomous drilling system may cover all well construction operations, including drilling, tripping in/out, circulating, reaming, casing/tubing running, and cementing. These operations can be performed using both open and closed loop methods. Closed loop drilling methods may include Managed Pressure Drilling (MPD) and Under Balanced Drilling (UBD) methods. The system is applicable to various types of drilling rigs, including rotary rigs (both land/offshore dry BOP stack rigs and offshore floating rigs with subsea BOPs and drilling risers) and coil tubing rigs (also known as COIL rigs). While the equipment lineups may differ slightly, the technology and methods are applicable to various rig types, ensuring versatility and broad applicability across different well construction scenarios.

In some aspects, the digital twin models within the autonomous drilling system may be designed to continuously generate forward-looking data covering numerous potential scenarios. These models may operate at high speeds locally, leveraging the on-site processing capabilities of the system. By constantly or periodically creating and updating simulations for various possible drilling conditions and outcomes, the system may be able to anticipate and prepare for a wide range of situations in real-time. This approach may enable the AI agent module to make rapid, informed decisions and recommendations, enhancing the system's ability to optimize drilling operations and respond proactively to changing well conditions.

In one example, the AI agent module can analyze the data and use physics models to predict potential issues. It can also integrate the outputs from multiple digital twins to detect complex patterns and interactions. In other words, agents are able to learn from historical data and simulation data which is described in more detail below.

Overall, the integrated rig control system can automate drilling operations based on the output (i.e. recommendations and/or instructions) from the AI agents and the real-time optimization engine. This system can allow for seamless transition between open loop and closed loop drilling configurations, enhancing the system's ability to adapt to changing well conditions without interruption of well construction. The HMI can provide a user-friendly interface for rig operators to monitor real-time data, receive AI output (i.e. recommendations and/or instructions), and manually intervene in the drilling process if desired.

Together, these components work in harmony to continuously monitor and control well conditions, leading to safer, faster, and more efficient well construction. The system's comprehensive well health monitoring capabilities can enable real-time assessment of wellbore stability, fluid dynamics, and formation characteristics. By integrating data from multiple sensors and digital twin simulations, the AI agent module can detect subtle changes in well health indicators, allowing for proactive adjustments to drilling parameters to maintain improved (e.g. optimal) well conditions throughout the construction process.

More specifically, as will be described in more detail, the described system and method leverages cloud-based learning, utilizing a vast historical well data repository to continuously improve machine learning algorithms. Real-time AI-driven optimization can combine live data, physics models, and AI to enhance drilling parameters for safer, more efficient well construction. Digital twin integration can provide real-time simulations of wellbore components, allowing the AI to anticipate and address potential issues proactively.

The system can operate autonomously based on the drilling plan, AI output (i.e. recommendations and/or instructions) while allowing human operators to intervene and maintain control when beneficial. Secure cloud connectivity can ensure safe data transfer between the wellsite and cloud-based learning platform. Safety capabilities can be enhanced through new mass flow meters and mud cutting measurement systems, providing a clearer understanding of well state and potential dysfunctions.

The system can monitor and report on various well dysfunctions, including but not limited to well control events, loss of circulation, stuck pipe, wellbore instability, and drilling vibrations cuttings buildup in the well, and well breathing. It can also identify bit wear and condition using detection of alkene (cracked) gases in return mud. Additionally, the system may monitor and report on other well dysfunctions such as formation fluid influx, differential sticking, and borehole washouts. It also can simplify output for drillers and supervisors using a traffic light system with corresponding output (i.e. recommendations and/or instructions) for action.

The system can continuously monitor both open hole or closed loop pressurized or non-pressurized drilling operations. This can be achieved through various sensors including a return flow venturi-type flow meter, which allows for seamless switching between open and closed hole drilling without loss of data or time. The addition of a high-pressure venturi-type mass flow meter on the input side of the well circulation system can provide a scalable and economic method for continuous mud system input measurement. It is noted that switching between open and closed hole drilling may be based on the drilling plan, manual intervention, AI instructions, AI recommendations or a combination of two or more of the above mechanisms.

The system can also incorporate cuttings measurement system data into digital twin models and AI-supported control systems. This can provide insights into hole conditions, potential collapses, poor cuttings transport/buildup and helps prevent issues like stuck pipe or circulation problems. It can also ensure accurate cement volume calculations for casing operations.

Continuous input of true mass flow into and out of the well in both closed and open systems, combined with cuttings measurements can maintain visibility and control of the well and circulating system. This real-time data can support the operation of the system effectively.

The details of the system will now be described with respect to the figures, which provide visual representations of various components and processes of the autonomous drilling system. These figures illustrate the drilling rig system, flow meters, mud cuttings measurement system, operational processes, network architecture, neural network module, and HMI, offering a comprehensive view of the system's structure and functionality. Through these illustrations, the workings of the autonomous drilling system and its components will be elucidated, providing a clear understanding of how the system optimizes well construction operations (i.e. activities).

Referring to FIG. 1A, the diagram illustrates an example drilling rig system 100 with components on both high-pressure and low-pressure sides of the drilling fluid circulation system. On the high-pressure side, the system can include a mud pump 106, a high-side pressure sensor (PSH), a high-side Venturi meter (VMH), a standpipe 108, and a rotary hose 110. These components are responsible for pumping and monitoring the drilling fluid as it enters the wellbore. Unless otherwise specified, in this disclosure the terms “wellbore” and “borehole” are used interchangeably, as it is assumed that the boreholes described herein are intended for connection into a well system.

The low-pressure side of the circulation system can include a choke 115, a low-side pressure sensor (PSL), a low-side Venturi meter (VML), a shale shaker 124, and mud tanks 126. These components can handle the drilling fluid as it returns from the wellbore. In some aspects, the mud tanks 126 can include a mud weight sensor (MWS) that measures physical characteristics of the cuttings and cuttings disposal 127 that receives the cuttings from the shale shaker and MWS. This sensor can provide real-time data on properties such weight of the drilling mud and cuttings, enabling continuous monitoring and analysis of the drilling process.

In general, the drilling fluid circulation process begins on the high-pressure side, where the mud pump 106 draws fluid from the mud tanks 126 and pumps it through the high-side pressure sensor PSH and VMH. The fluid then flows through the standpipe 108 and rotary hose 110 into the top drive 112 and downhole. As the fluid returns to the surface, it passes through the RCD and BOP 114 and enters the low-pressure side of the circulation system. In open loop drilling, the RCD may operate without a bearing assembly, effectively open to atmosphere. In closed loop drilling, the RCD may utilize a bearing assembly to create a seal around the drill pipe. In the closed loop configuration, the fluid then flows through a choke 115, which can be used to control the backpressure in the wellbore. After the choke 115, the fluid passes through the low-side pressure sensor PSL and VML, which measure the returning fluid's properties. In the open loop configuration, the choke is bypassed and the fluid flows through the VML. In either case, the fluid then passes over the shale shaker 124, which removes larger cuttings and solids and dumps them onto the MWS for weight monitoring. The cuttings are disposed of in cuttings disposal 127 while cleaned fluid returns to the mud tanks 126, completing the circulation cycle.

In some aspects, the autonomous drilling system may incorporate both MPD and UBD operations as well as traditional well control operations. MPD/UBD operations may utilize a choke to control wellbore pressure during normal drilling activities, while traditional well control operations may employ a separate choke manifold when dealing with more severe influxes or well control events. The system may be designed to seamlessly transition between these operational modes as needed.

In MPD/UBD operations, the return flow may pass through the RCD and then to a dedicated MPD/UBD choke manifold. However, in traditional well control scenarios, at least one of BOP elements may be closed, and the return flow may be directed through a separate well control choke manifold. In some implementations, the system may allow for return flow in a closed system to be routed to a pressure-controlling choke from multiple sources. This may include flow from the RCD, from below the RCD, or from an outlet below a closed BOP sealing element (such as an annular preventer or ram) via lines 122 in FIG. 1A. This flexibility may enable the system to handle a wide range of well control scenarios efficiently.

The fluid continues down the drill pipe 116, passes through down hole measurement sensors 121 until it reaches the bit 120 at the bottom of the wellbore. At this point, the fluid exits through nozzles in the bit, cleaning the cutting surface and cooling the bit. The fluid then carries the drill cuttings up the annular space (well anulus 119) between the drill pipe 116 and the wellbore wall, passing up past the down hole measurement sensors, drill collars 118 and drill pipe 116 towards the surface. In some cases, the system may incorporate Mud Gas Separators (MGS) if there is wellbore gas entrained in the return mud and cuttings flow. The MGS can be used to separate and safely handle any gas present in the drilling fluid before it reaches the shale shakers.

It is noted that the downhole measurement sensors within the drill string may measure conditions in the annulus and wellbore. These sensors may include Pressure While Drilling (PWD) devices, which measure pressure and potentially temperature at specific depths in the wellbore. While some sensors may read conditions internal to the drill string, PWD measurements may be beneficial for calibrating digital twin models with real-time data. This integration of downhole sensor data with surface measurements and digital simulations may enhance the system's ability to accurately model and respond to changing wellbore conditions throughout the drilling process.

As the fluid returns to the surface, it passes through the BOP and then the RCD and enters the low-pressure side of the circulation system. The low-side pressure sensor PSL may measure the surface back pressure before the choke. In some cases, during closed loop mode, an additional pressure sensor may be placed after the choke, enabling measurement of the pressure differential across the choke(s) in operation. It is noted that in a closed loop mode, before reaching the low-pressure side, the fluid may pass through a choke Manifold Assembly. This component can be used to control the wellbore pressure by adjusting the backpressure applied to the returning fluid. The choke Manifold Assembly may include multiple chokes (like choke 115) and valves that can be automatically or manually adjusted to maintain the desired wellbore pressure profile. In either case, the fluid then flows through the low-side pressure sensor PSL and VML, which measure the returning fluid's properties. The fluid then passes over the shale shaker 124, which removes larger cuttings and solids and dumps them onto the MWS for weight monitoring. The cuttings are disposed of in cuttings disposal 127 while cleaned fluid returns to the mud tanks 126, completing the circulation cycle.

This continuous circulation of drilling fluid helps to remove cuttings from the wellbore, cool and lubricate the drill bit, and maintain wellbore stability. The various sensors and meters on both the high-pressure and low-pressure sides of the system can allow for real-time monitoring of fluid properties and flow rates, which is beneficial for maintaining efficient and safe drilling operations. In some aspects, the sensors such as the flow rate sensors measure the velocity and/or amount (e.g. volume) of fluid flowing through the system. In addition, the sensors may also measure additional fluid properties such as density, temperature, and viscosity, providing a comprehensive characterization of the drilling fluid throughout the circulation system.

In addition to the sensors described in the drilling rig system 100, other sensors can also be present in the rig, either on the surface or downhole, to provide comprehensive monitoring and data collection for the autonomous drilling system. These additional sensors can include, but are not limited to, torque sensors, weight-on-bit sensors, rotary speed sensors, and directional sensors. For example, torque sensors can be installed on the top drive or drill string to measure the rotational force applied during drilling, while weight-on-bit sensors can be placed near the drill bit to measure the axial force applied to the formation.

It should be noted that in some aspects, top drive 112 may be hoisted up and down to perform drilling operations. This vertical movement may be accomplished by a drawworks 101, which can be a large motor powered winch-like device that controls the movement of the drilling equipment. The drawworks 101 may pull on a cable 103, which can be routed through a pulley system 105 mounted at the top of the derrick. By controlling the rotation of the drawworks 101, the system may raise or lower the top drive 112, allowing for precise control of the drill string's vertical position during drilling operations. This hoisting mechanism may enable the drilling system to adjust the weight on bit, make connections, and perform tripping operations as needed throughout the well construction process.

In some aspects, the rig can be equipped with downhole sensors integrated into the bottom hole assembly (BHA) or distributed along the drill string. These can include formation evaluation sensors such as gamma ray detectors, resistivity sensors, or neutron porosity sensors, which provide real-time data on the geological formations being drilled. Additionally, the system can incorporate downhole vibration sensors, acoustic sensors, or strain gauges to monitor drill string dynamics and wellbore conditions. In some cases, downhole pressure-while-drilling (PWD) sensors can be used to measure annular pressure and detect potential kicks or losses more accurately than surface measurements alone.

The rig controller 102 can play a role in managing and controlling the drilling rig system 100. It can receive data from various sensors throughout the rig, including the high-side pressure sensor PSH, VMH, low-side pressure sensor PSL, VML, and mud weight cuttings sensor MWS. In some cases, the rig controller 102 can also receive data from additional sensors such as torque sensors, weight-on-bit sensors, rotary speed sensors, and downhole sensors integrated into the bottom hole assembly.

In embodiments of this aspect, the rig controller 102 can be connected to the sensors and actuators of the rig through a combination of wired and wireless communication networks. The controller can receive data from various sensors, including surface sensors like pressure sensors and flow meters, as well as downhole sensors integrated into the bottom hole assembly, through data acquisition systems and telemetry networks. It can process this data in real-time and send control signals to actuators such as the drawworks, choke, mud pump, top drive, and rotating control device, adjusting drilling parameters based on the analyzed sensor data and AI output (i.e. recommendations and/or instructions). The controller can utilize industrial communication protocols to ensure reliable and secure data exchange between the control system and the rig's sensors and actuators.

The rig controller 102 can process sensor data in real-time, using it to monitor the drilling operation and make automated adjustments to various rig components. For example, it can control the mud pump 106 to adjust the flow rate of drilling fluid or modify the speed and torque of the top drive 112 based on the current drilling conditions. In some aspects, the rig controller 102 can manage the operation of the BOP 114, adjusting its settings based on the detected wellbore pressure and flow conditions. The BOP 114 may include seal rams that can be opened and closed during any operation at any time to secure or close the well if there is a well control issue. The RCD 114 may be configured in multiple states such as: either in use with a mechanical seal around the drill pipe, or without a seal/bearing assembly, effectively open to atmosphere. The rig controller 102 may monitor the state of the RCD 114 but may or may not adjust the RCD during active drilling operations. The RCD configuration may be set prior to beginning a particular drilling phase or operation.

The rig HMI 104 can serve as the primary interface between the operator and the autonomous drilling system. Through the rig HMI 104, the operator can view real-time data from sensors, monitor the status of various rig components, and observe output (i.e. recommendations and/or instructions) generated by the AI agent module (described in more detail later). In some cases, the rig HMI 104 can display graphical representations of the drilling operation, such as wellbore trajectory, formation data, and drilling parameter trends.

The operator can use the rig HMI 104 to input commands and adjust drilling parameters manually when desired. For instance, the operator can override automated settings, initiate specific drilling procedures, or respond to alerts and warnings generated by the system. In some aspects, the rig HMI 104 can allow the operator to switch between different operational modes, such as transitioning from automated drilling to manual control for specific operations or in response to unexpected situations.

To further elucidate the operation of the autonomous drilling system, a specific use case will now be described. This example will demonstrate how the various components of the system work together in a real-world scenario, illustrating the system's ability to detect and respond to potential drilling issues, optimize performance, and maintain wellbore stability. By walking through this use case, a clearer understanding is gained of how the system's data acquisition, digital twin simulations, AI analysis, and automated control functions interact to enhance drilling efficiency and safety.

In a specific use case, the operator can begin by inputting a well construction plan through the rig HMI 104 or the well construction plan may be received through other data channels. This plan can include details such as the target depth, expected formation characteristics, and planned trajectory of the wellbore. Once the plan is loaded, the digital twin framework incorporates this information into its models, creating a virtual representation of the planned well.

As drilling begins, the autonomous drilling system continuously monitors the well construction process using its array of sensors. The high-side pressure sensor PSH and VMH can measure the pressure and flow rate of the drilling fluid entering the wellbore, while the low-side pressure sensor PSL and VML monitor the returning fluid. The MWS can analyze the cuttings, providing real-time data on the formations being drilled.

It is noted that there is time lag before the cuttings reach the surface, ranging from a few minutes to more than an hour depending on the well depth. This lag time may be taken into account when interpreting the data from the MWS. Additionally, the MWS may sometimes measure materials from collapsed sections of the well, which can occur if wellbore stability is not maintained. Therefore, the system may also consider these factors when analyzing the cuttings data and correlating it with other sensor inputs to provide an accurate representation of the current downhole conditions.

In some aspects, the autonomous drilling system may incorporate a digital image analysis system (not shown) that can examine the cuttings coming off the shale shakers. This imaging technology may provide beneficial insights into the type of rock being drilled, the effectiveness of the drill bit, and potentially identify well collapse material. By analyzing the size, shape, and composition of the cuttings in real-time, the system may offer a more detailed understanding of the downhole conditions, enabling improved decision-making and potentially early detection of wellbore stability issues. This visual analysis capability may complement the existing sensor data, further enhancing the system's ability to optimize drilling operations and maintain wellbore integrity.

In this scenario, the system can detect a gradual increase in the weight of cuttings measured by the MWS, coupled with a slight decrease in the return flow rate measured by the VML. The AI agent module, analyzing this data in conjunction with the digital twin simulations, can identify a potential risk of hole cleaning issues and formation instability. It is noted that cleaning efficiency may be based on Equivalent Circulating Density (ECD), as excess cuttings in the wellbore increase return mud density, leading to higher back pressure and ECD, which in turn affects the overall well pressure profile at depth.

Based on this analysis, the AI agent module can recommend adjustments to the drilling parameters to mitigate the developing problem. These output (i.e. recommendations and/or instructions) can include increasing the mud flow rate to improve hole cleaning efficiency, adjusting the rotary speed to optimize cuttings transport, and modifying the mud properties to enhance wellbore stability.

It is noted that the ability to adjust the surface choke is beneficial in managing wellbore pressure and maintaining well control. The surface choke, which may be part of the choke manifold, can be opened or closed to increase or decrease the back pressure on the well. This mechanism plays a beneficial role in controlling the bottomhole pressure and overall wellbore pressure profile and hence can help in managing the wellbore stability. In practice, closing the surface choke restricts the flow of drilling fluid returning from the wellbore, which increases the back pressure in the system. This increased back pressure is transmitted down the wellbore, effectively increasing the bottomhole pressure. Conversely, opening the choke reduces the restriction on the return flow, decreasing the back pressure and, consequently, the bottomhole pressure. By manipulating the choke position, the drilling team can respond to changing downhole conditions, such as influxes or losses, in real-time. This capability is particularly beneficial in managed pressure drilling operations, where precise control of the wellbore pressure profile is beneficial for maintaining well stability and preventing kicks or lost circulation events.

The integrated rig control system, upon receiving these output (i.e. recommendations and/or instructions), can automatically implement the changes. For instance, it can increase the speed of the mud pump 106 to boost the mud flow rate, adjust the rotation speed of the top drive 112, control drawworks 101 to add or reduce weight of top drive 112, close the choke to increase annulus pressure, and signal for adjustments to the mud composition in the mud tanks 126.

Throughout this process, the operator can monitor the situation through the rig HMI 104, observing the real-time data, AI output (i.e. recommendations and/or instructions), and automated adjustments. The performance chart on the HMI display can show the changing trends in drilling parameters and predicted outcomes. If desired, the operator can use the well control and output (i.e. recommendations and/or instructions) to manually fine-tune the adjustments or override the automated actions. As a result of these proactive adjustments, the system can prevent the development of a serious hole cleaning problem or wellbore control and/or instability issue, allowing the drilling operation to continue safely and efficiently.

The autonomous drilling system can detect and compensate for a wide range of well construction issues, including but not limited to wellbore instability, hole cleaning problems, formation fluid influx (kicks), well breathing and ballooning, lost circulation, stuck pipe incidents, drill string vibrations, bit wear, formation pressure changes, wellbore trajectory deviations, and mud property imbalances. By continuously analyzing real-time data from surface and downhole sensors, the system can identify early signs of these issues and implement corrective actions. For example, it can adjust drilling parameters such as weight on bit, rotary speed, and mud flow rate to optimize drilling efficiency and maintain wellbore stability. The system can also modify mud properties (e.g., either automatically by injecting chemicals or changing fluid sources, or by providing modification instructions to the crew), adjust the wellbore trajectory, or initiate specialized drilling procedures to address specific challenges encountered during the well construction process.

Well control covers various types of well control. In one example, well control may be Managed Pressure Drilling (MPD) which is an adaptive drilling process that controls the annular pressure profile throughout the wellbore. This technique involves using specialized equipment and techniques to manage the bottomhole pressure within a narrow window between the pore pressure and fracture pressure of the formation. MPD systems may be particularly beneficial in challenging drilling environments, such as deepwater operations, high-pressure/high-temperature wells, or formations with narrow drilling windows.

In MPD operations, a closed-loop circulation system may be used, often including the RCD to seal the annulus. This setup allows for the application and adjustment of surface back pressure in real-time using automated chokes. The system may enable continuous monitoring of downhole pressures and fluid flows to detect and respond to minute changes in wellbore conditions. By manipulating surface back pressure, pump rates, and fluid properties, MPD systems may allow for precise control of the equivalent circulating density (ECD).

One of the advantages of MPD is its ability to enable drilling through narrow pressure margins or depleted zones that might be challenging with conventional methods. This capability may improve safety, reduce non-productive time, reduce well construction costs and potentially enable access to reserves that might be uneconomical with conventional drilling techniques. The precise control offered by MPD systems can be particularly beneficial in the context of autonomous drilling systems, as it allows for more accurate implementation of AI-generated recommendations and enhances the overall adaptability of the drilling process.

Some of the rig sensors utilized in the autonomous drilling system will now be described in detail in FIGS. 1B, 1C, and 1D. These figures provide a closer look at the VMH, VML, and mud cuttings measurement system MWS, respectively. These sensors can contribute to the real-time monitoring and analysis capabilities of the autonomous drilling system, enabling precise measurement of drilling fluid properties, flow rates, and cuttings characteristics throughout the well construction process.

Referring to FIG. 1B, the VMH 130 is depicted. This meter can be designed to measure the flow of drilling fluid entering the wellbore. The VMH 130 can include a conduit with a first tract T1 and a second tract T2. These tracts can be connected by a curved portion T3. The conduit provides a path for the drilling fluid to flow from the mud pumps 106 through the VMH 130 to the standpipe 108.

The VMH 130 also can include shut-off valves SV2 and SV3. These valves control the flow of drilling fluid through the conduit. In some cases, the shut-off valves SV2 and SV3 can be solenoid valves, which can be electronically controlled to open or close the flow path. In other cases, the shut-off valves SV2 and SV3 can be manually operated valves.

A port for removal of cuttings CRP can also be included in the VMH 130. This port allows for the extraction problematic materials or jetting of problem cuttings from the drilling fluid as it flows through the meter.

The VMH 130 operates based on the Venturi principle, which states that fluid speed increases when it flows through a constricted section of a pipe, causing a drop in fluid pressure. The VMH 130 can include a narrowed section, which includes frustoconical portions V1 and V2 and a linkage portion V3. As the drilling fluid flows through this narrowed section, it accelerates, causing a pressure drop. This pressure drop can be measured by differential pressure sensors p1, p2, p3, and p4, which are connected to the conduit at different points. The pressure differences measured by these sensors, particularly between p1 and p2, and between p3 and p4, are directly proportional to the square of the fluid flow rate, allowing the flow rate to be calculated. In some aspects, the VMH 130 can include additional features or components to enhance its performance. For example, it can include a thermal dispersion sensor D2, which measures the heat transfer rate from a heated sensor to the fluid. This measurement can provide additional information about the fluid flow rate, especially in low flow conditions. In other cases, the VMH 130 can be integrated with other sensors or devices, such as temperature sensors or flow switches, to provide a comprehensive monitoring system for the drilling fluid entering the wellbore from the sensors p4 and p3 the density of the fluid can be calculated.

Referring to FIG. 1C, the VML 140 is depicted. This meter can be designed to measure the mass flow rate of drilling fluid exiting the wellbore in both open loop and closed loop drilling configurations. The VML 140 can include a bypass conduit BC with a first tract T1 that can be substantially vertical and a second tract T2 that can be inclined. These tracts can be connected by a curved portion T3. The bypass conduit BC provides a path for the drilling fluid to flow through the VML 140.

The VML 140 also can include shut-off valves SV2 and SV3. These valves control the flow of drilling fluid through the main duct MD and the bypass conduit BC to optional paddle flow meter (PFM) and to the shale shaker, mud tanks and cuttings disposal. In some cases, the shut-off valves SV2 and SV3 can be solenoid valves, which can be electronically controlled to open or close the flow path. In other cases, the shut-off valves SV2 and SV3 can be manually operated valves.

A port for removal of cuttings CRP can also be included in the VML 140. This port allows for the extraction problematic materials or jetting of problem cuttings from the drilling fluid as it flows through the meter.

The VML 140 operates based on the Venturi principle as described above. In some aspects, the VML 140 can include additional features or components to enhance its performance. For example, it can include a thermal dispersion sensor D2, which measures the heat transfer rate from a heated sensor to the fluid. This measurement can provide additional information about the fluid flow rate, especially in low flow conditions. In other cases, the VML 140 can be integrated with other sensors or devices, such as temperature sensors or flow switches, to provide a comprehensive monitoring system for the drilling fluid exiting the wellbore.

A beneficial element of the VML is its ability to function effectively in both open and closed loop drilling configurations, providing continuous measurement capabilities where a Coriolis mass flow meter may be ineffective in open loop scenarios. This versatility allows for seamless transitions between open and closed well construction operations without any loss of real-time well data or condition monitoring. Furthermore, the VML's design permits the standard flow paddle meter in the flowline before the shale shaker to operate normally, ensuring that existing monitoring systems remain fully functional. This dual compatibility enhances the system's flexibility and reliability, maintaining consistent data acquisition across various drilling modes and conditions.

Referring to FIG. 1D, the mud cuttings measurement system 150 is depicted. This system can include a mud line 122 that receives mud flow from the well. The mud line 122 directs the mud flow to shale shaker 123 which filters certain solids from the mud dispenses the cuttings on a measurement device 152, which can be designed to analyze the weight of cuttings from the wellbore and determine parameters such as the volume of cuttings. In other embodiments, the mud cuttings measurement systems 150 can alternatively be a sensor system capable of measuring and calculating weight, volume and mass flow of cuttings (solids phase) at various locations in the flowline between the well head and the shale shakers using other sensor technology. For example, a cuttings/solids measurement system 150 can be based on enabling technologies including, but not limited to, nuclear, x-ray, ultrasonic, acoustic, Coriolis, delta pressure, electrical capacitance and resistance type sensors. The sensor system 150 is ideally installed in a non-intrusive manner to reduce the maintenance and recalibration requirements. The cuttings/solids sensor system 150 can also be used in systems for other phases of well construction, for example in the flowback system 600 or drill-out system 1000, where the cutting/solids sensor 150 can be referred to as a cuttings flow sensor 1046. The mud cuttings system 150 can also be used to measure plug cuttings and sand in return flow during plug drill-out and well flow back operations, similar to the sand volume and weight sensor 1048.

In some aspects, the autonomous drilling system may also use the cuttings data to generate a synthetic hole caliper survey, providing a detailed representation of the true shape and volume of the open hole that has been drilled. For example, the system may analyze the volume and characteristics of the cuttings, correlate this information with the drilling parameters and formation data, and use advanced algorithms to reconstruct the wellbore geometry. This synthetic caliper survey may provide valuable insights into wellbore stability, hole cleaning efficiency, and potential problem areas without the need for running a physical caliper tool. The system may continuously update this synthetic survey in real-time, allowing for proactive adjustments to drilling parameters and mud properties to maintain optimal wellbore conditions throughout the drilling process.

The measurement device 152 can include a tray 20 that is supported by a frame and can rotate between two positions. In the first position, the tray 20 is nearly horizontal, slightly tilted with the front end higher than the rear end. This position can be for collecting debris from the shale shaker. In the second position, the front end is lower than the rear end, allowing debris to be discharged by gravity.

The movement member controls the tray's rotation. The actuator 32 rotates the shaft 31, which in turn moves the arm 33. The arm's first end 33a is connected to shaft 31. The arm's second end 33b is connected to the first protrusion 23a on the tray, causing the tray to follow a circular trajectory T1. Simultaneously, the second protrusion 23b follows a curved trajectory T2 defined by the slot 42 in the guide 40. This combination of movements results in a roto-translation of the tray between the two positions.

The device continuously alternates between loading debris in the first position and discharging it in the second position, allowing for continuous quantitative analysis of the drilling debris. The measured debris is then dumped into cuttings disposal 127.

In some aspects, the mud cuttings measurement system 150 can include additional sensors or components to enhance its performance. For example, it can incorporate a moisture sensor to measure the wet content (e.g. mud/oil/water) content of the cuttings, providing additional information about the drilling fluid properties. In other cases, the system can include a particle size analyzer to measure the size distribution of the cuttings, which can provide insights into the efficiency of the drilling process and the characteristics of the geological formations being drilled.

In some aspects, the autonomous drilling system may be configured to optimize the accuracy of the cuttings volume output by incorporating input from operators or well plan information regarding rock density and the percentage of mud at density on the cuttings being weighed. The system may also utilize offset lithological and geological data, when available, to specify the rock cutting properties. This can include information about cutting size and shape, which can affect the amount of mud carried with the cuttings over the shakers. By incorporating these additional data points, the system can enhance its ability to accurately assess wellbore conditions, optimize drilling parameters, and maintain efficient hole cleaning throughout the drilling process.

In some aspects, the mud cuttings measurement system 150 can be configured to operate in different drilling modes. For instance, it can operate in a continuous mode during normal drilling operations, providing real-time cuttings data. Alternatively, it can operate in a batch mode during specific drilling operations, such as when making a connection or performing a drilling test. This flexibility allows the system to adapt to different drilling conditions and operational requirements.

As mentioned above, the mud cuttings measurement system may estimate the volume of cuttings by averaging measurements across the time cycle when it is dumping cuttings, and the shale shaker is still producing and dumping cuttings past the mud cuttings weight sensor. This approach can allow for a more accurate representation of the continuous flow of cuttings, as it accounts for the cuttings that bypass the weight sensor during the dumping process. By utilizing this averaging method, the system may provide a more comprehensive and reliable estimate of the total cuttings volume, enabling an improved assessment of formation characteristics, drilling efficiency, and potential wellbore stability issues throughout the drilling operation. It can also include safety features, such as overload protection or fault detection systems, to ensure safe and reliable operation. In some aspects, the system may provide an estimate of the volume of hole drilled at depth and the volume of hole collapsed into the well from some open hole depth. This estimation capability may enhance the system's ability to monitor wellbore stability and detect potential issues such as formation collapse or excessive hole enlargement. By comparing the estimated drilled volume with the measured cuttings volume, the system may identify discrepancies that could indicate wellbore instability or other drilling-related problems. This information may be used to adjust drilling parameters, mud properties, or well trajectory to maintain optimal wellbore conditions and improve overall drilling efficiency.

In some cases, the mud cuttings measurement system 150 can be designed to handle different types of drilling fluids. For example, it can be configured to work with water-based muds, oil-based muds, synthetic-based muds, or other types of drilling fluids. This versatility allows the system to be used in a wide range of drilling operations, from conventional vertical drilling to complex horizontal or directional drilling.

In some aspects, the mud cuttings measurement system 150 can include features for easy maintenance and cleaning. For example, it can include removable components or access ports for easy cleaning or replacement of parts. It can also include self-cleaning features, such as automatic flushing or backwashing systems, to maintain its performance and reliability over extended periods of operation.

In some cases, the mud cuttings measurement system 150 can be designed for robust operation in harsh drilling environments. For example, it can include ruggedized components or protective housings to withstand high pressures, high temperatures, corrosive fluids, or abrasive cuttings. It can also include safety features, such as overload protection or fault detection systems, to ensure safe and reliable operation.

Overall, the mud cuttings measurement system 150 provides a function in the autonomous drilling system, enabling continuous monitoring and analysis of the cuttings from the wellbore. This data can be beneficial for assessing wellbore stability, optimizing hole cleaning efficiency, and enhancing the overall performance and safety of the drilling operation.

With this cutting measurement system 150, the autonomous drilling system may be able to derive the volume of cuttings that have come out of the hole. By utilizing assumed density values and the percentage of mud on the cuttings being weighed, the system can calculate the volume of cuttings based on the measured weight. The system may also average the measurements across the time cycle when it is dumping and the cuttings are still flowing off the end of the shale shaker, providing a more accurate representation of the continuous flow of cuttings. This capability can enable real-time monitoring of formation characteristics, drilling efficiency, and potential wellbore stability issues, further enhancing the system's ability to optimize drilling operations and maintain well integrity.

In some cases, the mud cuttings measurement system 150 may provide beneficial insights into wellbore stability issues. Some materials measured by the system may not be cuttings from the current drilling operation, but rather wellbore materials coming from the wellbore due to instability or collapse, leading to more cuttings volume than expected. Additionally, the system may detect metal swarf, cement, and packer cuttings returned due to other periodic operations outside of normal geological well formation drilling. The measurement device 152 can provide beneficial operational insight during these operations as well, supporting the improved ability to maintain well stability and control. By analyzing the quantity, composition, and timing of these unexpected materials, the system may help identify potential wellbore stability issues, allowing for proactive measures to be taken to prevent further deterioration and maintain well integrity.

Having described the overall rig structure, the details of the system will now be described with respect to the remaining figures, which provide visual representations of various components and processes of the autonomous drilling system. These figures illustrate the operational processes, network architecture, neural network module, and HMI, offering a comprehensive view of the system's structure and functionality. Through these illustrations, the intricate workings of the autonomous drilling system and its components will be elucidated, providing a clear understanding of how the system optimizes well construction operations.

Referring to FIG. 2A, the flowchart illustrates an overall operational process 200 of an autonomous drilling system. The process 200 can begin with a step 202, where data collection occurs. This step can involve collecting real-time data on drilling parameters from surface and downhole sensors, measuring mass flow rate of mud into and out of the well using flow meters, and analyzing characteristics of cuttings from the wellbore using cutting weight sensors.

In some aspects, the data acquisition system of the autonomous drilling system can include various types of sensors to collect a wide range of real-time data. For instance, the system can include pressure sensors to measure the pressure of the drilling fluid in the wellbore, temperature sensors to monitor the temperature of the drilling fluid and the wellbore, and vibration sensors to detect vibrations in the drill string and the wellbore. These sensors provide data that can be used to monitor the drilling operation and detect potential issues in real-time. Additionally, the system may incorporate data collected by the rig itself, such as hook load, rotary torque, and pump pressure. In some cases, the system may also integrate data from third-party service providers involved in the well construction process, which may include directional drilling data, formation evaluation logs, or mud logging information. This comprehensive data collection approach allows for a more holistic view of the drilling operation, enabling more accurate analysis and decision-making.

In some cases, the data acquisition system can also include flow meters to measure the mass flow rate of the drilling fluid into and out of the well. These flow meters can be based on various technologies, such as Venturi meters, Coriolis meters, or thermal dispersion meters. The choice of flow meter technology can depend on various factors, such as the expected flow rates, the properties of the drilling fluid, and the operating conditions of the well.

In addition to the above, the data acquisition system can also include cutting weight sensors to analyze the characteristics of the cuttings and other materials emanating from the wellbore. These sensors can measure various properties of the cuttings, such as their weight, volume, size, shape, density, and composition. The data from the cutting weight sensors can provide insights into the geological formations being drilled, the efficiency of the drilling process, return mud in the anulus and the condition of the wellbore.

The process 200 can then move to a step 204 for data preprocessing. In this step, raw data can be cleaned, filtered, and formatted for compatibility with the system. Additionally, derived data points can be generated through calculations based on sensor data. In some aspects, the data preprocessing step can involve various data cleaning and filtering techniques to remove noise, outliers, or erroneous data points from the raw data. This can help to improve the quality and reliability of the data, which is beneficial for the subsequent analysis and decision-making steps. In some cases, the data preprocessing step can also involve data transformation or normalization procedures to convert the raw data into a suitable format for the digital twin models and the AI algorithms. This can include scaling the data to a range, converting categorical data into numerical data, or encoding time-series data into a suitable format.

In some cases, the data preprocessing step can also involve feature engineering, where new features are created from the existing data to capture more complex relationships or patterns. This can include techniques such as polynomial features, interaction terms, or domain-specific transformations based on expert knowledge of the drilling process. Additionally, the preprocessing step can incorporate data augmentation techniques to enhance the robustness of the AI models, such as adding controlled noise to the data or generating synthetic samples to balance underrepresented classes or conditions.

In addition to the above, the data preprocessing step can also involve the generation of derived data points based on the raw data. These derived data points can be calculated using various mathematical or statistical operations, such as averages, sums, differences, ratios, or integrals. The derived data points can provide additional information that cannot be directly measured by the sensors but can be inferred from the raw data.

In some aspects, the derived data points can include complex calculations that combine multiple sensor readings to provide more comprehensive insights into the drilling process. These calculations may involve the identification and labeling of operational states, trends, or events in both time and depth domains. For example, the system may use simple statistical methods or more advanced machine learning and AI techniques to classify drilling states such as “drilling ahead,” “making connections,” or “tripping in/out.”

The system may also identify and label trends in real-time, such as increasing or decreasing rate of penetration, changes in torque, or variations in mud properties. These labeled data points may be associated with both time stamps and depth measurements, allowing for multi-dimensional analysis of drilling operations.

In some cases, the system may employ pattern recognition algorithms to detect and label specific events, such as stick-slip occurrences, bit bouncing, or sudden changes in formation characteristics. These labeled events may provide beneficial context for understanding drilling performance and potential issues.

The autonomous drilling system may use both simple threshold-based rules and more sophisticated machine learning models to perform this labeling. For instance, a simple rule might label a period as “connection time” when the hook load exceeds a certain threshold and rotation stops. In contrast, a machine learning model might use a combination of multiple sensor inputs to identify more complex operational states or events.

By generating these labeled data points in both time and depth domains, the system may create a rich, annotated dataset that can be used for real-time decision making, post-well analysis, and continuous improvement of the AI models. This approach may enhance the system's ability to provide context-aware recommendations and optimize drilling operations based on a comprehensive understanding of past and current operational states and events.

For example, the system can calculate the rate of penetration (ROP) by combining depth measurements with time data or determine the equivalent circulating density (ECD) by considering the static mud weight, friction pressure losses, and annular velocity. In some cases, the system can also generate derived data points that represent trends or patterns over time, such as moving averages or rate of change calculations, which can help identify gradual shifts in drilling conditions or equipment performance.

The data preprocessing step can also involve the creation of dimensionless parameters or ratios that can provide information about the drilling process. For instance, the system can calculate the Reynolds number to characterize the flow regime in the wellbore, or the friction factor to assess the efficiency of fluid circulation. These derived parameters can help in making comparisons across different well sections or drilling operations, regardless of variations in well depth, mud properties, or other factors.

In some cases, the data preprocessing step can include the application of advanced signal processing techniques to extract meaningful information from complex sensor data. This can involve techniques such as Fourier transforms to analyze frequency components of vibration data, or wavelet analysis to detect transient events in pressure or flow rate measurements. The results of these analyses can be used to generate additional derived data points that capture specific characteristics of the drilling process, such as the presence of stick-slip vibrations or the occurrence of pressure pulses.

Next, the process 200 can proceed to a step 206 for digital twin update. This step can involve feeding real-time data and the well plan into a digital twin framework. Digital twins can simulate specific aspects of the well construction process, and the framework can be continuously updated to reflect the current state of the well.

In some aspects, the digital twin framework can include multiple digital twins, each simulating a specific aspect of the well construction process. For example, one digital twin can simulate the dynamics of the drill string, another can simulate the circulation of the drilling fluid, and another can simulate the geometry of the wellbore. Each digital twin can be updated in real-time with the latest data from the sensors, allowing the digital twins to provide accurate and up-to-date simulations of the drilling operation.

In some cases, the digital twin framework can also incorporate the well plan into the simulations. The well plan can provide information about the intended trajectory of the well, the target depth, the type of geological formations to be drilled, and the expected drilling parameters. By incorporating the well plan into the digital twin simulations, the system can compare the actual drilling operation with the planned operation and detect any deviations or anomalies.

In some aspects, the digital twin framework can include a complex network of interconnected digital twins, each focusing on a specific aspect of the drilling operation. For instance, a drill string dynamics digital twin can simulate the vibrations, stresses, and torque experienced by the drill string. A fluid dynamics digital twin can model the flow of drilling fluid, including its pressure, velocity, and temperature throughout the wellbore. A formation digital twin can represent the geological characteristics of the rock being drilled, including its porosity, permeability, and strength.

These digital twins can work together in a hierarchical or parallel structure, exchanging information and influencing each other's simulations. For example, the drill string dynamics digital twin can provide input on the bit rotation speed and weight on bit to the formation digital twin, which in turn can update the expected rate of penetration and formation characteristics. This information can then feed into the fluid dynamics digital twin to adjust the mud flow rate and properties.

In some cases, the digital twins can employ various modeling techniques, such as finite element analysis, computational fluid dynamics, or discrete element methods, depending on the specific aspect being simulated. These models can be continuously refined and calibrated based on the real-time sensor data, improving their accuracy and predictive capabilities over time.

The digital twin framework can also include an integration module that synthesizes the outputs from individual digital twins. This module can use advanced data fusion techniques, such as Kalman filtering or Bayesian inference, to combine the results from different simulations and resolve any discrepancies or uncertainties.

In some aspects, the digital twin framework can incorporate machine learning algorithms to enhance its predictive capabilities. These algorithms can analyze historical data and patterns to identify correlations and trends that cannot be apparent in physics-based models alone. For instance, a neural network can be trained to predict drill bit wear based on a combination of simulated stress data from the drill string dynamics digital twin and historical performance data.

In other words, the digital twin framework can be multifaceted and probabilistic in nature. For example, the framework can provide a range of possible outcomes for a given drilling scenario, each with an associated probability based on the confidence levels of the individual digital twins and their integrated analysis. These conclusions can include predictions of drilling performance, risk assessments for potential issues like stuck pipe or lost circulation, and output (i.e. recommendations and/or instructions) for improved (e.g. optimal) drilling parameters.

In some cases, the digital twin framework can also perform sensitivity analyses, running multiple simulations with varying input parameters to identify factors affecting the drilling operation. This can help in prioritizing which aspects of the operation to focus on for optimization or risk mitigation.

The process 200 can then advance to a step 208 for real-time analysis and prediction. In this step, AI agents analyze the data and use physics models to predict potential issues. Digital twins can run parallel simulations incorporating operational data, and predictive models forecast outcomes of drilling operations under various scenarios.

In some aspects, the AI agents can use various machine learning algorithms to analyze the data and make predictions. These algorithms can include supervised learning algorithms, unsupervised learning algorithms, reinforcement learning algorithms, or deep learning algorithms. The choice of algorithm can depend on the type of data, the complexity of the problem, and the available computational resources.

In some cases, the AI agents can also use physics models to make predictions. These physics models can be based on the laws of physics, such as the laws of motion, the laws of thermodynamics, or the laws of fluid dynamics. The physics models can provide a theoretical basis for the predictions, complementing the empirical learning of the machine learning algorithms.

In addition to the above, the digital twins can run parallel simulations incorporating the operational data. These simulations can provide a virtual representation of the drilling operation, allowing the system to explore various scenarios and predict the outcomes of different drilling strategies. The simulations can also provide a visual representation of the drilling operation, which can be useful for monitoring and troubleshooting purposes.

The predictive models in the autonomous drilling system can employ a combination of physics-based simulations and machine learning techniques to forecast outcomes of drilling operations under various scenarios. These models can utilize historical data, real-time sensor inputs, and digital twin simulations to make predictions about future drilling performance, potential issues, and improved (e.g. optimal) drilling parameters. In some cases, the predictive models can use ensemble methods, combining outputs from multiple algorithms to improve accuracy and robustness. The models can also incorporate uncertainty quantification techniques, providing confidence intervals or probability distributions for their predictions. As new data becomes available, the predictive models can continuously update and refine their forecasts, adapting to changing well conditions and improving their predictive capabilities over time.

The process 200 can then advance to a step 210 for integration of digital twin outputs. In this step, outputs from multiple digital twins can be aggregated by advanced AI models. The AI synthesizes information to determine well health and operational status. Patterns, correlations, and interactions are detected for comprehensive analysis.

In some aspects, the AI agent module of the autonomous drilling system can use various machine learning algorithms to analyze the data and make predictions. These algorithms can include supervised learning algorithms, unsupervised learning algorithms, reinforcement learning algorithms, or deep learning algorithms. The choice of algorithm can depend on the type of data, the complexity of the problem, and the available computational resources.

In some cases, the AI agent module can also use physics models to make predictions. These physics models can be based on the laws of physics, such as the laws of motion, the laws of thermodynamics, or the laws of fluid dynamics. The physics models can provide a theoretical basis for the predictions, complementing the empirical learning of the machine learning algorithms.

In some aspects, the AI agent module can employ advanced ensemble learning techniques to combine the outputs of multiple machine learning algorithms and physics models. This approach can leverage the strengths of different predictive methods, potentially improving the overall accuracy and robustness of the predictions. For example, a gradient boosting algorithm might be used in conjunction with a physics-based model to predict drill bit wear, with the machine learning algorithm capturing complex patterns in historical data while the physics model ensures adherence to known physical constraints.

The AI agent module can also incorporate adaptive learning capabilities, allowing it to continuously refine its predictive models based on new data and feedback from the drilling operation. This can involve techniques such as online learning or transfer learning, where the AI agent can quickly adapt to changing well conditions or apply knowledge gained from one well to improve predictions for another.

In some cases, the physics models used by the AI agent module can be enhanced with uncertainty quantification methods. These methods can provide probabilistic predictions that account for uncertainties in input parameters, model structure, and measurement errors. This can be particularly beneficial in drilling operations, where many factors are uncertain or difficult to measure precisely.

The digital twin simulations can be further enhanced with real-time optimization capabilities. As the simulations can run in parallel with the actual drilling operation, they can continuously explore different operational scenarios and suggest improved (e.g. optimal) drilling parameters. This can involve techniques such as model predictive control or reinforcement learning, where the AI agent learns to make sequential decisions that optimize long-term drilling performance.

The process 200 then proceeds to a step 212 for decision-making and adjustment output (i.e. recommendations and/or instructions) for well health and safety management. This step involves AI evaluating combined data to make predictions and recommend drilling adjustments. The system monitors for potential well control events and recommends corrective actions. Operators receive instructions to mitigate risks and prevent non-productive time. A real-time optimization engine suggests drilling parameters for efficiency and safety. The HMI displays output (i.e. recommendations and/or instructions) and allows for operator intervention.

In some aspects, the real-time optimization engine of the autonomous drilling system can use various optimization algorithms to suggest ideal drilling parameters. These algorithms can include linear programming, nonlinear programming, genetic algorithms, or swarm optimization algorithms. The choice of algorithm can depend on the type of problem, the complexity of the drilling operation, and the available computational resources.

In some cases, the real-time optimization engine can also use feedback control strategies to adjust the drilling parameters in real-time. These strategies can include proportional-integral-derivative (PID) control, model predictive control, or adaptive control. The choice of control strategy can depend on the dynamics of the drilling operation, the accuracy of the digital twin models, and the performance requirements of the drilling operation.

The process 200 then proceeds with a step 214 for autonomous operation and human oversight. The system autonomously adjusts drilling operations based on AI output (i.e. recommendations and/or instructions). Operators can intervene and manually adjust parameters if desired. The process can include continuous improvement through secure cloud connectivity and historical data learning.

In some aspects, the autonomous drilling system can include an integrated rig control system that can be configured to automate the drilling operations based on the AI output (i.e. recommendations and/or instructions). This control system can include various components, such as programmable logic controllers (PLCs), motor control centers, variable frequency drives, and safety systems. The control system can also interface with various drilling equipment, such as the top drive, the mud pumps, and the blowout preventer.

In some cases, the autonomous drilling system can also include an HMI that provides a user-friendly interface for the rig operators. The HMI can display real-time data, AI output (i.e. recommendations and/or instructions), and control options. The HMI can also allow the operators to manually intervene in the drilling process, providing a balance between automation and human oversight.

In some aspects, the autonomous drilling system can be configured to operate in both open loop and closed loop drilling configurations. In some aspects, the autonomous drilling system can be configured to utilize both high-pressure and low-pressure Venturi flow meters continuously, regardless of whether the drilling operation is in an open loop or closed loop configuration. This approach allows for comprehensive monitoring of drilling fluid flow rates throughout the system. The transition between open and closed loop configurations may be facilitated by the use of a choke manifold, which can include one or more adjustable chokes or a choke bypass. In an open loop configuration, the choke may be fully open, allowing the drilling fluid to flow freely to the atmosphere, or the choke may be bypassed completely in the open loop configuration. In a closed loop configuration, the choke may be partially closed to maintain backpressure on the wellbore.

The system may continuously monitor and adjust the choke position based on real-time data and AI recommendations to optimize drilling parameters and maintain wellbore stability. In cases where high-pressure flow meters are not installed or available, the system may estimate the input side flow rates using alternative data sources. These can include mud pump stroke counts multiplied by cylinder volume and efficiency factors, as well as mud report data for density calculations. The choke position and its effect on backpressure may also be factored into these calculations to ensure accurate flow rate estimations.

This flexibility in data acquisition, processing, and choke control enables the system to maintain accurate flow monitoring and adapt to various equipment configurations and operational conditions, thereby optimizing the drilling operation across different scenarios. The ability to seamlessly transition between open and closed loop configurations through choke manipulation allows the system to respond quickly to changing well conditions and maintain improved (e.g. optimal) drilling performance. As mentioned above, the ability to seamlessly transition between open and closed loop configurations may be achieved through a combination of RCD bearing seal assembly manipulation and valve position changes to redirect flow through the choke. This approach allows the system to respond quickly to changing well conditions and maintain improved (e.g. optimal) drilling performance. In open loop configuration, the RCD may operate without a bearing assembly, effectively open to atmosphere, while in closed loop configuration, the RCD may utilize a bearing assembly to create a seal around the drill pipe.

It is noted that the choke(s) used in the autonomous drilling system may be of various types, each capable of sustainably controlling flow in drilling operations. These may include linear chokes, which provide precise flow control through a linear valve movement; rotational orifice chokes, which adjust flow by rotating an orifice plate (some with multiple orifices); some choke valves can use automation and electric, air or Hydraulic actuators for local and remote operation; cage chokes, which use a cylindrical cage with flow ports for pressure control; wedge chokes, which employ a wedge-shaped plug for flow regulation; and ball chokes, which use a rotating ball with a bore to control flow. Other types of valves capable of sustainably choking flow may also be utilized, depending on the specific requirements of the drilling operation, the well conditions, and the desired level of control precision.

In some aspects, the AI's evaluation of combined data for predictions and drilling adjustment output (i.e. recommendations and/or instructions) can involve a sophisticated process of data fusion and multi-modal analysis. The AI can integrate real-time sensor data, historical well information, digital twin simulation outputs, and physics-based model predictions to create a comprehensive understanding of the current drilling state. This holistic approach can allow the AI to identify subtle patterns and correlations that might not be apparent when analyzing individual data streams in isolation. The AI can employ advanced techniques such as deep learning, reinforcement learning, or Bayesian inference to process this complex, high-dimensional data and generate actionable insights. These insights can include predictions of future drilling performance, potential risks or issues, and improved (e.g. optimal) drilling parameters for various scenarios. The AI can continuously refine its predictions and output (i.e. recommendations and/or instructions) as new data becomes available, adapting to changing well conditions and improving its decision-making capabilities over time.

It is noted that the AI agents described above can be deployed using a distributed architecture, with components running both on-site at the drilling rig and in cloud-based environments. The on-site deployment can include edge computing devices that process real-time sensor data and execute time-sensitive control algorithms, ensuring rapid response to changing drilling conditions. These edge devices can host lightweight versions of the AI models, optimized for low-latency decision-making. Meanwhile, more computationally intensive tasks, such as complex simulations and long-term trend analysis, can be performed in the cloud, leveraging scalable computing resources.

The deployment strategy can also incorporate a hierarchical structure, with different AI agents specializing in specific aspects of the drilling operation. For example, lower-level agents can focus on individual subsystems like mud pump control or drill string dynamics, while higher-level agents coordinate overall drilling strategy and risk management. This modular approach can allow for easier updates and maintenance of the AI system, as individual components can be refined or replaced without disrupting the operation. Additionally, the system can employ containerization technologies to ensure consistent performance across different hardware environments and facilitate seamless updates to the AI models as new algorithms or data become available.

Having described the overall operational process of the autonomous drilling system, FIG. 2B will be described which illustrates the specific actions taken by drilling operators within this advanced drilling system. This figure provides a detailed breakdown of the operator's role in the system, highlighting the points of human interaction and decision-making that complement the AI-driven processes. By examining these operator actions, we gain a clearer understanding of how human expertise and oversight are integrated with the autonomous capabilities of the system, ensuring safe, efficient, and effective drilling operations.

Referring to FIG. 2B, the flowchart illustrates operator actions process 220 in an autonomous drilling system. The process 220 can begin with a step 222 for system initialization, where the operator powers on the autonomous drilling system and the system performs self-diagnostics and sensor checks. This initialization step ensures that components of the system are functioning correctly before drilling operations commence. In some cases, the system can perform additional checks or calibrations, such as verifying the accuracy of the sensors or testing the communication links between different components.

The process 220 can then move to a step 224 for operator input. In this step, the operator inputs or loads the well construction plan, and the system integrates the plan into the digital twin framework. The well construction plan can include various details about the intended drilling operation, such as the target depth, the expected geological formations, and the desired drilling parameters. The operator can input this plan through the HMI 104, using various input devices such as a keyboard, a mouse, or a touchscreen. In some cases, the well construction plan can be pre-loaded into the system from a database or a file, or it can be received from remote data sources, thereby reducing manual input.

Next, the process 220 can proceed to a step 226 for real-time monitoring. The operator observes real-time data and AI output (i.e. recommendations and/or instructions) on the HMI 104, while the system continuously collects and processes sensor data. The real-time data can include various drilling parameters, such as the weight on bit, the rotation speed, the mud flow rate, and the pressure and temperature in the wellbore. The AI output (i.e. recommendations and/or instructions) can include suggested adjustments to the drilling parameters, warnings about potential drilling hazards, or predictions about the drilling performance. The operator can monitor this information in real-time on the HMI 104, allowing them to maintain situational awareness and make informed decisions.

The process 220 then advances to a step 228 for parameter adjustment. The operator reviews the AI-suggested drilling parameter adjustments and can either accept the automated adjustments or manually override them as needed. The AI-suggested adjustments are generated by the AI agent module, which uses the real-time data, the digital twin simulations, and the historical well data to produce informed output (i.e. recommendations and/or instructions). The operator can review these output (i.e. recommendations and/or instructions) on the HMI 104 and decide whether to accept them or make manual adjustments. This step provides a balance between automation and human oversight, allowing the operator to leverage the AI's capabilities while maintaining control over the drilling operation.

Following this, the process 220 can move to a step 230 for drilling progress tracking. The operator monitors the drilling progress against the well plan, and the system provides updates on well health and potential issues. The drilling progress can be tracked in terms of various metrics, such as the depth drilled, the rate of penetration, or the volume of cuttings removed. The system can provide updates on the well health based on the sensor data and the AI analysis, alerting the operator to any changes in the well conditions or potential drilling hazards. This step allows the operator to keep track of the drilling progress and respond promptly to any issues.

The process 220 can then proceed to a step 232 for well control event management. The system alerts the operator to potential well control events, and the operator follows the AI-recommended actions or implements manual intervention. Well control events can include situations such as kicks, losses, or stuck pipe incidents, which can pose risks to the drilling operation and the well integrity. The system can detect these events based on the sensor data and the AI analysis, and it can alert the operator through the HMI 104. The AI can also recommend corrective actions, such as adjusting the mud weight, changing the drilling platform parameters (e.g. control the drilling itself, the pressure via the choke, etc.), or shutting in the well. The operator can follow these AI-recommended actions or implement their own manual intervention, depending on their judgement and expertise.

Next, the process 220 can advance to a step 234 for end of section/well completion. The operator confirms the completion of the drilling section or the entire well, and the system generates final reports and data analysis. The operator can confirm the completion through the HMI 104, marking the end of the drilling operation. The system can then generate final reports summarizing the drilling performance, the well data, and any issues encountered. The system can also perform a final data analysis, comparing the actual drilling results with the well plan and the digital twin predictions. This step provides a comprehensive summary of the drilling operation, allowing for post-drilling review and learning.

The process 220 then proceeds with a step 236 for continuous learning. The operator can provide feedback on the system performance, and the system updates the AI models and the historical database for future operations. Throughout the drilling operation, operators continuously provide comments, data labeling, and feedback, enabling real-time system refinement and optimization. The operator can provide feedback through the HMI 104, commenting on the system's performance, the accuracy of the AI output (i.e. recommendations and/or instructions), or any issues encountered. The system can then use this feedback to update the AI models, improving their learning and prediction capabilities. The system can also update the historical database with the new well data, enriching the data set for future operations. This step ensures that the system continuously learns and improves over time, enhancing its performance and reliability for future drilling operations.

In some aspects, the continuous learning step can involve a multi-faceted approach to system improvement. The operator feedback can be categorized into different types, such as accuracy of predictions, timeliness of alerts, effectiveness of recommended actions, and overall system usability. This categorization can allow for targeted improvements in specific areas of the system's performance.

The AI models can employ advanced machine learning techniques, such as transfer learning or meta-learning, to efficiently incorporate new knowledge without compromising previously learned information. This approach can enable the system to adapt to new drilling scenarios or geological formations while retaining its expertise in familiar situations.

In some cases, the system can implement a form of active learning, where it identifies areas of uncertainty in its predictions and actively seeks operator input or input from offsite or third party entities on these specific issues. This targeted approach to gathering feedback can accelerate the learning process and improve the system's performance in challenging well construction activities.

The historical database update process can involve sophisticated data management techniques, such as data versioning and provenance tracking. These techniques can allow the system to maintain a comprehensive record of how its knowledge evolves over time, enabling rollbacks to previous states if beneficial and providing insights into the system's learning trajectory.

In some aspects, the continuous learning process can extend beyond individual well operations to incorporate cross-well and cross-field learning. The system can identify patterns and best practices across multiple drilling operations, potentially leading to broader insights into improved (e.g. optimal) drilling strategies for different geological formations or operational conditions.

The system can also implement a form of explainable AI, allowing operators to understand the reasoning behind the AI's output (i.e. recommendations and/or instructions) and predictions. This transparency can foster trust between the operators and the AI system, encouraging more frequent and detailed feedback and ultimately leading to more effective continuous improvement.

In some aspects, the operator actions process 220 can include additional steps or variations. For example, the process can include a step for equipment setup or calibration before the system initialization. The process can also include a step for data backup or archiving after the well completion. The process can include additional monitoring or control steps, such as monitoring the equipment health, controlling the rig systems, or coordinating with other personnel. The process can also include steps for handling emergencies or unexpected events, such as activating safety systems, implementing emergency procedures, or coordinating with emergency response teams. These additional steps or variations can enhance the operator's control over the drilling operation and the system's ability to handle a wide range of drilling scenarios.

The autonomous drilling system may operate with varying levels of autonomy, adapting to the operator's preferences and the system's capabilities. In some cases, the system might make fully autonomous adjustments to drilling parameters without requiring operator intervention based on AI instructions. In other instances, it may generate suggested adjustments that need confirmation or acceptance by the driller or operator before implementation. The system can also provide recommendations that require manual input from the operator, or issue alarms and alerts that may or may not necessitate operator action, depending on the severity and nature of the situation. As the digital twin models, AI components, and associated systems become more intelligent over time through continued training and programming, the system may evolve to become increasingly autonomous, potentially reducing the need for frequent operator intervention while maintaining the option for human oversight and control when beneficial.

For example, in the context of drilling operations, the spectrum of automation levels may be applied to various aspects of the drilling process. At the lower end of the spectrum, the system may function as a computer-based aid by providing real-time data on drilling parameters such as weight on bit, rotary speed, and mud flow rate, allowing the driller to make informed decisions. As the level of automation increases, the system may offer advisory functions, such as suggesting improved (e.g. optimal) drilling parameters based on formation characteristics and historical data.

Moving further along the automation spectrum, the system may make decisions and execute actions with varying degrees of human oversight in specific drilling operations. For example, in computer-based decision-making, the system may determine the improved (e.g. optimal) drilling trajectory based on real-time formation evaluation data but inform the driller who can override the decision if beneficial. At higher levels of automation, the system may perform computer-based execution by automatically adjusting drilling parameters such as weight on bit and rotary speed to maintain the improved (e.g. optimal) rate of penetration, while the driller monitors and has the option to intervene.

The most advanced level may involve full automation of certain drilling processes, such as tripping operations or pressure management during managed pressure drilling. In these cases, the system may perform all tasks without human intervention, including making connections, adjusting mud weight, and controlling wellbore pressure. However, the capability for human oversight and intervention may still be maintained as a safety measure, allowing the driller to take control in case of unexpected events or system anomalies. This range of automation levels in drilling operations allows the autonomous drilling system to be tailored to specific well complexities, crew expertise levels, and company risk management policies.

Referring to FIG. 3, the network diagram 300 illustrates an example cloud-based architecture of the autonomous drilling system. The diagram shows the interconnections between various components of the system, highlighting the role of the cloud network 308 in facilitating communication and data exchange.

On the drilling site, the network can include a rig controller 302A, a rig HMI 302B, and a PC 302C. These components are connected to the cloud network 308, allowing for remote monitoring and control of drilling operations.

The rig controller 302A can be responsible for controlling the drilling equipment and processes on the rig. The rig HMI 302B provides a user-friendly interface for the rig operators to monitor real-time data, receive AI output (i.e. recommendations and/or instructions), and manually intervene in the drilling process if desired. The PC 302C can be used for additional data analysis, reporting, or remote control functions.

On the system side, the network can include system models server 304A, system historical database 304B, and a system control center 304C. The system models server 304A can host the digital twin models and drilling well plans, providing the computational resources for the real-time simulations and predictions. The system historical database 304B can store past operational data, enriching the data set for the AI algorithms and enabling continuous learning. The system control center 304C can provide centralized management of the system, coordinating the data flow, the AI analysis, and the control actions.

In this example, the network diagram 300 also shows third party services 306 connected to the cloud network 308. This indicates the system's ability to integrate external data or services, such as geological data, weather forecasts, or third-party drilling software. This integration can enhance the system's capabilities and provide additional insights for the drilling operation.

The cloud network 308 can serve as the central hub, facilitating communication and data exchange between the components. This cloud-based architecture allows for real-time data processing, remote access, and scalability of the autonomous drilling system.

In some aspects, the network architecture of the autonomous drilling system can include additional components or features. For example, it can include redundant servers or databases for data backup and recovery. It can also include security features, such as firewalls, encryption, or authentication mechanisms, to protect the data and the system from cyber threats. In some cases, the network can also include wireless communication links, such as satellite or cellular connections, to enable remote operation in locations without wired internet access.

In some cases, the network architecture can be configured to support different types of cloud services, such as Infrastructure as a Service (IaaS), Platform as a Service (PaaS), or Software as a Service (SaaS). This flexibility allows the system to leverage the advantages of cloud computing, such as scalability, cost-effectiveness, and accessibility, while meeting the specific needs of the drilling operation.

In some aspects, the network architecture can be designed for robust operation in harsh drilling environments. For example, it can include ruggedized hardware, industrial-grade network equipment, or fault-tolerant systems to withstand high temperatures, vibrations, or other challenging conditions on the drilling rig. It can also include remote monitoring and diagnostic capabilities to ensure reliable operation and quick resolution of any network issues.

It is noted that the network configuration in FIG. 3 may incorporate a hybrid cloud architecture that combines on-site and off-site infrastructure for all devices shown. The drilling side components, including the rig controller 302A, rig HMI 302B, and PC 302C, may connect to a local “edge cloud” at the well site, which may comprise data integration and storage systems along with computing resources capable of running digital twin simulations and AI algorithms independently. This on-site infrastructure may synchronize with off-site cloud services when available, enabling seamless data exchange and remote operations. The system-side components, including the models server 304A, historical database 304B, and control center 304C, may be distributed between the on-site edge cloud and off-site cloud infrastructure, allowing for flexible allocation of computational resources. The third party services 306 may integrate with both the on-site and off-site components through the cloud network 308, which serves as the central hub for communication and data exchange. As computational demands grow, the on-site processing capabilities for all devices may be expanded, enhancing the system's ability to perform complex analyses and make real-time decisions directly at the well site.

The steps outlined in FIG. 2A and FIG. 2B can be executed by the devices shown in FIG. 3 through a coordinated effort leveraging the cloud-based architecture of the autonomous drilling system.

The data collection step 202 can primarily involve the rig controller 302A, which interfaces with various sensors and equipment on the drilling rig. This real-time data can be transmitted through the cloud network 308 to the system models server 304A for processing. The data preprocessing step 204 can occur on the system models server 304A, where raw data is cleaned, filtered, and formatted for use in the digital twin models.

The digital twin update step 206 can be executed on the system models server 304A, which hosts the digital twin models. The server can continuously update these models with the preprocessed real-time data and well plan information stored in the system historical database 304B.

For the real-time analysis and prediction step 208, the system models server 304A can run AI algorithms to analyze the data and generate predictions. The digital twin simulations can also be executed on this server, leveraging its computational resources.

The integration of digital twin outputs at step 210 and the decision-making process at step 212 can occur on the system control center 304C. This component can aggregate outputs from multiple digital twins, synthesize the information, and generate recommendations for drilling adjustments.

The autonomous operation and human oversight step 214 can involve both the system control center 304C and the drilling side components. The system control center 304C can send automated adjustment commands to the rig controller 302A, while the rig HMI 302B can display recommendations and allow for operator intervention.

The operator actions outlined in FIG. 2B can primarily involve interactions with the rig HMI 302B. The system initialization step 222 and operator input step 224 can be carried out through this interface, with the input data being transmitted to the system models server 304A and system historical database 304B via the cloud network 308.

Real-time monitoring step 226 and parameter adjustment step 228 can involve the rig HMI 302B displaying data and recommendations from the system control center 304C, and the operator using the interface to accept or override these recommendations/instructions.

Drilling progress tracking step 230 and well control event management step 232 can rely on the system control center 304C processing data from the rig controller 302A and sending alerts and updates to the rig HMI 302B.

In some aspects, the autonomous drilling system may incorporate multiple rig HMIs to facilitate effective information and control management across various locations and functions at the well site. These HMIs may serve different purposes, with some designed for comprehensive operator control at a central station, while others may be distributed throughout the rig site for specialized monitoring and control tasks. The system may include read-only HMIs for monitoring purposes, as well as HMIs with full control capabilities and data input rights. Examples of specialized HMIs may include interfaces for Managed Pressure Drilling (MPD), Measurement While Drilling (MWD) directional operations, mud logging, well control, activity performance management, and systems for providing recommendations and feedback. This flexible HMI configuration may enable efficient information dissemination and control across different drilling disciplines and operational areas.

The end of section/well completion step 234 can trigger the system control center 304C to generate final reports, which can be viewed on both the rig HMI 302B and the PC 302C.

The continuous learning step 236 can involve the operator providing feedback through the rig HMI 302B, which can be sent to the system models server 304A to update AI models, and to the system historical database 304B to enrich the dataset for future operations.

Throughout these processes, the cloud network 308 can facilitate seamless communication between the components, enabling real-time data exchange and remote access capabilities. The third party services 306 can provide additional data or functionality, integrating with the system through the cloud network 308.

Referring to FIG. 4, the block diagram illustrates a neural network module 400 for an autonomous drilling system. The neural network module 400 may include an input distribution module 404, three digital twin models, and a fusion and output module 406.

The input distribution module 404 can be responsible for distributing the collected real-time data to the digital twin models. This module can preprocess the data to ensure compatibility with the models, such as normalizing the data or converting it into a suitable format. In some cases, the input distribution module 404 can also perform data augmentation techniques, such as adding noise or generating synthetic data, to enhance the robustness of the models.

The autonomous drilling system may focus on well pressure, flow, and stability control. The digital twin models may therefore incorporate parameters such as pressure and flow dynamics, drilling rate, drill string RPM, weight on drill bit (WOB), trip rate (in/out), surface back pressure choke (position/flow pressure response), and surge and swap pressures. These model aspects can be beneficial for maintaining wellbore stability, optimizing drilling performance, and preventing well control incidents. By continuously monitoring and analyzing these parameters, the system may be able to predict and mitigate potential issues, adjust drilling operations in real-time, and ensure safe and efficient well construction.

In some aspects, the input data for the autonomous drilling system can include a wide range of parameters and measurements from various components of the drilling operation. This comprehensive data set can encompass, but is not limited to, wellbore and drilling parameters, such as rate of penetration, weight on bit, and torque; pump data, including pressure and flow rate; standpipe data, such as pressure and temperature; and top drive data, like rotary speed and torque. The input data can also include wellbore data, such as inclination, azimuth, and depth; Rotating Control Device (RCD) data, including seal pressure and wear; and Managed Pressure Drilling (MPD) and well control choke(s) data, such as choke position and backpressure data; High Pressure (HP) and Low Pressure (LP) mass flow meter data, providing accurate measurements of drilling fluid flow rates; mud data, including density, viscosity, and electrical stability; tank data, such as volume and temperature; flare data, like gas composition and flow rate; casing running data, including string weight and makeup torque; cementing data, such as slurry density and pump rate; and rig operation state data, which can include information about the current phase of the drilling operation, equipment status, and crew activities. This wide range of input data allows the autonomous drilling system to maintain a holistic view of the entire drilling operation, enabling more accurate predictions and informed decision-making.

The digital twin models in the neural network module 400 can include a well physics model digital twin 402A, a cuttings management model digital twin 402B, and a well control model digital twin 402C among other applicable models. Each of these digital twins can simulate a specific aspect of the well construction process, providing a virtual representation of the well and the drilling operations.

In some aspects, the autonomous drilling system may incorporate additional or alternative digital twin models to enhance its capabilities and address specific drilling challenges. For example, a reservoir characterization digital twin may be implemented to simulate and predict formation properties, fluid behavior, and production potential in real-time. Another example could be a drill string dynamics digital twin, which may model the complex vibrations, stresses, and interactions of the drill string with the wellbore, potentially improving drilling efficiency and reducing the risk of equipment failure.

For example, the well physics model digital twin 402A simulates the physical properties and dynamics of the well and the drilling process. This can include the behavior of the drill string, the fluid dynamics of the drilling mud, the thermal properties of the wellbore, and the mechanical properties of the geological formations. In some cases, the well physics model digital twin 402A can also simulate the interactions between these components, such as the effect of drill string vibrations on the mud flow or the impact of temperature changes on the wellbore stability.

The cuttings management model digital twin 402B can simulate the generation, transport, and removal of cuttings from the wellbore. This can include the fragmentation of the rock formations by the drill bit, the entrainment of the cuttings in the mud flow, the deposition and erosion of cuttings in the wellbore, and the separation of cuttings at the surface. In some cases, the cuttings management model digital twin 402B can also simulate the effect of drilling parameters on the cuttings management, such as the impact of mud flow rate or drill string rotation speed on the cuttings transport efficiency.

The well control model digital twin 402C can simulate the control of the well pressure and the prevention of well control events. This can include the regulation of the mud weight and the surface backpressure, the detection and control of fluid influxes or losses, and the response to well control events such as kicks or losses. In some cases, the well control model digital twin 402C can also simulate the effect of drilling parameters on the well control, such as the impact of drill string movement or mud composition on the well pressure.

The fusion and output module 406 can aggregate the outputs from the digital twin models and synthesizes the information to determine the well health and operational status. This module can use various data fusion techniques, such as weighted averaging, voting, or machine learning algorithms, to combine the outputs from the digital twins. The fusion and output module 406 can also generate an output (i.e. recommendations and/or instructions) for drilling parameter adjustments based on the aggregated outputs and the well health status.

In some aspects, the outputs of digital twin models may be sequentially input to one another, creating a cascading flow of information through the system. For instance, the fusion and output module 406 may also output its results to another digital twin model for further analysis. This sequential input-output process may allow for more complex, multi-stage simulations and analyses, where the output of one model becomes a beneficial input for another, potentially enabling the system to capture interdependencies and feedback loops within the drilling operation. Such an approach may enhance the system's ability to model and predict complex drilling scenarios, improving overall decision-making and operational efficiency.

The output data from the autonomous drilling system can include a wide range of indicators and analyses for efficient and safe drilling operations. Digital twin model verification and calibration data can be provided, ensuring the accuracy of the simulations by comparing predicted values with actual measurements. Formation evaluation outputs can include porosity, permeability, and fluid content estimates, while the geological model can provide updated stratigraphic information and structural interpretations based on real-time drilling data.

The system can generate various well health indicators, such as hole instability indicators (e.g., cavings or washouts), kick early and positive indicators (e.g., sudden increases in return flow rate or pit volume), and ballooning indicators (e.g., cyclic variations in downhole pressure). Sticking indicators can include torque fluctuations or hook load variations, while cuttings build-up or loss of circulation indicators can involve changes in return flow rate or drilling fluid properties. Well control indicators might encompass gas levels in the mud, formation pressure trends, or sudden changes in downhole pressure. Formation breakdown and collapse indicators can include sudden drops in standpipe pressure or unexpected changes in hole diameter. Drilling vibration indicators can cover stick-slip, whirl, or axial vibrations, while potential twist-off indicators might involve progressive increases in torque or sudden changes in drill string tension.

The system can provide data on various drilling equipment in use and potential issues that may arise during operations. For example, it may detect and report on drilling motor stalls, which can be indicated by sudden increase in differential pressure. In some cases, the system may monitor mud pump performance and identify potential piston, valve, or liner failures that could result in lower flow to the well. The system may also be capable of detecting drillstring washouts, which can be indicated by unexpected changes in pressure or flow rates. Additionally, the system may provide data on bit conditions, such as rate of penetration trends or torque responses, which can help improve (e.g. optimize) drilling performance and predict bit wear. In some aspects, the system may monitor and report on top drive performance, including rotary speed and torque data. The system may also track and analyze data, such as hook load and block position. Furthermore, the system may provide control system setpoints or advisory information for improved (e.g. optimal) drilling parameters based on the aggregated data and analysis from various equipment and sensors throughout the drilling operation.

In some aspects, the neural network module 400 can include additional components or features. For example, it can include a learning module that updates the digital twin models based on the feedback from the fusion and output module 406. This learning module can use various machine learning algorithms, such as supervised learning, unsupervised learning, deep learning or reinforcement learning, to improve the accuracy and robustness of the digital twin models. In some cases, the neural network module 400 can also include a visualization module that generates graphical representations of the digital twin simulations and the well health status, providing a visual aid for the operators and the decision-makers.

The fusion and output module 406 can also incorporate temporal aspects in its analysis. It can use time series analysis techniques to identify trends and patterns in the aggregated data over time, potentially enabling early detection of developing issues or optimization opportunities. In some implementations, the module can employ recurrent neural networks or long short-term memory (LSTM) networks to capture and utilize temporal dependencies in the data.

When generating output (i.e. recommendations and/or instructions) for drilling parameter adjustments, the fusion and output module 406 can utilize optimization algorithms. These algorithms can consider multiple objectives simultaneously, such as maximizing drilling efficiency, minimizing risk, and optimizing cost. In some cases, the module can employ reinforcement learning techniques, where the system learns to make sequences of decisions to optimize long-term outcomes.

The fusion and output module 406 can also include explainable AI features, providing transparency into how it arrives at its conclusions and output (i.e. recommendations and/or instructions). This can involve techniques such as SHAP (SHapley Additive explanations) values or LIME (Local Interpretable Model-agnostic Explanations) to help operators understand the factors influencing the system's decisions.

In some aspects, the neural network module 400 can include a learning module that updates the digital twin models based on the feedback from the fusion and output module 406. This learning module can use various machine learning algorithms, such as supervised learning, unsupervised learning, deep learning or reinforcement learning, to improve the accuracy and robustness of the digital twin models. In some cases, the neural network module 400 can also include a visualization module that generates graphical representations of the digital twin simulations and the well health status, providing a visual aid for the operators and the decision-makers.

Referring to FIG. 5, the diagram illustrates a front view of an HMI display 500 for an autonomous drilling system. The HMI display 500 provides a user-friendly interface for rig operators to monitor real-time data, receive AI output (i.e. recommendations and/or instructions), and manually intervene in the drilling process if desired. The layout of the HMI display 500 can be designed to provide a comprehensive overview of the drilling operation, enabling operators to maintain situational awareness and make informed decisions.

In the upper left quadrant of the HMI display 500, a performance chart 504 is shown. This chart includes graphical representations of various performance metrics and predictions generated by the AI agents and the digital twin models. The performance chart 504 can display information such as the drilling rate, the mud flow rate, the well pressure, and the cuttings volume. It can also display predictions for future drilling performance, such as the expected drilling rate, the optimal mud weight, or the estimated time to reach the target depth. The performance chart 504 allows operators to monitor the drilling performance in real-time and assess the accuracy of the AI predictions.

To the right of the performance chart 504, a well selection 506 section is depicted. This section allows operators to select the well that they want to monitor or control. The well selection 506 can include a list of wells, a map of the drilling site, or a search function to find a specific well. Once a well is selected, the HMI display 500 updates to show the real-time data, AI output (i.e. recommendations and/or instructions), and control options for the selected well. This feature allows operators to easily switch between different wells, which is particularly useful in multi-well drilling operations or drilling rigs with multiple drilling stations for redundancy, supervision or other activity performance management.

In the lower left quadrant of the HMI display 500, a well control, activity performance management, and recommendations section 502 is displayed. This section contains AI-generated recommendations for adjusting the drilling parameters, as well as control options for the operator to implement these recommendations. The recommendations can include suggested adjustments to the weight on bit, the rotation speed, the mud flow rate, choke adjustment, or the mud weight. The control options can include buttons, sliders, or input fields for the operator to adjust the drilling parameters manually. The well control, activity performance management and recommendations section 502 provides a direct interface for the operator to control the drilling operation based on the AI recommendations, enhancing the operator's control over the drilling process.

In the lower right quadrant of the HMI display 500, an alerts section 508 is shown. This section displays alerts or warnings about potential drilling hazards or well control events. The alerts can be generated by the AI agents based on the real-time data and the digital twin simulations. They can include warnings about potential kicks, losses, stuck pipe incidents, or equipment malfunctions. The alerts section 508 provides a visual indication of potential issues, allowing operators to respond promptly and take corrective actions.

At the bottom of the HMI display 500, a row of HMI buttons 510 is visible. These buttons provide quick access to various functions or settings of the autonomous drilling system. The buttons can include navigation buttons to switch between different screens or modes, control buttons to start or stop the drilling operation, and setting buttons to adjust the system settings or preferences. The HMI buttons 510 enhance the usability of the HMI display 500, providing a convenient and intuitive interface for the operators.

In some aspects, the HMI display 500 can include additional features or components to enhance its functionality. For example, it can include a data logging feature to record the real-time data, AI output (i.e. recommendations and/or instructions), and operator actions for later analysis or review. It can also include a help feature to provide guidance or instructions to the operators, or a diagnostic feature to troubleshoot any issues with the system. In some cases, the HMI display 500 can be customizable, allowing operators to configure the layout, the data displays, or the control options to suit their preferences or the specific requirements of the drilling operation.

To further illustrate the comprehensive operation of the autonomous drilling system, a detailed use case will be presented, incorporating steps from the flowcharts in FIG. 2A and FIG. 2B, and the devices shown in FIG. 3.

In this specific scenario, the autonomous drilling system is being used to drill a horizontal well in a shale formation with an extended horizontal section.

The process begins with system initialization step 222, where the operator powers on the autonomous drilling system through the rig controller 302A. The system performs self-diagnostics and sensor checks, ensuring that the components are functioning correctly. This can include verifying the calibration of the high-side pressure sensor PSH, VMH, low-side pressure sensor PSL, VML, and MWS.

Next, in the operator input stage step 224, the operator uses the rig HMI 302B to load the well construction plan. This plan specifies the target depth, the horizontal section, expected shale formation characteristics, and the planned wellbore trajectory including the kickoff point at a kickoff depth. This information can be transmitted via the cloud network 308 to the system models server 304A.

The data collection phase step 202 commences as drilling begins. The rig controller 302A gathers real-time data from the sensors. For example, the VMH measures an initial mud flow rate, while the MWS indicates a mud weight.

In the data preprocessing stage step 204, the system models server 304A cleans and formats this raw data. It can, for instance, filter out noise from the pressure sensor readings and calculate derived data points such as equivalent circulating density based on the mud weight and flow rate.

During the digital twin update phase step 206, the system models server 304A incorporates this preprocessed data into its digital twin framework, creating an up-to-date virtual representation of the well. This might include updating the simulated wellbore pressure profile based on the current mud weight and flow rate.

Real-time analysis and prediction step 208 can then be performed. For example, as the drill bit approaches the planned kickoff point, the AI agents might predict an increased risk of wellbore instability based on the formation characteristics and current drilling parameters.

In the integration of digital twin outputs step 210, the system control center 304C aggregates this information. It might determine that the well health is generally good, but that adjustments can be beneficial for the upcoming directional drilling phase.

During the real-time monitoring phase step 226, the operator observes on the rig HMI 302B that the rate of penetration in a number of feet per hour and that the parameters are within expected ranges as the well approaches the kickoff point.

In the decision-making stage step 212, as the well approaches the kickoff point, the system control center 304C recommends reducing the weight on bit and increasing the mud weight in preparation for the directional drilling phase.

In the parameter adjustment phase step 228, the operator reviews these recommendations/instructions on the rig HMI 302B and accepts them. The integrated rig control system, managed by the rig controller 302A, then autonomously implements these changes.

Throughout the drilling progress tracking stage step 230, the operator monitors the well's progress. As the well successfully kicks off and begins its horizontal section, the system control center 304C provides updates on the wellbore trajectory and formation characteristics.

If a well control event is detected, such as a sudden increase in gas levels in the returning mud as the horizontal section is drilled, the well control event management process step 232 is initiated. The system control center 304C might alert the operator and recommend increasing the bottom hole pressure by closing the choke and increasing the Surface Back Pressure (SBP). This automated process, part of the MPD or Well Control System, can provide a rapid response to the detected well control event. Alternatively, the system might suggest increasing the mud weight as a longer-term solution. The system's ability to consider both immediate and long-term solutions demonstrates its comprehensive approach to well control management.

The autonomous operation and human oversight phase step 214 continue as the horizontal section is drilled, with the system making continuous adjustments to optimize the drilling process while the operator maintains oversight.

A choke may also be controlled by the autonomous drilling system as part of its well control strategy. The system may continuously monitor downhole pressure, flow rates, and other relevant parameters through sensors like the VMH and VML. Based on this real-time data and predictions from the digital twin models, the AI agent module may recommend choke adjustments to maintain the desired bottomhole pressure, especially during operations like transitioning to the horizontal section. The integrated rig control system may then automatically adjust the choke position, while the operator can monitor these changes through the well control activity performance management and recommendations section 502 on the HMI display 500. In case of sudden pressure fluctuations or gas influx, the system may rapidly adjust the choke to prevent well control incidents, with the operator having the option to intervene manually if needed.

Upon reaching the target depth of the and horizontal section, and successfully tripping out of the well after conditioning the hole for the completion, the end of well completion process step 234 is initiated. The operator confirms completion through the rig HMI 302B, and the system models server 304A generates final reports detailing the well path, formation data, and drilling performance metrics.

In the continuous learning phase step 236, the operator provides feedback on the system's performance during the horizontal drilling phase. This feedback, along with the operational data from the successful well, is stored in the system historical database 304B to improve AI models for future horizontal shale wells.

Throughout this process, the cloud network 308 enables real-time communication between the components, while third party services 306, such as real-time geo-steering data providers, are integrated to enhance the system's decision-making capabilities during the horizontal drilling phase.

While the foregoing is directed to example embodiments described herein, other and further example embodiments can be devised without departing from the basic scope thereof. For example, aspects of the present disclosure can be implemented in hardware or software or a combination of hardware and software. One example embodiment described herein can be implemented as a program product for use with a computer system. The program(s) of the program product defines functions of the example embodiments (including the methods described herein) and can be contained on a variety of computer-readable storage media. Illustrative computer-readable storage media include, but are not limited to: (i) non-writable storage media (e.g., read-only memory (ROM) devices within a computer, such as CD-ROM disks readably by a CD-ROM drive, flash memory, ROM chips, or any type of solid-state non-volatile memory) on which information is permanently stored; and (ii) writable storage media (e.g., floppy disks within a diskette drive or hard-disk drive or any type of solid-state random-access memory) on which alterable information is stored. Such computer-readable storage media, when carrying computer-readable instructions that direct the functions of the disclosed example embodiments, are example embodiments of the present disclosure.

It will be appreciated by those skilled in the art that the preceding examples are exemplary and not limiting. It is intended that all permutations, enhancements, equivalents, and improvements thereto are apparent to those skilled in the art upon a reading of the specification and a study of the drawings are included within the true spirit and scope of the present disclosure. It is therefore intended that the following appended claims include all such modifications, permutations, and equivalents as fall within the true spirit and scope of these teachings.

Although the preferred embodiment has been described in detail, it should be understood that various changes, substitutions and alterations can be made therein without departing from the spirit and scope of the invention as defined by the appended claims.

Referring now to the drawings including FIGS. 6, 7A, 7B, 8, and 9, wherein like reference numbers are used to designate like elements throughout, the various views and embodiments of an automated well flowback and production system are illustrated and described, and other possible embodiments are described. The figures are not necessarily drawn to scale, and in some instances the drawings have been exaggerated and/or simplified in places for illustrative purposes only. One of ordinary skill in the art will appreciate the many possible applications and variations based on the following examples of possible embodiments.

Referring to FIG. 6, there is illustrated an automated well flowback and production system (“AWFPS”) that supports an optimal method to bring recently stimulated (i.e., “fracked”) oil, gas or geothermal wells online. The AWFPS 600 is disposed on a surface site 602 (or “pad”) from which a wellbore 604 has been drilled into an underground reservoir 606 containing hydrocarbons or geothermal water. A tubing string 608 is installed in the wellbore 604 and the lower portion of the annulus between the tubing and wellbore is sealed using cement 617. In some parts of the wellbore 604, the tubing string 608 can be surrounded by intermediate casing 610 or surface casing 612, forming respective annular spaces 614, 616 therebetween.

Prior to flowback, the well is first stimulated or “fracked.” This fracking is often performed in stages, with each stage involving one or more zones of the reservoir 606, starting with the zones at the farthest end of the tubing 608 and proceeding sequentially back towards the surface end. Each stage involves using a special tool (the “perf gun”) to form one or more holes or perforations (“perfs”) 618 in the tubing 608 of the target zone. Each perf extends through the tubing 608 and cement 617 and reaches a short distance into the surrounding rock of the reservoir 606. Next, fracking materials, primarily water and sand, are pumped from the surface site 602 down the tubing 608 at very high pressure. The pressurized fracking materials exits the tubing 608 through the perforations 618 and enters the surrounding rock of the target zone, where the pressurized water creates fissures in the rock and holds the fissures open as the sand particles are carried in. The trapped sand particles serve to hold the fissures open even when the water pressure is reduced, thereby greatly increasing the permeability of the rock in the fracked zone. After the desired amount of fracking materials have been pumped into the target zone, a drillable frac plug or bridge plug (not shown) is set in the tubing 608 to seal the just-fracked target zone from the remainder of the tubing 608. This pattern of perfing, fracking, and plugging is repeated for each subsequent zone until the well is fully fracked. Fracking can also be performed in open hole using slotted liners or mechanical sleeves, isolating sections using mechanical packers or ball-drop systems

After fracking, the well must be “drilled out” to initiate flowback. For drill-out, a snubbing, drilling, workover, or coil tubing rig uses a drill-out bit or mill to drill through the frac plugs/bridge plugs of each stage. As each plug is drilled out, trapped frac water, loose sand, and oil, gas, and/or water from the formation (collectively denoted by arrows 619) will begin to flow from the adjacent zone via the fissures and perfs 618 into the tubing 608, thus creating a combined flow (denoted by arrow 621) if the well is allowed to achieve an underbalanced state. Typically, today most all wells and their frac plugs are drilled out in an over pressure mode so only the fluid being circulated from the surface returns along with plug parts and sand that might have been in the wellbore. If the well begins to flow during plug drilling operations, the rig personnel will typically increase the surface back pressure to reduce and stop the well flow before proceeding to the next plug. In contrast, some aspects of the current inventive process include configuring the system so that each frac plug will be drilled out while utilizing state of the art pressure management methods, thus allowing a period of well flow in a controlled manner after each plug is drilled out before the next plug is drilled out, and collecting flowback data for each drill-out zone for future analysis and for use as modeling inputs . . .

Referring still to FIG. 6, in the illustrated embodiment, the AWFPS 600 comprises a wellhead assembly 620 operably connected to the surface end of the tubing string 608 for a well. A well is generally considered to include the outputs of all the zones exiting from a single wellhead assembly. The wellhead assembly 620 includes: a) an automated choke 622 to control the flowrate and pressure of production flow (denoted by arrow 623) exiting the wellhead; b) a wellhead production (upstream) pressure sensor 624; c) an after-wellhead-choke (downstream) pressure sensor 626; d) a wellhead production (upstream) flow temperature sensor 628; e) a production-tubing-to-intermediate-casing annulus pressure sensor 630; f) a production-tubing-to-intermediate-casing annulus temperature sensor 633, and g) an intermediate-casing-to-surface-casing annulus pressure sensor 632.

In some embodiments, the AWFPS 600 further incorporates a sealing system configured to selectively seal the wellbore 604 to pressurize the wellbore to facilitate a closed-loop flowback configuration and to unseal the wellbore to unpressurize the wellbore to facilitate an open-loop flowback configuration. In one implementation, a choke manifold (e.g., choke 622 or 662) can be throttled to provide back-pressure for closed-loop flowback operation when the wellbore is pressurized to enable managed pressure control, whereas wide-open operation of the choke or bypass of the choke manifold using a bypass valve (not shown) can allow open-loop flowback operation when the wellbore is unpressurized. The system 600 can transition between the open-loop and closed-loop flowback configurations through coordinated choke valve setting and/or bypass valve/choke routing to apply or remove surface backpressure while maintaining continuous monitoring and control of returns.

In some embodiments, the system 600 can automatically transition between open-loop and closed-loop flowback by selectively opening and closing the chokes 622 or 662 and/or the automated block valves 651 using a central control unit 700 as further described herein. In some embodiments, the central control unit system 700 can be configured to autonomously change the flowback operations between open-loop and closed-loop configurations based on the AI output (i.e. recommendations and/or instructions) as further described herein.

The AWFPS 600 further comprises a wellhead flow line 634 that carries the production flow 623 from the wellhead assembly 620 to one or more sand separators 636, also known as sand knockout units (“SKUs”). The wellhead flow line 634 can include: a) a sand volume detection sensor 638; and b) a flow rate sensor 640. The SKUs 636 can be of cyclonic, spherical or other design known to those in the art. In some embodiments, the flow rate sensor 640 can be a low-pressure flow meter on the returns line configured to measure the flow rate (e.g., mass or volumetric) of fluid exiting the wellbore 604 in both the open-loop (wellbore unpressurized) configuration and the closed-loop (pressurized wellbore) configuration to ensure seamless data continuity during transitions between operating modes. In some other embodiments, the flow rate sensor can be a flow meter (e.g., sensor 150, 640, or 657) on the returns line configured to measure the mass- or volumetric-flow rate of fluid exiting the wellbore 604 in both the open-loop (wellbore unpressurized) configuration and the closed-loop (pressurized wellbore) configuration.

The wellhead flow line 634 can further include one or more (redundant) junk/plug catcher 642 in the line to remove larger pieces of debris from production flow 623 and protect equipment down line. These junk/plug catchers 642 can include a downstream pressure sensor 643 and a differential pressure sensor 644 to monitor the delta pressure across the catcher to determine when to clean the debris from them. The system 600 can be configured with automated valves and bypass lines (not shown) on the wellhead flow line 634 to switch from one junk/plug catcher 642 to an alternate catcher or to temporarily bypass the catcher when the delta pressure measured by the pressure sensor 644 exceeds a predetermined limit.

As described above, the AWFPS 600 can further comprise a sand separator/SKU 636 on the wellhead flow line 634. A SKU inlet pressure sensor 646 and a SKU outlet pressure sensor 648 can be provided on the respective flow inlet and outlet of the sand separator 636, thus making available inlet, outlet and differential pressure data for each sand separator. A separate SKU differential pressure sensor 647 can also be provided between the inlet and outlet lines of the SKU 636. Further, each sand separator 636 can have a sand level sensor or sensors 650, to detect the level of sand 637 in the separator. In the illustrated embodiment, the sand separator 636 has a first sand level sensor 650a of direct measuring type, e.g., radar or ultrasonic, and a second sand level sensor 650b of indirect measuring type, e.g., densitometer or resistive paddle.

In some embodiments, the SKU 636 can be a single-entry SKU. Basic cyclonic and spherical single-entry SKUs can function well within a predetermined design range of flow rates and pressures. However, when the flow and/or pressure are not matched to the SKU's design range, a single-entry SKU may allow sand to flow over. Therefore, in other embodiments, the SKU 636 can be a multi-entry SKU (i.e., having more than one inlet). Cyclonic separators having more than one inlet allow for improved sand and solids removal over a greater range of flow rates and pressures.

The SKU 636 illustrated in FIG. 6 has two inlets (denoted inlet “1” and inlet “2”). The two inlets can be of similar size or different sizes. Each inlet can have a specified design flow rate and pressure range. In the illustrated example, the SKU 636 has a first nominal design flow rate of 300 BBL/Hr., 3 MMCF@3500 psi using inlet 1 only, a second nominal design flow rate of 100 BBL/Hr., 2 MMCF @3100 psi using inlet 2 only, and a third nominal design flow rate of 450 BBL/Hr., 8 MMCF @ 3800 psi using both inlet 1 and inlet 2 simultaneously. Block valves 651 can be provided on the dual inlet lines to inlet 1 and inlet 2 to select which inlets are used according to the expected flow rate and pressure. The block valves 651 can be manual valves or automated valves. In the illustrated embodiment, block valves 651 are automated valves, wherein block valve 651a controls SKU inlet 1 and block valve 651b controls SKU inlet 2. The automated block valves 651a, 651b can be operably connected to a central control unit 700 for automated or autonomous control as further described herein.

In some other embodiments, a SKU 636 can have three or more inlets and three or more respective manual or automated block valves 651 proving even greater flow rate and pressure flexibility. In still other embodiments the system 600 can include multiple single-entry SKUs 636 per well that are selectively accessed using respective manual or automatic block valves 651. In still further embodiments the system 600 can include two or more multi-entry SKUs 636 per well, with a manifold of respective manual or automated block valves 651 to control the flows and pressures at each SKU inlet to optimize sand removal over a wider range of flow regimes.

When automated SKU inlet block valves (e.g., 651a and 651b) are connected to a central control unit 700, the active inlets (e.g., inlet 1 and inlet 2) of the SKU 636 can be automatically selected based on the pressure and flowrates flowing into the SKU (e.g., determined from real-time signals received from pressure sensors P1, P2, flow sensor 640, sand sensor 638, etc.). Selective control of the active inlets in a multi-inlet SKU can provide optimal sand removal, improved separation efficiency, stabilized flow in multiphase applications, and/or reduced erosion. In some embodiments, the wellhead choke 622 upstream of the SKU 636 can also open and close under automatic control of the central control unit 700 to adjust the flow rate in the wellhead flowline 634 if the SKU 636 is receiving insufficient flow or excessive flow for the available inlet configurations of the system 600 (i.e., provided by one or more multi-inlet SKUs or multiple single-inlet SKUs).

The AWFPS 600 can further comprise a output production flowline 652 that carries the output production flow (denoted by arrow 653) from the fluid outlet of the sand separator 636 to a well testing choke manifold (WTCM) 654. This output production flowline 652 can include a sand volume detection sensor 656 that can detect the volume of sand bypassing the sand separator 636 and flowing through the flowline. The output production flowline 652 can include a multi-phase flow meter 657 that can detect the volume of fluid and/or gas that is flowing through the flowline.

In preferred embodiments, the well testing choke manifold 654 of the AWFPS 600 can be an automated well testing choke manifold (AWTM) including a test manifold inlet pressure sensor 658 at the inlet of the manifold assembly and a test manifold outlet pressure sensor 660 at the outlet. The AWTM test manifold 654 can further include at least one automated choke 662 to control the pressure and flowrate of the output production flow 653 from well through the production flowline 652. In some embodiments, the AWTM 654 includes a AWTM control unit 661 that can receive choke control signals from a remote device, change the choke settings of the automated choke 662 in accordance with the received choke control signals, collect data from external sensors including, but not limited to, the automated choke itself (e.g., current outlet size), the inlet and outlet pressure sensors 658, 660 (e.g., current pressures), and flow rate sensor 657 (e.g., current flow rates of gas, oil and/or water), and transmit signals representing the collected data to a remote device.

The AWFPS 600 can further comprise a downstream flowline 664 that carries the output production flow 653 from the outlet of well testing choke manifold/AWTM 654. The downstream flowline 664 can carry the output downstream flow 653 from the choke manifold/AWTM 654 to addition production equipment 666 including, but not limited to, heaters, production separators, dehydrators, and other treating equipment. The additional production equipment 666 can release a treated flow (denoted by arrow 668) including a mixture of gas, oil, and water, or it can release individual treated flows including, but not limited, to a line for gas flow 670, a line for oil flow 672, and/or a line for water flow 674. The treated flows 668, 670, 672, 674 can subsequently be routed to storage pits and tanks, flare line or pipelines.

The AWFPS 600 can include more than one sand separator 636 and more than one AWTM 654 if required to handle the volume of flow from a wellhead assembly 620.

The sand separator (SKU) 636 of the AWFPS 600 has a sand outlet 676 for releasing a sand flow (denoted by arrow 680) containing sand 637 separated by the SKU from the production flow 623. A blowdown line 678 is connected to the sand outlet 676 to carry the sand flow 680 from the SKU 636 to an automated sand dump choke 682 (ASDC). The ASDC 682 includes an automated sand dump choke 684 to control the rate of sand flow 680 released from the SKU 636 through the blowdown line 678. In some embodiments, there are multiple ASDCs 682 per blowdown line 678, so one ASDC can be bypassed for repair while another ASDC continues control of the sand flow 680. In other embodiments, a single ASDC 682 can include multiple automated sand chokes 684 so that one sand choke can be bypassed for repair while another sand choke continues control of the sand flow 680. This minimizes delay in automated SKU sand dumping operations.

The AWFPS 600 can further include one or more dump valves 686, 688 on the blowdown line 678. A sand outlet pressure sensor 677 can be provided on the blowdown line 678 between the sand outlet 676 and the dump valves 686, 688. In the illustrated embodiment, dump valve 686 is a manual valve and dump valve 688 is an automated valve. The dump valves 686, 688 are typically used to fully block or fully open the blowdown line 678 as needed but are not typically intended to modulate the rate of sand flow 680.

In some embodiments, the ASDC 682 includes a ASDC control unit 683 that can receive sand dump choke control signals from a remote device, change the sand dump choke settings of the automated dump sand choke 683 in accordance with the received sand dump choke control signals, collect data from external sensors including, but not limited to, the automated sand dump choke itself (e.g., current outlet size), the inlet and outlet ASDC pressure sensors 679, 696 (e.g., current pressures), and sand volume flow sensor 694 (e.g., current or cumulative sand flow), and transmit signals representing the collected data to a remote device.

The AWFPS 600 can further include a sand dump line 690 leading from the ASDC 682 to a dump tank/pit 692 for carrying the flow 680 of sand waste 691 (i.e., sand along with any waste water or oil). The sand dump line 690 can include a sand volume flow meter 694 and a sand line pressure sensor 696. The dump tank/pit 692 can include a dump tank level sensor 693 to sense the level of sand waste 691 in the tank. In some embodiments, data from dump tank level sensor 693 can be collected and compared to the data collected from the sand volume flow meter 694 for calibrating the system. The sand dump tank/pit 692 can include a gas detection sensor 699, to detect any gas being release from any given well's SKU during blowdown. The measured gas value from the gas detection sensor 699 is another potential input for the ASDC 682 and/or CCU 700 to respond to, and also to alert operations of the presence of potentially hazardous gas in the tank/pit. The sand dump line 690 can further include one or more automated block valves 698 to fully open or fully block the sand flow 680 in the sand dump line. In some embodiments, the automated block valve 698 can comprise a portion of an automated test subsystem that opens and closes the block valve (either alone or in coordination with additional valves) to determine if any of the SKU valves or the ASDC 682 are leaking.

In some embodiments, The AWFPS 600 can further include an ASDC test system that can open and close automated valves and/or chokes to determine if any of the SKU valves or the ASDC 682 or the automated sand choke 684 are leaking. In one embodiment, the automated dump valve 688 can comprise a portion of an automated ASDC test subsystem. In other embodiments, the automated ASDC test subsystem further includes the sand outlet pressure sensor 677 (upstream of the automated dump valve 688). In still other embodiments, the automated ASDC test subsystem further includes the blow down line pressure sensor 679 (between the automated dump valve 688 and the automated sand choke 684). In still other embodiments, the automated ASDC test subsystem further includes the sand line pressure sensor 696 (downstream of the ASDC 682). In yet other embodiments, the automated ASDC test subsystem further includes the automated block valve 698 (downstream of the sand line pressure sensor 696).

For reference, a description of an automated sand dump system for oil and gas wells is disclosed in United States Patent Application Publication US2023/0313660 A1, now issued as U.S. Patent No. U.S. Pat. No. 12,209,491B2, assigned to ADS Services, Inc. U.S. Pat. No. 12,209,491, issued Jan. 28, 2025, is hereby incorporated by reference in its entirety.

In different embodiments of the AWFPS 600, the automated sand dump choke (ASDC) 682 can have various different modes of operation in terms of: a) when the automated choke 684 opens to dump sand from the SKU 636; and b) how the automated choke opens to dump sand from the SKU.

Examples of when (i.e., under what conditions) the ASDC 682 initiates a sand dump sequence to dump sand from the SKU 636 include, but are not limited to:

• • a) On a time period basis (i.e., the sand dumps are executed according to a repeating time interval, wherein a single time interval is specified and the dump sequence is repeated upon expiration of each occurrence of the specified interval).
  • b) On a time scheduled basis (i.e., the sand dumps are executed according to a predefined schedule of input times, such that specific times, dates, or other temporal indicators are designated, and a sand dump is executed at each of the designated times. The schedule can comprise one or more discrete time points, including calendar dates, clock times, or combinations thereof);
  • c) On a level of sand in the SKU basis (i.e., the sand dumps are executed each time the level of sand in the SKU 636 reaches a designated level);
  • d) On the volume of sand measured into the SKU (i.e., the sand dumps are executed each time a designated volume of sand is determined to have entered the SKU 636);
  • e) On the delta pressure across the SKU (i.e., the sand dumps are executed each time the differential pressure between SKU inlet 646 and SKU outlet 648 reaches a designated value);
  • f) Manually initiated by a system user, locally or remotely; and
  • g) Triggered by some a smart algorithm estimating the SKU's needs to be dumped.

In embodiments where the ASDC 682 is connected to multiple SKUs 636, the ASDC can be configured to finish an ongoing sand dump sequence for a first SKU before starting another sand dump sequence for a different SKU.

Examples of how (i.e., the modes of opening and closing the automated choke 684) the ASDC 682 executes a sand dump sequence to dump sand from the SKU 636 include, but are not limited to:

• • a) The speed of opening and closing the automate choke;
  • b) The acceleration of opening and closing the automatic choke;
  • c) How far the automated choke opens before stopping or closing again;
  • d) Opening to hold a set flow rate of gas, liquid or sand
  • e) Opening and closing to maintain a set pressure at:
    • 1) the choke inlet;
    • 2) the SKU outlet 648;
    • 3) the SKU inlet 646;
    • 4) the well head 620;
    • 5) test manifold inlet pressure 658;
    • 6) test manifold outlet pressure 660;
    • 7) in the reservoir 606 (e.g., based on wellhead upstream pressure 624); and
    • 8) any combination of the above modes.
  • f) Opening and closing to:
    • 1) Maintain a set flow rate 680 from the SKU 636 to dump tank 692;
    • 2) To stay below a set pressure 696 in the choke dump/vent line 690;
    • 3) Closing if gas is detected in the sand dump flowline 690 after the choke;
    • 4) Closing after a set amount of sand has been flowed out of the SKU to the dump
    • tank (e.g., measured at 694);
    • 5) Other smart algorithms to control and opening and closing of the automated choke 684 during the blowdown of the SKU 636;
  • The opening and closing modes of the ASDC 682 can be selected and changed.

In some embodiments there can be multiple wells and multiple SKUs 636 being blown down by the one ADSC 682, and in other embodiments there can be one ASDC for each SKU.

In some embodiments, a multi-phase (e.g., 3 or 4 phase) flow meter 657 is included downstream of the SKU 636 (e.g., on line 652) or downstream of the AWTM 654 (e.g., on line 664) to measure the oil, water and gas mix flowing from the well and SKU.

In some embodiments, the AWFPS 600 includes a central control unit 700 that is operatively connected to one or more sensor points of the system, where a sensor point is any device capable of measuring a value of a parameter of interest (e.g., pressures, temperatures, flows, levels, orifice sizes, etc.) or detecting a value of a status of interest (e.g., open/closed status, on/off status, etc.) and producing a signal representing the measure value or detected status. For example, sensor points in the illustrated embodiment include devices 624, 626, 628, 630, 632, 633, 638, 640, 644, 646, 648, 650, 656, 657, 658, 660, 693, 694, 699, valves 688, 698, and chokes 622, 662 and 684. Sensor points can be operably connected to the central control unit 700 to transmit respective signals representing their respective measured values. Preferably, the central control unit 700 is connected to all available sensor points of the AWFPS system, however, for purposes of illustration, only the connections from the central control unit to the wellhead choke 622, AWTM 654, ASDC 682, and automated valves 688, 696 are shown in FIG. 6. It will be appreciated that some sensor points are associated with actuated devices as described herein. The sensor points can be operably connected to the central control unit 700 using any known method for transmitting signals including, but not limited to, electrical wires, optical fibers, radio, Wi-Fi, Bluetooth, or via a cloud application or distributed network such as the internet. Preferably, all of the sensors in the AWFPS 600 are adapted to communicate signals to the central control unit 700.

In some embodiments, the central control unit 700 can be a single discrete computer, whereas in other embodiments, the central control unit can comprise multiple computers operating in a distributed or networked fashion. The computer(s) comprising the central control unit 700 can be physically located at the surface site/pad 602, disposed at one or more remote location(s), or a combination of both. In some cases, the computer(s) comprising the central control unit 700 can be on-site or remote computers running AI engines, machine learning, expert systems or other platforms accessible via the internet, cloud or other distributed systems.

The central control unit 700 is further configured to communicate with one or more actuated devices of the AWFPS 600, where an actuated device is any device that can be remotely triggered or commanded to perform an operation or provide information regarding the device's status (e.g., actuated devices in the illustrated embodiment include the AWTM 654, ASDC 682, chokes 622, 662, 684, and valves 688, 698). Actuated devices can be operably connected to the central control unit 700 to receive control signals to allow the central control unit to control their function and to send telemetry signals representing their status or configuration to the central control unit. The actuated devices can be operably connected to the central control unit 700 using any known method for transmitting signals including, but not limited to, electrical wires, optical fibers, radio, Wi-Fi, Bluetooth, or via a cloud application or distributed network such as the internet. It will be appreciated that some actuated devices include, or are associated with, sensor points as described herein. Preferably, all of the actuated devices in the AWFPS 600 are adapted to communicate signals to and from the central control unit 700.

In some embodiments of the AWFPS 600, the central control unit 700 logs the data from each sensor point (“channel”) in the system at a minimum of 1 second frequency. In some embodiments, the frequency of logging each data channel can be adjusted from a few milliseconds to a few minutes depending on the control algorithm and reporting requirements.

In some embodiments, the AWFPS 600 is configured to utilize machine learning artificial intelligence (“AI”) algorithms to learn to control the well flow and pressures so that sand and water product are minimized while produced hydrocarbons are optimized

In some embodiments, the AWFPS 600 further includes one or more human-machine interface 702 (“HMI”) configured to interchanged data, signals, or control inputs from a human user to the system, e.g., via the central control unit 700. The HMI 702 can comprise known type of display units (screens, lights, alarms horns, etc.) and known types of input devices (keyboards, touch screens, joysticks, switches, etc.). In the illustrated embodiment of FIG. 6, HMI 702a includes a keyboard for input and display screen for output, whereas HMI 702b is a mobile device including a touch screen combining both input and output functions. The HMI 702 can be disposed at surface site/pad 602 or at a remote location and connected through In some embodiments of the AWFPS 600, the choke valve 662 of the automated well test manifold (AWTM) has the ability to control both flows and pressures of the production from SKU 636 and from the associated wellhead 620.

In some embodiments of the AWFPS 600, the system is configured to provide online reporting of historical and live data/activities from the well pad. Such online reporting can include, but is not limited to:

• • a) Sand production per well over time;
  • b) SKU Sand bypass per well over time;
  • c) Well pressure over time;
  • d) Well flow over time;
    • 1) Gas flow;
    • 2) Oil/Liquids flow;
    • 3) Water flow;
  • e) Reservoir Temperatures;
  • f) Annuli pressure and temperature;
  • g) Estimated production decline rate by well; or
  • h) Estimated time until well sand flow rate is low enough to remove SKUs.

In some embodiments of the AWFPS 600, the system is configured to provide system health alerts and/or alarms. Examples of the health alerts and alarms include, but are not limited to, the following:

• • a) A Choke or valve in the system is leaking and need service;
  • b) There is faulty sensor in the system;
  • c) A sensor needs to be calibrated;
  • d) A sensor is missing;
  • e) Sand is bypassing a SKU (and identity of the relevant SKU);
  • f) A plug/junk catcher needs cleaning (and identity of the relevant plug/junk catcher);
  • g) A line is leaking; or
  • h) Gas is detected in the sand storage pit/tank, coming from the sand dump line.

In some embodiments of the AWFPS 600, the system is configured to control well production pressures and flows by selectively controlling the AWTM 654 and/or wellhead choke 622 to optimize specified performance parameters. Examples of performance parameters to be optimized include, but are not limited to, the following:

• • a) Minimize Water production;
  • b) Optimize oil and gas production;
  • c) Minimize sand production; or
  • d) Minimize the time to clean up the well (i.e., time to reduce the sand production below a preset volume or weight per day such that the well can be placed on production without the need for SKUs and AWTM).

In some embodiments of the AWFPS 600, the system is configured to perform automated well testing procedures by selectively controlling the AWTC and or AWTM. Examples of well test procedures to be performed include, but are not limited to, the following:

• • a) Open to a set bean size (choke diameter) for a set time or until the pressure drops to a set level or a set flow;
  • b) Open and close to a set of bean sizes over time;
  • c) Open to maintain a set flowrate for a set time;
  • d) Open to maintain a set pressure (i.e., at well head or AWTM inlet or AWTM outlet) for a set time;
  • e) Open to a set position for a time until flow drops to a set flow and then close;
  • f) Open to a set position for a time to maintain a set pressure and then close when the pressure drops below the set pressure; or
  • g) Close until pressure reaches a set or steady level and then open to a set bean size, choke position, pressure or flow.

Aspects of the structure and operation of AWFPS 600 involve incorporation of new and existing multi-phase flow measurement sensor (MPFS) technologies and the use of data and measurements available from such MPFM technology. Some of the capabilities of MPFM technology were heretofore unknown and/or underappreciated in the industry and have been incorporated into the current system 600 and related processes in new and unexpected ways. Such MSFT may include, but are not limited to: a) advanced mass flow sensors for fluids with high percentage gas; b) advanced fluid flow sensors for high-pressure, non-intrusive use; c) advanced cuttings flow sensors measuring weight percent and volume in degassed return drilling mud; and d) advanced sand sensors for measuring sand volume and sand weight in flow lines. Many of these MPFM sensors have non-intrusive configurations, meaning that the sensor's presence does not obstruct the fluid flow in the flowline, e.g., either the sensor can function without being in physical contact with the measured fluid or the sensor elements are “wetted” but flush with the ID of the pipe and thus again does not obstruct the flow line.

In some embodiments, the AWFPS 600 can comprise advanced mass flow sensors (AMFS) for fluids with high percentage gas. AMFS sensors can measure total mass flow rate of fluid and gas, percent fluid/gas split and overall density measurement. These sensors can operate across a wide range of fluid types, densities, and gas conditions. During drilling, drill-out, or flowback there may be some percentage of entrained cuttings or sand measured as fluid, increasing apparent overall density. The AMFS can measure mass flows with high percentages of entrained gas up to 50%, and operating pressures from atmospheric to 3000 psi. In some embodiments of the AWFPS, either or both of flow sensors 640 and 657 can be AMFS sensors embodying MPFM technology.

In some embodiments, the AWFPS 600 can comprise advanced fluid flow sensors for high-pressure, non-intrusive use (HPNIS). Such sensors can be used as flow meters for high pressure drilling mud, water, gas, frac fluids, drill-out fluids, and production fluids. The HPNIS sensor detects the flow rate of mud and other fluids being pumped into or flowing from a well at high pressure, for example, in the range from greater than 500 psi up to 30,000 psi.

In some embodiments, the AWFPS 600 can comprise advanced cuttings flow sensors (ACFS) measuring weight percent and volume of cuttings in degassed return drilling mud or other fluids. The ACFS sensor can be a non-intrusive sensor. The ACFS sensor can measure the overall flow rate, and percentage and volume of cuttings in return flow drilling mud.

In some embodiments, the AWFPS 600 can comprise advanced sand sensors (ASDS) for measuring sand volume and sand weight in flow lines. The ASDS sensor can be non-intrusive sensor. The ASDS sensor can detect the flow rates, volume and weight of sand during fracking, plug drill-out, well clean-up, well production testing, and in system process flow lines. Some ASDS sensors can measure the overall process flow rate in addition to sand volume and weight, whereas other ASDS sensors measure sand only and utilize other sensors for measuring process flow. ASDS sensors can also be used in conjunction with multiphase flow meters (e.g., AMFS or other flow meters) to provide enhanced measurement of sand flow rates and process flows. In some embodiments of the AWFPS, any or all of the sand sensors 638, 656, and 694 can be ASDS sensors embodying MPFM technology.

Having described the system apparatus at the surface site 602 and tubing string 608, the details of the AWFPS system 600 will now be described with respect to the remaining figures, which provide visual representations of various components and processes of the automated flowback and production system. These figures illustrate the operational processes, network architecture, neural network module, and HMI, offering a comprehensive view of the system's structure and functionality. Through these illustrations, the intricate workings of the automated flowback and production system and its components will be elucidated, providing a clear understanding of how the system optimizes well construction operations.

Referring to FIG. 7A, the flowchart illustrates an overall operational process 701 of an automated flowback and production system. The process 701 can begin with a step 703, where data collection occurs. This step can involve collecting real-time data on wellhead conditions, production characteristics, sand recovery parameters, sand dump parameters, etc. from surface site apparatus, sensors, and actuated devices of the AWFPS 600 as described herein. For example, collecting data on pressure, temperature, mass flow and volume flow rates, fluid composition, sand weight etc. moving through different parts of the wellhead assembly 620, flowing into and out of the sand separators 636, and then to further processing 666 or the sand dump pit 692.

In some aspects, the data acquisition system of the AWFPS system can 600 include all of the various types of sensors described herein in connection with the surface site 602 to collect a wide range of real-time data. These sensors provide data that can be used to monitor the flowback operation and detect potential issues in real-time. In some cases, the system may also integrate data from third-party service providers involved in other aspects of the well construction process, which may include information from drilling, fracking, and drill-out operations for the subject wells or associated wells (e.g., from the same area, same formation, or with similar geology). May also include data from in hole pressure and temperature sensors. This comprehensive data collection approach allows for a more holistic view of the of the well construction and flowback operation, enabling more accurate analysis and decision making.

The process 701 can then move to a step 704 for data preprocessing. In this step, raw data can be cleaned, filtered, and formatted for compatibility with the system. Additionally, derived data points can be generated through calculations based on sensor data. In some aspects, the data preprocessing step can involve various data cleaning and filtering techniques to remove noise, outliers, or erroneous data points from the raw data. This can help to improve the quality and reliability of the data, which is beneficial for the subsequent analysis and decision-making steps. In some cases, the data preprocessing step can also involve data transformation or normalization procedures to convert the raw data into a suitable format for the digital twin models and the AI algorithms. This can include scaling the data to a range, converting categorical data into numerical data, or encoding time-series data into a suitable format.

In some cases, the data preprocessing step 704 can also involve feature engineering, where new features are created from the existing data to capture more complex relationships or patterns. This can include techniques such as polynomial features, interaction terms, or domain-specific transformations based on expert knowledge of the flowback process. Additionally, the preprocessing step 704 can incorporate data augmentation techniques to enhance the robustness of the AI models, such as adding controlled noise to the data or generating synthetic samples to balance underrepresented classes or conditions.

In addition to the above, the data preprocessing step 704 can also involve the generation of derived data points based on the raw data. These derived data points can be calculated using various mathematical or statistical operations, such as averages, sums, differences, ratios, or integrals. The derived data points can provide additional information that cannot be directly measured by the sensors but can be inferred from the raw data.

In some aspects, the derived data points can include complex calculations that combine multiple sensor readings to provide more comprehensive insights into the flowback process. These calculations may involve the identification and labeling of operational states, trends, or events in time, pressure, flow, and/or cumulative volume domains. For example, the system may use simple statistical methods or more advanced machine learning and AI techniques to classify flowback pressure states (“overpressure,” “open pressure”), sand production rate states (“nominal sand recovery,” “excessive sand recovery”) e.g., based on total sand used in fracking, or sand dump states (“no dump,” “modulated dump, full dump”).

The system may also identify and label trends in real-time, such as increasing or decreasing the well head pressure, the rate of production, the conditions for initiating a sand dump from the sand separator. These labeled data points may be associated with both time stamps and flowback depth measurements (when available), allowing for multi-dimensional analysis of flowback operations.

In some cases, the system may employ pattern recognition algorithms to detect and label specific events such as “sand dump valve opening (dynamic),” “sand dump valve open (static),” “sand dump valve closing (dynamic),” and “sand dump valve closed (static)”. These labeled events may provide beneficial context for understanding flowback performance and potential issues.

The automated flowback system may use both simple threshold-based rules and more sophisticated machine learning models to perform this labeling. For instance, a simple rule might label a sand dump event period as “elapsed time from initiation of dump until conclusion of dump.” In contrast, a machine learning model might use a combination of multiple sensor inputs to identify more complex operational states or events, for example the “dump valve opening phase” with time and acceleration parameters, “dump valve dwell phase” with time and size parameters, “dump valve closing phase” with time and deceleration parameters.

By generating these labeled data points in the time, pressure, flow and/or cumulative volume domains, the system may create a rich, annotated dataset that can be used for real-time decision making, post-well analysis, and continuous improvement of the AI models. This approach may enhance the system's ability to provide context-aware recommendations and optimize flowback operations based on a comprehensive understanding of past and current operational states and events.

The data preprocessing step can also involve the creation of dimensionless parameters or ratios that can provide information about the flowback process. For instance, the system can calculate numbers to characterize the flow regime in the wellbore, potential for entrainment of sand in the flow up the tubing, or the friction factor to assess the efficiency of fluid circulation. These derived parameters can help in making comparisons across different well zones or flowback operations, regardless of variations in absolute depth, absolute oil, gas and water production, absolute sand production, or other factors.

In some cases, the data preprocessing step can include the application of advanced signal processing techniques to extract meaningful information from complex sensor data. This can involve techniques such as Fourier transforms to analyze frequency components of vibration data, or wavelet analysis to detect transient events in pressure or flow rate measurements. The results of these analyses can be used to generate additional derived data points that capture specific characteristics of the flowback process.

Next, the process 701 can proceed to a step 706 for digital twin update. This step can involve feeding real-time data and the flowback plan into a digital twin framework. Digital twins can simulate specific aspects of the flowback process, and the framework can be continuously updated to reflect the current state of the well.

In some aspects, the digital twin framework can include multiple digital twins, each simulating a specific aspect of the flowback process. For example, one digital twin can simulate the dynamics of sand-rich fluid flow from the formation to the wellhead, another can simulate the dynamics of fluid flow of the produced flow exiting the wellhead and flowing through the surface production equipment, and yet another can simulate the removal of sand in the sand separator and periodic dumping of sand to the sand pit. Each digital twin can be updated in real time with the latest data from the sensors, allowing the digital twins to provide accurate and up-to-date simulations of the flowback operation.

In some cases, the digital twin framework can also incorporate the drilling plan, fracking plan, and/or drill-out plan into the simulations. These plans can provide information about the expected flowback parameters. By incorporating these additional plans into the digital twin simulations, the system can compare the actual flowback operation with the planned flowback operation, make predictions regarding expected values, and detect any deviations or anomalies.

In some aspects, the digital twin framework can include a complex network of interconnected digital twins, each focusing on a specific aspect of the flowback operation. For instance, a mechanical dynamics digital twin can simulate the vibrations, stresses, and torque experienced by the actuated chokes and block valves of the system during different flowback events. A fluid dynamics digital twin can model the flow of production fluids, water, and sand, including their pressure, velocity, and temperature throughout the wellbore and surface equipment. A formation digital twin can represent the geological characteristics of the formations from which the production flow and sand are being recovered, including its porosity, permeability, and strength.

These digital twins can work together in a hierarchical or parallel structure, exchanging information and influencing each other's simulations. For example, the mechanical dynamics digital twin can provide input on the choke or dump valve speed and duration to the fluid dynamics digital twin, which in turn can update the expected rate of emptying the sand separator. This information can then feed into the formation digital twin to adjust the wellhead choke flowrate to avoid recovering sand from the formation too quickly and clogging the separator equipment.

In some cases, the digital twins can employ various modeling techniques, such as finite element analysis, computational fluid dynamics, or discrete element methods, depending on the specific aspect being simulated. These models can be continuously refined and calibrated based on the real-time sensor data, improving their accuracy and predictive capabilities over time.

The digital twin framework can also include an integration module that synthesizes the outputs from individual digital twins. This module can use advanced data fusion techniques, such as Kalman filtering or Bayesian inference, to combine the results from different simulations and resolve any discrepancies or uncertainties.

In some aspects, the digital twin framework can incorporate machine learning algorithms to enhance its predictive capabilities. These algorithms can analyze historical data and patterns to identify correlations and trends that cannot be apparent in physics-based models alone. For instance, a neural network can be trained to predict choke wear based on a combination of simulated sand production from the well using data from the formation digital twin and historical performance data.

In other words, the digital twin framework can be multifaceted and probabilistic in nature. For example, the framework can provide a range of possible outcomes for a given flowback scenario, each with an associated probability based on the confidence levels of the individual digital twins and their integrated analysis. These conclusions can include predictions of flowback performance, risk assessments for potential issues like sanded out tubing, clogged sand separators, leaking chokes, and output (i.e. recommendations and/or instructions) for improved (e.g. optimal) flowback parameters.

In some cases, the digital twin framework can also perform sensitivity analyses, running multiple simulations with varying input parameters to identify factors affecting the flowback operation. This can help in prioritizing which aspects of the operation to focus on for optimization or risk mitigation.

The process 701 can then advance to a step 708 for real-time analysis and prediction. In this step, AI agents analyze the data and use physics models to predict potential issues. Digital twins can run parallel simulations incorporating operational data, and predictive models forecast outcomes of flowback operations under various scenarios.

In some aspects, the AI agents can use various machine learning algorithms to analyze the data and make predictions. These algorithms can include supervised learning algorithms, unsupervised learning algorithms, reinforcement learning algorithms, or deep learning algorithms. The choice of algorithm can depend on the type of data, the complexity of the problem, and the available computational resources.

In some cases, the AI agents can also use physics models to make predictions. These physics models can be based on the laws of physics, such as the laws of motion, the laws of thermodynamics, or the laws of fluid dynamics. The physics models can provide a theoretical basis for the predictions, complementing the empirical learning of the machine learning algorithms.

In addition to the above, the digital twins can run parallel simulations incorporating the operational data. These simulations can provide a virtual representation of the flowback operation, allowing the system to explore various scenarios and predict the outcomes of different flowback strategies. The simulations can also provide a visual representation of the flowback operation, which can be useful for monitoring and troubleshooting purposes.

The predictive models in the automated well flowback and production system can employ a combination of physics-based simulations and machine learning techniques to forecast outcomes of flowback operations under various scenarios. These models can utilize historical data, real-time sensor inputs, and digital twin simulations to make predictions about future flowback performance, potential issues, and improved (e.g. optimal) flowback parameters. In some cases, the predictive models can use ensemble methods, combining outputs from multiple algorithms to improve accuracy and robustness. The models can also incorporate uncertainty quantification techniques, providing confidence intervals or probability distributions for their predictions. As new data becomes available, the predictive models can continuously update and refine their forecasts, adapting to changing well conditions and improving their predictive capabilities over time.

The process 701 can then advance to a step 710 for integration of digital twin outputs. In this step, outputs from multiple digital twins can be aggregated by advanced AI models. The AI synthesizes information to determine well health and operational status. Patterns, correlations, and interactions are detected for comprehensive analysis.

In some aspects, the AI agent module of the automated flowback system can use various machine learning algorithms to analyze the data and make predictions. These algorithms can include supervised learning algorithms, unsupervised learning algorithms, reinforcement learning algorithms, or deep learning algorithms. The choice of algorithm can depend on the type of data, the complexity of the problem, and the available computational resources.

In some cases, the AI agent module can also use physics models to make predictions. These physics models can be based on the laws of physics, such as the laws of motion, the laws of thermodynamics, or the laws of fluid dynamics. The physics models can provide a theoretical basis for the predictions, complementing the empirical learning of the machine learning algorithms.

In some aspects, the AI agent module can employ advanced ensemble learning techniques to combine the outputs of multiple machine learning algorithms and physics models. This approach can leverage the strengths of different predictive methods, potentially improving the overall accuracy and robustness of the predictions. For example, a gradient boosting algorithm might be used in conjunction with a physics-based model to predict choke wear, with the machine learning algorithm capturing complex patterns in historical data while the physics model ensures adherence to known physical constraints. Another example would be to predict produced sand rates based on current and historical pressure and flow measurements supporting effective management of sand removal and handling by the AWFPS.

The AI agent module can also incorporate adaptive learning capabilities, allowing it to continuously refine its predictive models based on new data and feedback from the flowback operation. This can involve techniques such as online learning or transfer learning, where the AI agent can quickly adapt to changing well conditions or apply knowledge gained from one well to improve predictions for another.

In some cases, the physics models used by the AI agent module can be enhanced with uncertainty quantification methods. These methods can provide probabilistic predictions that account for uncertainties in input parameters, model structure, and measurement errors. This can be particularly beneficial in flowback operations, where many factors are uncertain or difficult to measure precisely.

The digital twin simulations can be further enhanced with real-time optimization capabilities. As the simulations can run in parallel with the actual flowback operation, they can continuously explore different operational scenarios and suggest improved (e.g. optimal) flowback parameters. This can involve techniques such as model predictive control or reinforcement learning, where the AI agent learns to make sequential decisions that optimize long-term flowback performance.

The process 701 then proceeds to a step 712 for decision-making and adjustment output (i.e. recommendations and/or instructions) for well and AWFPS health and safety management. This step involves AI evaluating combined data to make predictions and recommend flowback adjustments. The system monitors for potential well and AWFPS control events and recommends corrective actions. Operators receive instructions to mitigate risks and prevent non-productive time. A real-time optimization engine suggests flowback parameters for efficiency and safety. The HMI 702 displays output (i.e. recommendations and/or instructions) and allows for operator intervention.

In some aspects, the real-time optimization engine of the automated flowback system can use various optimization algorithms to suggest ideal flowback parameters. These algorithms can include linear programming, nonlinear programming, genetic algorithms, or swarm optimization algorithms. The choice of algorithm can depend on the type of problem, the complexity of the flowback operation, and the available computational resources.

In some cases, the real-time optimization engine can also use feedback control strategies to adjust the flowback parameters in real-time. These strategies can include proportional-integral-derivative (PID) control, model predictive control, or adaptive control. The choice of control strategy can depend on the dynamics of the flowback operation, the accuracy of the digital twin models, and the performance requirements of the flowback operation.

The process 701 then proceeds with a step 714 for autonomous operation and human oversight. The system autonomously adjusts flowback operations based on AI output (i.e. recommendations and/or instructions). Operators can intervene and manually adjust parameters if desired. The process can include continuous improvement through secure cloud connectivity and historical data learning.

In some aspects, the automated flowback system can include 700, a Central Control Unit system that can be configured to automate the flowback operations based on the AI output (i.e. recommendations and/or instructions). This control system can include various components, such as programmable logic controllers (PLCs), hydraulic power units, valve actuators, advanced automation devices and safety systems. The control system can also interface with various flowback equipment, such as the wellhead choke 622, automated well test manifold 654, automated sand dump choke 682, and automated block valves 688, 698.

In some cases, the automated flowback system can also include an HMI 702 that provides a user-friendly interface for the system operators. The HMI 702 can display real-time data, AI output (i.e. recommendations and/or instructions), and control options. The HMI 702 can also allow the operators to manually intervene in the flowback process, providing a balance between automation and human oversight.

In some aspects, the AI's evaluation of combined data for predictions and flowback adjustment output (i.e. recommendations and/or instructions) can involve a sophisticated process of data fusion and multi-modal analysis. The AI can integrate real-time sensor data, historical well information, digital twin simulation outputs, and physics-based model predictions to create a comprehensive understanding of the current flowback state. This holistic approach can allow the AI to identify subtle patterns and correlations that might not be apparent when analyzing individual data streams in isolation. The AI can employ advanced techniques such as deep learning, reinforcement learning, or Bayesian inference to process this complex, high-dimensional data and generate actionable insights. These insights can include predictions of future flowback performance, potential risks or issues, and improved (e.g. optimal) flowback parameters for various scenarios. For purposes of objectively measuring the improvements provided by the AWFPS system and improvements provided by the AI and digital twin aspects of the system, flowback performance can be quantifiably defined in terms of values for one or more measurements (i.e., “metrics”) including, but not limited to, the measures/metrics listed below.

One quantitative measure of flowback performance is the number of occurrences of well control events, e.g.;

• • a) overall number of well control events,
  • b) number of well control events per time unit (i.e., events per day), or
  • c) or number of well control events per flowback interval (i.e., events per hour of active flowback)).

Another quantitative measure of flowback performance is the number of health, safety and environmental (HSE) incidents or near misses, e.g.,

• • a) overall HSE/safety incidents,
  • b) number of HSE/safety incidents per unit time (i.e., incidents per day), or
  • c) number of HSE/safety incidents per flowback interval (i.e., incidents per hour of active flowback)).

Another quantitative measure of flowback performance is efficiency of sand management, e.g.,

• • a) weight of sand bypassing SKUs (absolute or relative measure),
  • b) percentage of sand entering each SKU separated from fluid flow, or
  • c) percentage of frac sand recovered during flowback.

Another quantitative measure of flowback performance is minimization of equipment erosion/wear, e.g.,

• • a) mean time between failure (MTBF) for chokes, valves, SKUs and other equipment;
  • b) maximize equipment uptime (e.g., absolute operation life during flowback, or
  • c) percentage of flowback time operationally ready).

Another quantitative measure of flowback performance is improved cleanup and transition speed, e.g.,

• • a) absolute time for cleanup,
  • b) cleanup time relative to frac sand/water load, or
  • c) cleanup time normalized to the amount of injected frac water or sand load
  • d) cleanup time relative to onset of stable flowback production
  • e) the time required to reach a stable, sand-free production regime.

Another quantitative measure of flowback performance is minimization of formation damage, e.g.,

• • a) percentage of actual production relative to engineering predictions
  • b) percentage of formation productive zones actually producing, or
  • c) percentage of percentage of production
  • d) the relative reduction in permeability or production efficiency due to fines or over-drawdown.

Another quantitative measure of flowback performance is minimization of environmental discharges and costs, e.g.,

• • a) total number of environmental discharges,
  • b) total cost for remediation of environmental discharges,
  • c) total weight or volume of environmental discharges,
  • d) environmental discharges per unit time of flowback.

Another quantitative measure of flowback performance is maximization of reservoir and flowback data gathering and learning,

• • a) total gigabytes of real-time measurement data collected,
  • b) gigabytes of real-time measurement data collected per sensor on system,
  • c) number of simultaneous simulations (including digital twins) running using real-time updates during real-time operations,
  • d) total gigabytes of simulation data collected during real-time operations,
  • e) gigabytes of simulation measurement data collected per simulation,
  • f) gigabytes of real-time measured control data collected, or
  • g) gigabytes of real-time control signals recorded during real-time operations.

Another quantitative measure of flowback performance is maximization of initial production at hand over, e.g.,

• • a) absolute production at hand over; or
  • b) percentage of predicted production available at hand over.

Another quantitative measure of flowback performance is pressure and flow stability performance, e.g.:

• • a) bottomhole pressure deviation relative to target setpoint,
  • b) magnitude and frequency of flowrate oscillations,
  • c) choke differential pressure variance (%),
  • d) average pressure ramp rate (dP/dt), or
  • e) the number of pressure excursions beyond allowable envelopes.

Another quantitative measure of flowback performance is fluid recovery and cleanup efficiency, e.g.:

• • a) the percentage of total injected frac fluid recovered,
  • b) the rate of fluid recovery per day or per stage, or
  • c) the number of days required to reach 80% total fluid recovery.

Another quantitative measure of flowback performance is reservoir characterization and diagnostic learning, e.g.:

• • a) calculated stage contribution index (SCI),
  • b) fracture connectivity index (FCI),
  • c) reservoir connectivity index (RCI),
  • d) inferred pore pressure accuracy (psi deviation vs model), or
  • e) number of diagnostic events or underbalance tests completed successfully.

Another quantitative measure of flowback performance is operational and AI learning system metrics, e.g.:

• • a) total number of digital twin state updates per hour,
  • b) real-time error between measured and simulated pressure or flowrate,
  • c) number of AI-driven control recommendations executed autonomously,
  • d) model prediction accuracy (e.g., mean absolute percentage error, MAPE), or
  • e) the cumulative number of adaptive control cycles completed per operation.

Another quantitative measure of flowback performance is economic and performance efficiency, including for example:

• • a) total operational cost per barrel of fluid recovered ($/bbl),
  • b) overall cost per day of flowback relative to production volume, or
  • c) reduction in non-productive time (NPT) or downtime percentage relative to baseline operations.

It will be appreciated that the quantitative measures of flowback performance described herein are not exhaustive, but are merely exemplary of the measures that can be used by the AI to continuously refine its predictions and output (i.e. recommendations and/or instructions) as new data becomes available, adapting to changing well conditions and improving its decision-making capabilities over time.

It is noted that the AI agents described above can be deployed using a distributed architecture, with components running both on-site at the well pad surface site 602 and in cloud-based environments. The on-site deployment can include edge computing devices that process real-time sensor data and execute time-sensitive control algorithms, ensuring rapid response to changing flowback conditions. These edge devices can host lightweight versions of the AI models, optimized for low-latency decision-making. Meanwhile, more computationally intensive tasks, such as complex simulations and long-term trend analysis, can be performed in the cloud, leveraging scalable computing resources.

The deployment strategy can also incorporate a hierarchical structure, with different AI agents specializing in specific aspects of the flowback operation. For example, lower-level agents can focus on individual subsystems like wellhead chokes 622 or sand dump chokes 684, while higher-level agents coordinate overall flowback strategy using the automated well test manifold 654 and risk management. This modular approach can allow for easier updates and maintenance of the AI system, as individual components can be refined or replaced without disrupting the operation. Additionally, the system can employ containerization technologies to ensure consistent performance across different hardware environments and facilitate seamless updates to the AI models as new algorithms or data become available.

Having described the overall operational process of the automated flowback system, FIG. 7B will be described which illustrates the specific actions taken by flowback operators within this advanced flowback system. This figure provides a detailed breakdown of the operator's role in the system, highlighting the points of human interaction and decision-making that complement the AI-driven processes. By examining these operator actions, we gain a clearer understanding of how human expertise and oversight are integrated with the autonomous capabilities of the system, ensuring safe, efficient, and effective flowback operations.

Referring to FIG. 7B, the flowchart illustrates operator actions process 720 in an automated flowback system. The process 720 can begin with a step 722 for system initialization, where the operator powers on the automated flowback system and the system performs self-diagnostics and sensor checks. This initialization step ensures that components of the system are functioning correctly before flowback operations commence. In some cases, the system can perform additional checks or calibrations, such as verifying the accuracy of the sensors or testing the communication links between different components.

The process 720 can then move to a step 724 for operator input. In this step, the operator inputs or loads the well flowback plan, and the system integrates the plan into the digital twin framework. The well flowback plan can include various details about the intended flowback operation, such as the number of zones, frac sand pumped into each zone, the expected production composition and flow from each zone, and the desired flowback parameters. The operator can input this plan through the HMI 702, using various input devices such as a keyboard (702a), a mouse, or a touchscreen (702b). In some cases, the well flowback plan can be pre-loaded into the system from a database or a file, or it can be received from remote data sources, thereby reducing manual input.

Next, the process 720 can proceed to a step 726 for real-time monitoring. The operator observes real-time data and AI output (i.e. recommendations and/or instructions) on the HMI 702, while the system continuously collects and processes sensor data. The real-time data can include various flowback parameters, such as the wellhead pressure, temperature and flow, the wellhead choke settings, production flow rates and composition, sand production rates, sand separation rates, sand volumes, sand level in the SKU, and sand dump status. The AI output (i.e. recommendations and/or instructions) can include suggested adjustments to the wellhead choke settings, to the automated well test manifold 654 parameters, intervals and duration of sand dump events and operation of the automated sand dump choke 682, warnings about potential flowback hazards, or predictions about the flowback performance. The operator can monitor this information in real-time on the HMI 702, allowing them to maintain situational awareness and make informed decisions.

The process 720 then advances to a step 728 for parameter adjustment. The operator reviews the AI-suggested flowback parameter adjustments and can either accept the automated adjustments or manually override them as needed. The AI-suggested adjustments are generated by the AI agent module, which uses the real-time data, the digital twin simulations, and the historical well data to produce informed output (i.e. recommendations and/or instructions). The operator can review these output (i.e. recommendations and/or instructions) on the HMI 702 and decide whether to accept them or make manual adjustments. This step provides a balance between automation and human oversight, allowing the operator to leverage the AI's capabilities while maintaining control over the flowback operation.

Following this, the process 720 can move to a step 730 for flowback progress tracking. The operator monitors the flowback progress against the flowback plan, and the system provides updates on well health and potential issues. The flowback progress can be tracked in terms of various metrics, such as wellhead pressure and temperature, casing annuli temperatures and pressures, the production rates of oil, gas, water, and sand, the sand recovery volume/weight relative to the volume/weight of frac sand injected into the well, the separation effectiveness of the sand separator, the rate of sand dumping, etc. The system can provide updates on the well health based on the sensor data and the AI analysis, alerting the operator to any changes in the well conditions or potential flowback hazards. This step allows the operator to keep track of the flowback progress and respond promptly to any issues.

The process 720 can then proceed to a step 732 for well control event management. The system alerts the operator to potential well control and flowback events and the operator follows the AI-recommended actions or implements manual intervention. Well control and flowback events can include situations, e.g., loss of pressure or flow control, loss of well flow, clogged sand separator, leaking chokes and block valves, which can pose risks to the flowback operation and the well integrity. The system can detect these events based on the sensor data and the AI analysis, and it can alert the operator through the HMI 702. The AI can also recommend corrective actions, such as adjusting the wellhead choke 622, well test choke 662, sand dump choke 684, bypassing one junk/plug catcher 642 so it can be cleaned out, switching to an alternate SKU 636, or changing the flowback platform parameters (e.g. control the AWTM 654 or ASDC 682, etc.), or shutting in the well. The operator can follow these AI-recommended actions or implement their own manual intervention, depending on their judgement and expertise.

Next, the process 720 can advance to a step 734 for completion of flowback for a selected well/well section/zone. The operator confirms the completion of the flowback section/zone or the entire well, and the system generates final reports and data analysis. The operator can confirm the completion of flowback through the HMI 702, marking the end of the flowback operation. The system can then generate final reports summarizing the flowback performance, the well data, and any issues encountered. The system can also perform a final data analysis, comparing the actual flowback results with the flowback plan and the digital twin predictions. This step provides a comprehensive summary of the flowback operation, allowing for post-flowback review and learning.

The process 720 then proceeds with a step 736 for continuous learning. The operator can provide feedback on the system performance, and the system updates the AI models and the historical database for future operations. Throughout the flowback operation, operators continuously provide comments, data labeling, and feedback, enabling real-time system refinement and optimization. The operator can provide feedback through the HMI 702, commenting on the system's performance, the accuracy of the AI output (i.e. recommendations and/or instructions), or any issues encountered. The system can then use this feedback to update the AI models, improving their learning and prediction capabilities. The system can also update the historical database with the new well and flowback data, enriching the data set for future operations. This step ensures that the system continuously learns and improves over time, enhancing its performance and reliability for future flowback operations.

In some aspects, the continuous learning step 736 can involve a multi-faceted approach to system improvement. The operator feedback can be categorized into different types, such as accuracy of predictions, timeliness of alerts, effectiveness of recommended actions, and overall system usability. This categorization can allow for targeted improvements in specific areas of the system's performance.

The AI models can employ advanced machine learning techniques, such as transfer learning or meta-learning, to efficiently incorporate new knowledge without compromising previously learned information. This approach can enable the system to adapt to new flowback scenarios, fracking histories, or geological formations while retaining its expertise in familiar situations.

In some cases, the system can implement a form of active learning, where it identifies areas of uncertainty in its predictions and actively seeks operator input or input from offsite or third-party entities on these specific issues. This targeted approach to gathering feedback can accelerate the learning process and improve the system's performance in challenging well flowback operations.

The historical database update process can involve sophisticated data management techniques, such as data versioning and provenance tracking. These techniques can allow the system to maintain a comprehensive record of how its knowledge evolves over time, enabling rollbacks to previous states if beneficial and providing insights into the system's learning trajectory.

In some aspects, the continuous learning process 636 can extend beyond individual well flowback operations to incorporate cross-well and cross-field learning. The system can identify patterns and best practices across multiple flowback operations, potentially leading to broader insights into improved (e.g. optimal) flowback strategies for different geological formations, fracking histories, or operational conditions.

The system can also implement a form of explainable AI, allowing operators to understand the reasoning behind the AI's output (i.e. recommendations and/or instructions) and predictions. This transparency can foster trust between the operators and the AI system, encouraging more frequent and detailed feedback and ultimately leading to more effective continuous improvement.

In some aspects, the operator actions process 720 can include additional steps or variations. For example, the process can include a step for equipment setup or calibration before the system initialization. The process can also include a step for data backup or archiving after the completion of flowback operations. The process can include additional monitoring or control steps, such as monitoring the equipment health, controlling the equipment and sensor systems, or coordinating with other personnel. The process can also include steps for handling emergencies or unexpected events, such as activating safety systems, implementing emergency procedures, or coordinating with emergency response teams. These additional steps or variations can enhance the operator's control over the flowback operation and the system's ability to handle a wide range of flowback scenarios.

The automated flowback system 600 may operate with varying levels of autonomy, adapting to the operator's preferences and the system's capabilities. In some cases, the system might make fully autonomous adjustments to flowback parameters without requiring operator intervention based on AI instructions. In other instances, it may generate suggested adjustments that need confirmation or acceptance by the operator before implementation. The system can also provide recommendations that require manual input from the operator, or issue alarms and alerts that may or may not necessitate operator action, depending on the severity and nature of the situation. As the digital twin models, AI components, and associated systems become more intelligent over time through continued training and programming, the system may evolve to become increasingly autonomous, potentially reducing the need for frequent operator intervention while maintaining the option for human oversight and control when beneficial.

For example, in the context of flowback operations, the spectrum of automation levels may be applied to various aspects of the flowback process. At the lower end of the spectrum, the system 600 may function as a computer-based aid by providing real-time data on flowback parameters such as the wellhead pressure, temperature and flow, the wellhead choke settings, production flow rates and composition, sand production rates, sand separation rates, sand volumes, sand level in the SKU, and sand dump status, allowing the driller to make informed decisions. As the level of automation increases, the system may offer advisory functions, such as suggesting improved (e.g. optimal) flowback parameters based on formation characteristics and historical data.

Moving further along the automation spectrum, the system 600 may make decisions and execute actions with varying degrees of human oversight in specific flowback operations. For example, in computer-based decision-making, the system may determine the improved (e.g. optimal) flowback parameters based on real-time testing and evaluation data (e.g., from the AWTM 654) but inform the operator who can override the decision if beneficial. At higher levels of automation, the system may perform computer-based execution by automatically adjusting flowback parameters such as wellhead choke settings, production flow rates, permissible sand level in the SKU, and sand dump parameters, to maintain the improved (e.g. optimal) result of flowback (i.e., maximizing long term productivity of the well), while the operator monitors and has the option to intervene.

The most advanced level may involve full automation of certain flowback processes, such as sand dumping operations or pressure management during managed pressure flowback. In these cases, the system may perform all tasks without human intervention, including adjusting chokes, dumping sand, test for valve leakage, and maximizing production. However, the capability for human oversight and intervention may still be maintained as a safety measure, allowing the operator to take control in case of unexpected events or system anomalies. This range of automation levels in flowback operations allows the automated flowback system to be tailored to specific well complexities, crew expertise levels, and company risk management policies.

In some embodiments of the AWFPS 600, the system is configured to perform automated wellbore annulus pressure relief. During well flowback and production operations the heat of the well reservoir production can lead to over pressurization of fluids sealed in the outer casing annuluses 614, 616 of the well. If these pressures get too high it can lead to production tubing failure, casing failure, or cement failure; potentially resulting in well damage, unnecessary well service work, or a leaking well head that could cause loss of well control and environmental damage. To prevent such overpressure from occurring, the AWFPS 600 is configured to monitor the pressures of the outer well annuluses (e.g., via sensors 630, 632) and to automatically bleed off pressure from the relevant annulus to a lower level, thereby avoiding any damages or leakages. The bleed-off operation can be controlled by the central control unit 700, which actuates one or more automated valves to open the relevant well anulus to an automated choke or relief valve that releases fluid to controllably bleed the pressure down to the desired level. Once the desired pressure level is reached the central control unit 700 closes the automated valves to reseal the well annulus. There can be one or multiple wells being managed by the system on each surface well pad/site.

In another aspect, the AWFPS involves the novel use of Digital Twins (“DTs”) for automating and optimizing well flowback, cleanup, and testing operations. Some of these DTs may be related to other phases (besides flowback) of the overall well construction process, but the data, models, and AI agents associated with these DTs of the other phases may still be utilized in connection with the automated well flowback operations. For example, the following DTs can be incorporated into automating and optimizing flowback, cleanup, and testing operations of the AWFPS: a) reservoir DT; b) wellbore/hydraulics DT; c) flowback system processes DT; d) integrated operations supervisory/predictive layer DT; and e) AI agent control layer DT.

In one aspect, the AWFPS comprises the combination of a historical data build, digital twins simulations, machine learning (“ML”), and AI agents to provide autonomous well flowback operations. Further, the incorporation of quantitative sand metering (i.e., contrasted with conventional qualitative metering) and the addition of improved sensors (e.g., using MPFM technology) and automation components in the AWFPS yield improvements in efficiency and optimization of processes that exceed conventional control systems.

For example, the AWFPS may incorporate a “Reservoir Digital Twin” simulating formation, wellbore and frac interaction. The purpose of this DT is to simulate pressure and saturation transients during cleanup to optimize how quickly the well is “opened up” to the reservoir. The core DT models for this DT can include: a porous-media flow model; a frac connectivity model to estimate which stages are cleaning up first; a reservoir pressure and saturation tracking based on surface responses (i.e., observable from the wellhead); and a dynamic reservoir twin comprising a continuously updated full 3D reservoir model. The outcomes on flowback system management and reservoir data from this modeling can include: optimized drawdown scheduling, avoiding damage by adjusting drawdown rates to protect fines, collapse of proppant pack, and fracture integrity; estimating contribution of each fracture stage during early production (drill-out data supported); optimized choke schedule for balanced cleanup across stages. (drill-out data supported); stage-by-stage cleanup forecasting, estimating which fracture stages clean up first, gas breakthrough timing, relative contribution (drill-out data supported); pressure transient matching, i.e., matching the well test or flowback decline response to infer formation permeability, fracture conductivity, skin, and connectivity; production forecasting, i.e., projecting stabilized flow rates and cumulative recovery from the early transient behavior; anomaly detection and diagnosis (for example, if flow deviates from forecast, the DT can suggest whether the deviation is due to channeling, plugging, crossflow or other potential cause; and defining well operating boundaries including probabilistic forecasts (e.g. 90% confidence intervals) to guide safe operating envelopes.

For another example, the AWFPS may incorporate a “Wellbore Digital Twin” simulating downhole hydraulics and transients. The purpose of this DT is to model transient multiphase flow inside the wellbore during unloading and production stabilization, from total depth (TD) to surface. The core DT models for this DT can include: a transient multiphase flow simulator; a gas-liquid-solid transport model to estimate slugging, and sand carryover; a thermal model (i.e., where temperature affects viscosity or hydrate generation risk); coupled with a surface twin for full-system pressure and mass balance. The outcomes on flowback system management and reservoir data from this modeling can include: optimized pressure and flow management to support well clean up, well testing, and ultimate production “open up”; predicted choke response to bottomhole drawdown; optimized flowback ramp-up rate to avoid formation damage or water hammer; detecting and preventing onset of gas coning, sand slugging, or liquid loading; and predicting heating and over pressure of outer annuli.

For another example, the AWFPS may incorporate a “Flowback System Processes Digital Twin” simulating surface flow and separation. The purpose of this DT is real-time optimization of flowback choke and separator systems, surface equipment, manifold pressures, chokes, separators, tanks, flare, and sand traps. The core DT models for this DT can include: a dynamic multiphase flow model that simulates choke pressure drops, separator residence time, gas-liquid ratios, and sand separation effectiveness and erosion; a sensor inputs model for single and multi-phase flow rate (mass/volumetric), pressure, temperature, sand rate, valve positions, separator levels; a control layer model for PID or AI-based control to adjust choke openings, pump speeds (drill-out), and sand dump valve sequences. The outcomes on flowback system management and reservoir data from this modeling can include: preventing separator overflow during high flow surge conditions; preventing sand erosion of equipment and flowlines; automatically tuning choke to maintain constant wellhead pressure; optimizing sand dump timing to eliminate (or minimize) manual intervention; optimizing flow to SKUs to avoid sand bypass and protect downstream equipment; quantifying in real time four-phase mass and volumetric flow rates from well and through process stages; and optimizing choke positions to support reservoir management digital twin and well behavior.

For another example, the AWFPS may incorporate a “Integrated Operational AI-Driven Digital Twin” simulating supervisory and predictive layers. The purpose of this DT is to integrate data-driven machine learning (“ML”) with physics-based twins to predict and prevent upsets. The core DT models for this DT can include: one or more machine learning models, e.g., trained on historical reservoir, wellbore and surface flowback datasets (e.g., pressure, flow, gas, oil water, sand, ECD, delta-flow)); an anomaly detection and predictive maintenance model for valves, separators, and chokes; and an optimization engine for automated wellhead, test and sand dump choke control, SKU, tank management, and flare minimization, the outcomes on flowback system management and reservoir data from this modeling can include: predicting when sand load will exceed SKU capture and automated dump thresholds; forecasting separator pressure surges ahead of time; recommending choke step adjustments based on drawdown curve; identifying and predicting flow back system equipment failures; and optimizing surface flowback system operational behavior to support requirements of reservoir, wellbore and process digital twin/AI agent automation.

For another aspect, the AWFPS 600 incorporates an “integrated AI-driven cognitive well cleanup, testing and optimization control agent layer” (the “Agent”). Utilizing the well design and historical data, planned and real-time digital twins, the Agent can ultimately autonomously design, execute, and adapt well cleanup and testing operations that minimize damage, maximize reservoir connectivity, and accelerate the well's transition from transient to an optimal stable production state, while protecting surface and subsurface equipment. The Agent learns and “understands” each well's cleanup behavior, predicts what it will do next, and controls it to safely and optimally reach peak production. The Agent learns over time: multiphase transient dynamics; reservoir-well interaction pattern; sand production management and erosion risk; flow back separation and measurement system component behaviors; evaluates and improves completion design; well flowback response and optimal flowback program

In some embodiments, the integrated AI-driven cognitive well cleanup, testing and optimization control agent layer can perform the following operations: simulate before acting (via digital twin ensemble detailed in prior steps); predict reservoir response and surface outcomes; learn from every well, forever improving; generate automated well test reports with analytics and KPI scoring; minimize safety hazard and risk exposure during operations. The Agent can produce the quantifiable outcomes over current manual and basic control automation including, but not limited to: reducing clean up time; reducing sand carryover and separator plugging; avoiding reservoir damage incidents (respecting drawdown limits); improving reservoir early production rates; improving reservoir estimated ultimate recovery (EUR) accuracy; reducing staff on site and their safety exposure; evolve to fully autonomous well cleanup, testing, and commissioning system.

Referring to FIG. 8, the flowchart illustrates an overall operation process 850 for automated wellbore annulus pressure relief. The process 850 begins with a step 852 representing initiation of the process, which can be part of the overall startup of the AWFPS 600, or it can be separately initiated by inputs from the HMI 702. The process 850 can then move to a step 854, wherein pressure limit (PL) values for each annulus (e.g., annuli 614, 616) can be loaded into the system. In some embodiments, the loading step 854 can include loading pre-defined PL values from a well design data set, whereas in other embodiments, the PL values for each annulus can be entered via the HMI 702. The process 850 can then move to a step 856, wherein data is received from a sensor on each annulus, for example annulus pressure sensors 630, 632, the data including actual pressure (AP) values for each annulus.

After performing step 346, the process 850 can then move to a decision step 858, wherein the respective AP values are compared to respective PL values for each annulus. When the comparing step 858 determines the AP value exceeds the PL value for a selected annulus, the process 850 moves along the “YES” path to a step 860, wherein a command is transmitted to open a relief valve on the selected annulus to release fluid from the annulus. The relief valve can be an automated valve or choke operatively connected to the annulus space (e.g., 614 or 616) and controlled by the central control unit 700. The “YES” path also leads to a step 862, wherein the overpressure event results in sending an alarm message and/or logging data regarding the overpressure event. The alarm message of step 862 can cause sounding a physical alarm at the surface site 602, displaying an alarm signal on the HMI 702 or other actions. The data logging of step 862 can cause recording of selected data parameters of the AWFPS 600 at the time of the overpressure event for future analysis, or for input into the digital twin framework of the AWFPS or other well control systems. After performing steps 860 and 862, the process 850 proceeds back to the comparing step 858 for another comparison of the AP value to the PL value for the selected annulus.

When the comparing step 858 determines the AP value does not exceed the PL value for a selected annulus, the process 850 moves along the “NO” path to a step 864, wherein a command is transmitted to close the relief valve on the selected annulus to seal the annulus. After performing step 864, the process 850 can proceed to a step 866, wherein the AI analysis or real-time digital twin model(s) running in the digital twin framework can be updated with data on the overpressure event. The step 866 can improve the performance of the AI analysis or digital twin in predicting the likely behavior of the flowback operation. After performing step 866, the process 850 proceeds back to the receiving data step 856 to receive updated data from the sensors on the annulus. It will be appreciated that during each cycle through the steps of data receiving (856), comparing (858), and controlling (860, 862, 864), the process 850 can monitor the same annulus as previously monitored or a different annulus than previously monitored.

In some embodiments of the AWFPS 600, the system is configured to perform well site data acquisition and utilize AI/digital twin control schema. In these embodiments, the AWFPS 600 provides an automated well flowback system designed to optimize well flowback, cleanup and testing and production operations in the oil, gas, and geothermal industries. This system integrates a data acquisition system, a digital twin framework, an artificial intelligence (AI) agent module, an integrated well and flowback control system, and an optional human-machine interface (HMI). The AWFPS 600 of such embodiments is configured to monitor and control multiple wells on a given well pad site and with different equipment configurations. The system can also be configured to provide data and communication integration and remote control across multiple well pad sites.

Referring to FIG. 9, the flowchart illustrates an overall operation process 900 for automated management one or more sand separators (SKUs) collectively having multiple inlets to support variable well flowrates during flowback. The method is suitable for managing a single SKU 636 having multiple inlets or multiple SKUs having one or more inlets each, provided that collectively, the controlled SKUs have multiple inlets. Each of the inlets to be controlled is equipped with a respective automated block valve 651 that allows the process to open or close each inlet. The process 900 begins in block 902 by measuring the flow rate from the well, which can comprise gas, fluids and sand. In some embodiments, block 902 further includes measuring the pressure from the well. The process then proceeds to block 904 wherein the controller 700 evaluates the data, e.g., measured flow rate and (if available) pressure to size and select SKU inlet flow path(s) using predetermined data regarding each inlet's capacity over a range of conditions. The process then proceeds to block 906 wherein the controller adjusts the automated inlet block valves to route the flow to one or more of the available SKUs (if there are multiple SKUs available). The process then proceeds to block 908 wherein the controller further adjusts the automated inlet block valves to route the flow to one or more inlets on each SKU (for SKUs having multiple inlets). For example, depending on the measured flow rate, the flow to one multi-inlet SKU could be routed entirely to inlet A, entirely to inlet B, or simultaneously to inlet A and inlet B to effectively process the flow rate measured.

In block 908, the process further checks for abnormal conditions including, but not limited to: a) well flowrate exceeding total SKU(s) capacity; b) well flowrate below minimum SKU operational limit (e.g., too low to effectively remove sand); and c) detection of sand bypassing the SKU(s) into the output line. When no abnormal condition is detected in block 908, the process proceeds via return line 914 back to block 902 to repeat the main process loop.

When block 908 detects the well flowrate is exceeding the total SKU(s) capacity, the process proceeds to block 910, wherein the process further determines if sand bypassing the SKU is detected. The process then proceeds to block 913 wherein the process automatedly closes the well head (e.g., using the wellhead choke 622) or test manifold choke (e.g., AWTM choke 662) enough to reduce flow to within the collective SKU(s) capability. The process then proceeds via return line 914 back to block 902 to repeat the main process loop.

When block 908 detects the well flowrate is below the minimum SKU operational limit, the process branches down line 911 to block 912, wherein the process further determines if sand bypassing the SKU is detected. The process then proceeds to block 915 wherein the process automatedly opens the well head (e.g., using the wellhead choke 622) or test manifold choke (e.g., AWTM choke 662) enough to increase the flow to within the effective sand removal operating range of the SKU(s). The process then proceeds via return line 916 back to block 902 to repeat the main process loop.

When block 908 detects sand bypassing the SKU(s) into the output line, the process branches down line 917 to block 918, wherein the process determines that sand is bypassing the SKU(s) even though the total flow is between the system's maximum and minimum capacities. The process then proceeds to block 920, for evaluating the well flow, sand readings, and system settings. When block 920 determines the SKU inlet path setup is out of the ideal range for the flow (i.e., the flow is not within the capacity range of the currently active SKU inlets), the process branches to block 922. Branching to block 922 indicates the process should change the inlet setting selection and/or flow rate to return the SKU(s) into effective operation. The process then proceeds via return line 423 back to block 902 to repeat the main process loop.

When block 920 determines the SKU inlet path setup is within the ideal range for the flow (i.e., the flow is within the capacity range of the currently active SKU inlets), the process branches to block 924. Branching to block 924 indicates the process has detected a problem with the system and/or operation of the SKU(s). The process then proceeds to block 926 wherein the process alerts the user/operator that the SKU(s) and/or the system is not operating correctly and needs maintenance or repair. The process next proceeds to block 928 wherein the affected SKU units are bypassed (if possible) or the system is shut down pending the user/operators intervention to resolve the mechanical/physical system issues.

In some embodiments of the AWFPS 600, the system incorporates a version of the cloud infrastructure located on-site at the well(s), enabling local processing and storage capabilities. Additionally, the system 600 can be configured to connect to off-site infrastructure, such as cloud services or other remote systems, to support off-site learning, updates, and modifications to digital twin models and well plans. This hybrid architecture allows for remote work and collaboration while ensuring that the system 600 can operate independently without an internet connection. The ability of the system 600 to function autonomously, without relying on external connectivity, is beneficial for the system's commercial viability and scalability, particularly in remote or challenging environments where internet access may be limited or unreliable.

In some embodiments of the AWFPS 600, the system further includes a data acquisition system that collects real-time data from surface and downhole sensors, including but not limited to multiphase flow meters, pressure sensors, temperature sensors, valve and choke positions, and sand weight and flow sensors. The collected data can then be preprocessed to clean, filter, and format it for compatibility with the system 600.

In some embodiments of the AWFPS 600, the system is configured to run digital twin models within the automated flowback system to continuously generate forward-looking data covering multiple potential (predicted) operational scenarios. The models may operate at high speeds locally, leveraging the on-site processing capabilities of the system 600. By constantly or periodically creating and updating simulations for multiple possible (predicted) flowback conditions and outcomes, the system can learn to anticipate and prepare for a wide range of situations in real-time. This operational mode enables the AI agent module to make rapid, informed decisions and recommendations, enhancing the system's ability to optimize flowback operations and respond proactively to changing well conditions.

In one exemplary configuration, the AI agent module of the AWFPS 600 can analyze the collected data from the sensor points and actuated devices and use physics models to predict potential issues. The system can further integrate the outputs from multiple digital twin simulations to detect complex patterns and interactions. In this manner, the system's digital agents are able to learn from historical data and simulation data which is described in more detail below.

The integrated flowback control system of AWFPS 600 can automate operations based on the output (i.e. recommendations and/or instructions) from the AI agents and the real-time optimization engine. This allows the system 600 to smoothly transition between SKUs 636 being dumped and well testing objectives, enhancing the system's ability to adapt to changing well conditions without interruption of well cleanup, testing and production. The HMI 702 can provide a user-friendly interface for well operators to monitor real-time data, receive AI output (i.e. recommendations and/or instructions), and manually intervene in the process if desired.

The components of the AWFPS 600 operate with one another to continuously monitor and control well conditions, leading to safer, faster, and more efficient well cleanup, testing and production. The system's comprehensive well health monitoring capabilities can enable real-time assessment of wellbore multiphase production output, fluid dynamics, and formation characteristics. By integrating data from multiple sensors and digital twin simulations, the AI agent module can detect subtle changes in well health indicators, allowing for proactive adjustments to flowback parameters to maintain optimally improved (e.g. optimal) well conditions throughout the flowback process.

More specifically, as will be described in more detail, the AWFPS 600 and associated methods can leverage cloud-based learning, utilizing a vast historical well data repository to continuously improve machine learning algorithms. Real-time AI-driven optimization can combine live data, physics models, and AI to enhance flowback parameters for safer, more efficient well clean up, testing and production. Digital twin integration can provide real-time simulations of well and site components, allowing the AI to anticipate and address potential issues proactively.

In another exemplary embodiment, the AWFPS 600 is configured with a digital twin framework to create sophisticated virtual models of the wellbore flowback operations. These models can be continuously updated with real-time data, allowing for accurate simulations of various components and scenarios. In some aspects, the system 600 can also collect external data inputs from the well and other well site service companies. For instance, in some embodiments, the system can receive and incorporate data regarding heating, dehydration, and/or multiphase separation from specialized service providers. The integration of external data sources to data collected directly can enhance the accuracy and comprehensiveness of the digital twin models, potentially enabling more precise simulations and predictions for the flowback operations.

In some embodiments of the AWFPS 600, the system is configured with artificial intelligence and one or more digital twin frameworks for optimization of the well flowback process. The AWFPS 600 can then operate according to a method using the output of digital twin(s) to emulate and control the flow from and pressure of the well. Next, outputs from multiple digital twins can be aggregated by the system's advanced AI models. In another step, the system's AI synthesizes information to determine the producing well health, cleanliness and operational status. In another step, patterns, correlations, and interactions are detected by the system's AI for comprehensive analysis.

In some embodiments, the system 600 can be configured with an AI agent module of the automated flowbacks system and can use various machine learning algorithms to analyze the data and make predictions. These algorithms can include supervised learning algorithms, unsupervised learning algorithms, reinforcement learning algorithms, or deep learning algorithms. The algorithm incorporated for a particular embodiment can be selected based on the type of data, the complexity of the problem, and the available computational resources.

In some embodiments, the system's AI agent module can be configured to include physics models to make predictions. These physics models can be based on the laws of physics, such as the laws of thermodynamics and the laws of fluid dynamics. The physics models can provide an additional (theoretical) basis for the prediction made by the AI agent, complementing the empirical learning of the machine learning algorithms.

In some embodiments, the AI agent module of the system 600 can be configured to employ advanced ensemble learning techniques to combine the outputs of multiple machine learning algorithms and physics models. This configuration synthesizes a new prediction based on the strengths of multiple different predictive methods, potentially improving the overall accuracy and robustness of the predictions. For example, a production boosting algorithm might be used in conjunction with a physics-based model to minimize sand and water production, with the machine learning algorithm capturing complex patterns in historical data while the physics model ensures adherence to known physical constraints.

In some embodiments, the AI agent module of the system 600 can be configured to incorporate adaptive learning capabilities, allowing the system to continuously refine its predictive models based on new data and feedback from the sensors and actuated devices on the actual flowback operation. This can involve techniques such as online learning or transfer learning, where the AI agent can quickly adapt to changing well conditions or apply knowledge gained from one well to improve predictions for another.

In some cases, the physics models used by the system's AI agent module can be configured with uncertainty quantification methods. These uncertainty quantification methods can provide probabilistic predictions that account for uncertainties in input parameters, model structure, and measurement errors. Such configurations can be particularly beneficial during control of flowback operations, where some parameters and factors have high uncertainty or difficult-to-measure values.

When the system is configured to include one or more digital twin frameworks, the digital twin simulations can be further enhanced with real-time optimization capabilities. As the simulations can run in parallel with the actual flowback operation during real time, they can continuously explore different operational scenarios and suggest (predicted) optimal improved (e.g. optimal) flowback parameters. By continuously comparing the results of the digital twin predictions to the actual measured values in real time or near-real time, the AI agent can utilize model predictive control or reinforcement learning, where the AI agent learns to make sequential decisions that optimize long-term well production performance and well clean up.

The systems and methods described can further be configured to include a step for decision-making and adjustment output (i.e. recommendations and/or instructions) for well health and safety management. This step involves AI evaluating combined data to make predictions and recommend flowback system adjustments. The system 600 monitors for potential unoptimized water and sand production levels or high-pressure flow and pressure events and recommends or automates corrective actions.

The system 600 can be configured to receive instructions, e.g., from human operators via the HMI 702 or from internal or external automated agents, to mitigate risks, optimal well health and prevent non-productive time during flowback, testing and cleanup operation. A real-time optimization engine can suggest flowback parameters for efficiency and safety. The local and remote HMI 702 can display output (i.e. recommendations and/or instructions) and allow for operator intervention when needed or desired.

The system will also use these AI and digital twin methods to monitor and report on the health of all the sensors and hardware in the system.

The system 600 can further be configured to analyze the data collected from each well's flowback operation with regard to the time taken and procedures used until the produced sand was reduced to acceptable levels (i.e., where normal production equipment can operate such that the expensive flowback and testing equipment can be removed). This analysis can compare data on the flowback/testing pressures used, flow rates used, and fracking process used on each well to corresponding data from other wells that have been fracked, flowed back, and cleaned up and use AI tools to and make recommendations for optimum frack methods and/or flowback procedures to use on subsequent wells to minimize the time to “clean up the well”, i.e., minimize the produced sand and maximize production volumes.

It will be appreciated that incorporation of machine learning, AI agents, and digital twin modeling frameworks within an automated flowback and production system such as AWFPS 600 can result in improved fracking effectiveness, shorter well testing and cleanup activities, improved operational safety and hardware life and optimized well production over the life of the well.

Referring now to FIG. 10, wherein like reference numbers are used to designate like elements throughout, the various views and embodiments of an automated frac plug drill-out and testing system 1000 are illustrated and described, and other possible embodiments are described. The figures are not necessarily drawn to scale, and in some instances the drawings have been exaggerated and/or simplified in places for illustrative purposes only. One of ordinary skill in the art will appreciate the many possible applications and variations based on the following examples of possible embodiments.

Referring first to FIG. 10, the fluid flow path (i.e., circulation path”) of the plug drill-out (often referred to as “drill-out”) and testing system 1000 is illustrated. As in all well-drilling-related processes, whether in hydrocarbon reservoir formation drilling, geothermal rock drilling, plug drill-out (e.g., post frac), sidetrack drilling, multi-lateral drilling, the same basic subsurface circulation path is followed, namely: pumping a fluid (“mud”) from a wellhead 1001 at the surface site/pad 1002 down through the internal bore of a tubular string 1004, out the end of the string, and then back up to the wellhead through the annular space (i.e., “annulus”) 1006 between the outer surface of the string and the inside of a wellbore 1008 formed into the hydrocarbon reservoir 1010 or other target formation. Once back to the wellhead 1001 at the surface site 1002, the fluid/mud circulation path can continue through surface equipment including pressure maintenance devices, debris separators, sand separators, gas separators, oil separators, temporary storage pits. The circulation path is completed by pumps or compressors that send the fluid back downhole via a tubular running system 1009, e.g., a rig's top drive, snubbing unit, or coil injector, drawworks, etc. operatively connected to the drill string 1004.

The drill string 1004 used with the drill-out system 1000 can include jointed pipe, coil tubing, or casing and be injected into the wellbore 1008 by the rig with top drive, snubbing unit or coil injector unit with the well's surface pressure (i.e., measured at the surface) at zero or at a positive pressure (e.g., when managed pressure is being maintained). The drill string 1004 used for plug drill-out may be the same string or a different string from the drill string used during primary drilling of the initial wellbore. The wellbore 1008 used with the drill-out system 1000 can comprise a cased hole, with or without cement, an open hole, or other well configurations.

The drill string 1004 may include one or more sensors disposed downhole to sense and measure the downhole conditions including, but not limited to, downhole pressure, temperature, geological values (e.g., resistivity, density, porosity, etc.), and steering values (e.g., direction, angle, etc.). For example, a bottom hole sensor unit 1012 can be mounted on a downhole motor unit 1014 or bottom hole assembly (“BHA”) disposed at the lower end of the drilling string 1004. The sensor unit 1012 can measure downhole pressure, temperature, geological, and/or steering values, and data signals regarding such value measurements can be transmitted by the sensor unit (e.g., using fluid pulses, wired pipe, etc.) to receivers at the surface site 1002, which are also parts of the system 1000. In some embodiments, the downhole pressure data and temperature data from the sensor unit 1012 can be used along with surface pressure data measured by other elements of the drill-out system 1000 (or other systems, e.g., drilling system 100, AWFPS system 600) to estimate additional characteristics of the reservoir 1010 including reservoir size, performance (i.e., flowrate/pressure) and reserves. The downhole pressure data and temperature data from the sensor unit 1012 along with surface pressure data can be collected while drilling out plugs, as well as during drilling, tripping, and during other stages of the well construction.

At the surface site 1002, the drill-out system 1000 can include injection side equipment including pumps 1016 and/or compressors to energize and/or pressurize the fluid/mud 1018 circulating down the drill string 1004. The circulating drilling fluids/mud 1018 can be made from a wide range of materials, for example compressed air and nitrogen, hydrocarbon foams, water, oils, density and viscosity additives, etc. During the mud's return to surface, the pressure in the wellbore annulus 1008 can be maintained by means of a flow barrier or restrictor placed along the circulation path including, but not limited to, a blow-out preventer (“BOP”) 1020, a rotating control device (“RCD”) 1022, injector heads, snubbing BOPs, as well as a choking system 1024 to hold various surface pressures on the well. The aforementioned devices are portions of a wellbore pressure management system. Collectively, these devices form a sealing system configured to selectively seal the wellbore 1008 to pressurize the wellbore to facilitate a closed-loop drilling/plug drill-out configuration and to unseal the wellbore to unpressurize the wellbore to facilitate an open-loop drilling/plug drill-out configuration. Thus, the system 1000 enables closed-loop managed pressure drilling/plug drill-out operations with a pressurized wellbore 1008 when sealed and conventional open-loop drilling/plug drill-out operations with an unpressurized wellbore when unsealed.

During closed-loop drill-out (i.e., wellbore is pressurized), the RCD bearing seal 1022 establishes annular sealing so that the choke manifold 1024 can regulate surface backpressure to track a BHP setpoint, while the low-pressure returns meter (e.g., 150, 1042, 1044) continuously measures outflow to support mass balance and control decisions. When the RCD bearing seal 1022 is removed or bypassed, the system 1000 reverts to open-loop drill-out operations (i.e., wellbore is unpressurized) without loss of outflow measurement.

In some embodiments, the system 1000 can automatically transition between open-loop drill-out configuration (i.e., with unsealed/unpressurized borehole) and closed-loop drill-out configuration (i.e., with sealed/pressurized borehole) by selectively unsealing and sealing the RCD's bearing assembly and/or opening and closing the chokes 1024 and/or the automated block valves 1065 using a central control unit 1070 as further described herein. In some embodiments, the central control unit system 1070 can be configured to autonomously change the drill-out operations between open-loop and closed-loop configurations based on the AI output (i.e. recommendations and/or instructions) as further described herein.

In some embodiments, the drill-out system 1000 can be provided with pre-frac dynamic testing data collected during pre-frac drilling operations (e.g., from drilling system 100). For example, pre-frac dynamic testing data can be collected as the drill string was injected into selected sections of the formation, controlled flow was established (e.g., using managed pressure drilling (“MPD”)) and the corresponding downhole pressure and temperature data and surface flow data were collected to assess or determine pre-frac formation characteristics including, but not limited to, pre-frac pressure profiles, radial permeability values and/or porosity values, for the selected formation sections. In some embodiments, the drill-out system 1000 can be provided with in-frac dynamic testing data collected during frac operations, including but not limited to section by section plug and perforation details, types and volumes of fluids and proppants pumped, pressures, flow rates, etc. The drill-out system 1000 can collect post-frac dynamic testing data. For example, while drilling out frac plugs 1025, the drill-out system 1000 can obtain post-frac dynamic testing data by controlling circulation flow into different sections of the formation during drill-out, establishing controlling flow from the open sections of formation (e.g., using the choke 1024), collecting corresponding downhole pressure and temperature data (e.g., from sensor unit 1012) and surface flow data, and assessing or determining post-frac formation characteristics including, but not limited to, post-frac pressure profiles, radial permeability values and/or porosity values, for the selected formation sections. In some embodiments, the controlled flow for dynamic testing while drilling and/or while drilling out frac plugs 1025 is achieved in underbalanced mode with the surface equipment configures as described herein.

The surface equipment used in various drilling and plug drill-out configurations on the flow out side (i.e., after return from the wellhead 1001) can include, but are not limited to, chokes 1024, valves, junk catchers 1026, hydro-cyclones/sand separators 1028, multi-phase pressurized separators (e.g., gas/liquid separator 1030 and oil/water separator 1032), multiphase flow meters, non-intrusive sand meters, non-intrusive mass and/or flow meters. The surface equipment of the drill-out system 1000 can accurately quantify (e.g., by weight and/or volume) the sand, solids, and produced hydrocarbon fluids or geothermal fluids received from the well during operations. The multiphase separators 1030, 1032 of the drill-out system 1000 (or the drilling system 100 or other systems described herein) can receive and separate the produced fluids (e.g., gas, oil, and produced water) and circulating return flow (e.g., mud) so the produced fluids can be directed into production flow lines 1034, flares 1036, surface/waste tanks 1038 holding excess water/mud 1039 and drilling tanks/suction pit 1040 holding reusable water/mud 1041, etc. In some embodiments of the drill-out system 1000, one or more of the multiphase separators 1030, 1032 can be four phase pressurized separators.

In some embodiments, the drill-out system 1000 includes a returns flow meter (e.g., sensor 150, 1042, 1044) positioned on the flow-out line to measure the flow rate of fluid exiting the wellbore 1008 in both open-loop (wellbore unpressurized) and closed-loop (wellbore pressurized) drill-out configurations. In some embodiments, this sensor 150, 1042, 1044 can be a low-pressure flow meter that provides continuous mass- or volumetric-flow measurement across open-loop/closed-loop mode transitions. In other embodiments, the sensor 150, 1042, 1044 can be a flow meter for inflow measurements for flow balance, influx/loss detection, and managed pressure control during drill-out.

The drill-out system 1000 can use data acquired by the autonomous drilling system 100 during drilling operations, for example, measured downhole pressure data, surface pressure data, and flow data acquired while drilling at known locations (i.e., depths) in a near balanced or underbalanced condition (“drilling system data”), to identify and/or predict the location of micro fractures and productivity of reservoir rock and additional reservoir characteristics. This drilling system data can be used for current and future frac program design, development, and optimization, through all phases of well construction through to well commissioning. The drill-out system 1000 can compare the drilling system data to measured downhole pressure data, surface pressure data, and flow data acquired during drill-out of each successive frac plug 1025 (“drill-out system data”) to determine quantitative measures of well construction performance including, but not limited to, frac efficiency by section, plug placement efficiency, reservoir production by section, decline curves, reservoir clean-up program efficiencies, more efficient drawdown programs to maximize frac efficiencies, production curves, and reserve estimates.

In some embodiments, the drill-out system 1000 includes a central control unit 1070 that is operatively connected to one or more sensor points of the system, where a sensor point is any device capable of measuring a value of a parameter of interest (e.g., pressures, temperatures, volume flows, mass flows, multiphase flows, gas %, liquids %, solids %, levels, orifice sizes, revolutions, weights, etc.) or detecting a value of a status of interest (e.g., open/closed status, on/off status, rpm status, etc.) and producing a signal representing the measure value or detected status. For example, sensor points in the illustrated embodiment include sensor devices (“INS”) 1012, 1014, 1015, 1042, 1044, 1046, 1048, 1050, 1051, 1056, 1057, 1058, 1059, 1060, 1061, 1062, 1063, automated valves 1065, chokes 1024, and pumps 1016. Sensor points can be operably connected to the central control unit 1070 to transmit respective signals representing their respective measured values. Preferably, the central control unit 1070 is connected to all available sensor points of the drill-out system 1000. The sensor points can be operably connected to the central control unit 1070 using any known method for transmitting signals including, but not limited to, electrical wires, optical fibers, radio, Wi-Fi, Bluetooth, or via a cloud application or distributed network such as the internet. Preferably, all of the sensors in the drill-out system 1000 are adapted to communicate signals to the central control unit 1070.

In some embodiments, the central control unit 1070 can be a single discrete computer, whereas in other embodiments, the central control unit can comprise multiple computers operating in a distributed or networked fashion. The computer(s) comprising the central control unit 1070 can be physically located at the surface site/pad 1002, e.g., on the rig 1009, disposed at one or more remote location(s), or a combination of both. In some cases, the computer(s) comprising the central control unit 1070 can be on-site or remote computers running AI engines, machine learning, expert systems or other platforms accessible via the internet, cloud or other distributed systems.

The central control unit 1070 is further configured to communicate with one or more actuated devices of the drill-out system 1000, where an actuated device is any device that can be remotely triggered or commanded to perform an operation or provide information regarding the device's status. In the illustrated embodiment, the actuated devices include the pump 1016, actuated valves 1065 on the inlet, outlet, or dump lines of the separators 1028, 1030, 1032, junk catcher 1026, and at other locations, chokes 1024, rig winch (for WOB), turntable (for torque, RPM, etc.). Actuated devices can be operably connected to the central control unit 1070 to receive control signals to allow the central control unit to control their function and to send telemetry signals representing their status or configuration to the central control unit. The actuated devices 1016, 1024, 1065, etc. can be operably connected to the central control unit 1070 using any known method for transmitting signals including, but not limited to, electrical wires, optical fibers, radio, Wi-Fi, Bluetooth, or via a cloud application or distributed network such as the internet. It will be appreciated that some actuated devices include, or are associated with, sensor points as described herein. Preferably, all of the actuated devices in the drill-out system 1000 are adapted to communicate signals to and from the central control unit 1070.

Referring now to FIGS. 11A and 11B, there is illustrated a method 1100 for automated well evaluation and determination of quantitative measures of well construction performance in accordance with another aspect. The method 1100 utilizes data obtained from pre-frac dynamic testing as described in blocks 1102 to 1117 and data from post-frac dynamic testing as described in blocks 1118 to 1133. The pre-frac dynamic testing begins with block 1102 and involves drilling into successive formation sections n_F=1, 2, . . . , N_F (block 1104), establishing controlled flow from the open formation section 1 to section n_F (block 1106), and collecting real-time pre-frac downhole conditions data and surface conditions data corresponding to the controlled flow from the open formation section 1 to section n_F (block 1108). The drilling, flowing, and data collection steps are repeated (i.e., via blocks 1112, 1114 and return path 1116) for each successive formation section until drilling of all N_F sections is completed with n_F=N_F (block 1112). The pre-frac real-time data from the pre-frac dynamic testing is used to determine pre-frac well characteristics for the formation for each section n_F=1, 2, . . . , N_F (block 1117). The determining block 1117 of the pre-frac testing can take into account that when establishing flow from each new section nr, the sections (n_F−1) through 1 will still be open. For example, when dynamically testing formation section 2, section 1 will still be open, thus the test values collected will be for the combined sections 1+2. To determine the pre-frac test values attributable to section 2 alone, the pre-frac test values previously collected from testing section 1 alone must therefore be subtracted from the pre-frac test values of combined sections 1+2. Similar calculations can be used to determine the pre-frac test values for all n formation sections where n=1, 2, . . . , N_F.

In some embodiments, the pre-frac testing data to be used for the method 1100 of automated well evaluation and determination of quantitative measures is collected (for example, by the autonomous drilling systems 100 described herein) from the same well that the post-frac testing data is collected from by the automated drill-out system 1000. However, in other embodiments, the pre-frac testing data may be associated with a different well in the same reservoir, a different well in a nearby location, or another well having similar geological characteristics. In still other embodiments, the pre-frac testing data can be synthesized data created by an AI agent based on training from pre-frac testing of other wells or parameters based on other pre-frac testing, e.g., seismic mapping, etc.

The post-frac dynamic testing begins at block 1118 (FIG. 11A) and involves drilling out successive frac plugs n_P=1, 2, . . . , N_P (block 1120, FIG. 11B), establishing controlled flow from the open formation section 1 to section n_P (block 1122), and collecting real-time post-frac downhole conditions data and surface conditions data corresponding to the controlled flow from the open formation section 1 to section n_P (block 1124). The drilling, flowing, and data collection steps are repeated (i.e., via blocks 1128, 1130 and return path 1132) for each successive frac plug section until drilling out of all N_P frac plugs is completed with n_P=N_P (block 1128). The post-frac real-time data from the post-frac dynamic testing is then used to determine post-frac well characteristics for the formation for each section n_P (block 1133). The post-frac testing can take into account that when establishing flow from each new frac plug section n_P, sections (n_F−1) through 1 will still be open. For example, when dynamically testing frac plug formation section 2, section 1 will still be open, thus the test values collected will be for the combined sections 1+2. To determine the post-frac test values attributable to section 2 alone, the post-frac test values previously collected from testing section 1 alone must therefore be subtracted from the post-frac test values of combined sections 1+2. Similar calculations can be used to determine the post-frac test values for all n formation sections where n_P=1, 2, . . . , N_P.

After the pre-frac test data, when available, and the post-frac test data are obtained, the method continues to block 1134, wherein the system determines actual well efficiency values for all sections by comparing pre-frac dynamic testing data to post-frac dynamic testing data for equivalent sections. In some embodiments, the locations, lengths, and spacing of the pre-frac formation sections n_F=1, 2, . . . , Nr will correspond identically to the locations, lengths, and spacing of the post frac plug drill-out sections n_P=1, 2, . . . , N_P, however, in other embodiments, the locations, lengths, or spacing may differ. In the latter case, the block 1134 can include a transform function substep wherein each post frac plug drill-out sections n_P=1, 2, . . . , N_P is mapped to one or more of the corresponding the formation sections n_F=1, 2, . . . , Nr based on collected information regarding the locations, lengths, and spacing if the respective tested sections. For example, if the two frac plugs were set in each formation section covered by a single pre-frac drilling test, then the transform function of block 1134 could map post-frac sections 1 and 2 (combined) to pre-frac section 1, map post-frac sections 3 and 4 (combined) to pre-frac section 2, and so on. In some embodiments, the pre-frac data can be obtained from pre-frac dynamic testing (e.g., performed by drilling system 100) as previously described. In other embodiments, the pre-frac data can be obtained from the reservoir's owner or developer, wherein data on production, reservoir pressure, temperature, pressure declines, etc. can be synthetic data based on reservoir modelling, actual data based on logging, or a mix of synthetic and actual data. In some embodiments, the pre-frac data can be supplied from the frac company with estimated improvements from the frac that was performed.

After the post-frac testing sections are mapped to the equivalent pre-frac testing sections, block 1134 continues comparing pre-frac dynamic testing data, when available, to post-frac dynamic testing data for equivalent sections to determine well efficiency parameters. Such parameters can include, but are not limited to, percent flow increase post-frac, percent increase in post-frac BHP retention for a given flow period, or other increases in post-frac production measures. These increases can be determined by comparing the expected flow or expected reservoir performance, i.e., the flow or performance that was expected based on data from the pre-frac or in-frac data received from the frac company or reservoir department, to the actual data from each section post frac, e.g., during drill-out. Thus, the system provides the ability to evaluate each section's production data in multiple ways, e.g., production versus pressure decline, production versus sand/proppant being produced, etc. The system also allows the evaluation of the frac efficiency by section, evaluation of reservoir production efficiency versus drilled wellbore length, and productivity by section versus lateral length, etc. The system then proceeds to block 1136, wherein actual well efficiency values (from the dynamic tests) are compared to the predicted well efficiency values from digital twin models. The difference between the actual efficiency values versus predicted efficiency values can be used as feedback to update (or “tune”) the digital twin models to improve the accuracy and performance of the digital twin models. The difference between the actual efficiency values versus predicted efficiency values also be used for modifying the frac design (i.e., for subsequent frac operations), e.g., by adjusting frac parameters including, but not limited to, the type/grade of proppant pumped, the weight/volume of proppant pumped, flow rates, pressures, lateral lengths, perforation and plug positions and number of fracs per lateral length

Referring again to FIG. 10, some of the new innovative capabilities of the drill-out system 1000 are achieved through application and use of the combination of new and existing sensor technologies. Innovative operations, testing and data capabilities are enabled by multi-phase sensor technology (“MPST”), which can provide real-time outflow cuttings measurement, sand measurement, oil, water, and gas measurements in conjunction with the “ESP” methods and automation to enable continuous operational control and monitoring from open to closed to open loop flow system utilizing continuous pressure and flow management technology (“PMT”).

Enabling improvements in automated well construction and control by novel combination and use of new and existing multi-phase flow measurement sensor technologies (“MPST”) is illustrated in FIG. 10. The accuracy and combination of sensor/flow metering and how it is being applied to drilling systems 100, drill-out systems 1000, and flowback system 600 is new and novel and enabling new method and capabilities. Using historical and real-time data to improve the performance of drill-out and flowback operations as described herein represents a fundamental improvement compared to conventional drill-outs and flowback.

Multiphase flow meters were previously known, but such devices had been developed and rated for high-pressure use and, more importantly, utilized nuclear (i.e., radioactive) sources for their multiphase measurements. As a result, their cost was typically between $200,000 and $400,000 per unit. The high cost and complexity (both logistical and regulatory) of managing nuclear sources used in these earlier multiphase flow meters made them uneconomical for use in the commercial drilling industry such as oil and gas or geothermal operations, so the development of automated systems based on these devices was not pursued. More recently, non-nuclear multiphase flow meters priced in the range of $40,000 to $80,000 per unit have been developed. Although still expensive, the applicants recognized that the (relatively) lower cost of such non-nuclear meters, and also their use of non-nuclear sensing sources, could support the creation of cost-effective new systems and methods for the commercial drilling industry, including automated systems 100, 600, and 1100 described herein, which had previously been overlooked.

Referring still to FIG. 10, some embodiments of the drill-out system 1000 include MPST technology comprising a mass flow sensor for fluids with high percentage gas (“MFS-HG”). The MFSHG 1042 can measure total flow rate of fluid and gas, percentage fluid, gas split, and overall density measurement. The MFSHG 1042 can function accurately across a wide range of fluid types, densities, and gas conditions. During drilling, drill-out, or flowback there may be some percentage of entrained cuttings or sand measured as fluid, increasing overall density. The MFSHG 1042 sensor can handle high percentages of entrained gas up to 50%, and operate from atmospheric to 3000 psi.

Some embodiments of the drill-out system 1000 include MPST technology comprising a fluid flow meter of high-pressure non-intrusive design (“FFS-HP/NI”). The FFS-HP/NI 1044 allows the measurement of fluid flow for high pressure drilling mud, water, gas, frac fluids, drill-out fluids, and production fluids. This non-intrusive sensor 1044 detects the flow rate of mud and other fluids being pumped into or flowing from a well at high pressure. In this context, “high pressure” is considered pressure in the range exceeding 500 psi up to 30,000 psi. In this context, “non-intrusive” relates to a sensor that does not obstruct the fluid flow along the flow path, e.g., the sensor 1044 is disposed recessed or flush with the interior pipe wall or senses through-wall.

Some embodiments of the drill-out system 1000 include MPST technology comprising a cuttings flow sensor (“CFS”). The CFS 1046 can measure the percent and volume of cuttings in degassed return drilling mud and cuttings measurement in return drilling mud. The cutting sensor 1046 can have a non-intrusive configuration and can detect the overall flow rate, percentage, and volume %/volume of cuttings in return flow drilling mud. In the embodiment of FIG. 10, a CFS 1046 can be provided on the return flowline downstream of the choke 1024. In some embodiments, a CFS 1046 can be provided on the return flowline downstream of the RCD 1022, including upstream of the choke 1024. In the embodiment of FIG. 1A, a CFS 1046 (not shown) can be provided on the line between the low side venturi meter “VML” and the shale shaker 124.

Some embodiments of the drill-out system 1000 include MPST technology comprising a sand volume and weight sensor (“SVWS”). The SVWS 1048 can be installed on flow lines adjacent to a sand separators 1028 or sand knock out unit to measure sand flowrates in flow lines and in sand dump lines leading to the sand disposal pit 1029 for holding dumped sand 1031. The SVWS 1048 can also be installed before the choke 1024 and junk catcher 1026 to provide total returning well sand and cuttings data. The SVWS 1048 can have a non-intrusive configuration. The SVWS 1048 detects the flow rates, volume and weight of sand during fracking, plug drill-out, well clean-up, well production testing, and in system process flow lines. In some embodiments, the SVWS 1048 can measure the overall process flow rate in addition to the sand volume/weight, while on other embodiments, the SWVS uses other sensors for the overall process flow this measurement. The SWVS 1048 can be used in conjunction with multiphase flow meters 1042 to provide more accurate sand flow rates process flows.

In some embodiments of the drill-out system 1000, venturi sensors or meters can be used for measurement of density and flow volume from and into the wells. The venturi meters can be used in addition to, or in lieu of, the MPST sensors described herein.

In conjunction with current on-the-market flow meters, high cost multiphase flow meters, pressure and density sensors, the previously described MPST devices support and enable many of the innovative methods and capabilities listed in this disclosure.

The following is a simple well construction example incorporating the disclosures provided herein. A gas well is drilled using ESP, PMT, and MPST and methods. The system(s) 100, 600, and/or 1000 can identify where the water producing zones, dead zones, and productive gas fractures are along the reservoir section without the use of expensive logging while drilling (“LWD”) equipment or the additional cost and time of conventional wireline logging services. The system(s) 100, 600, and/or 1000 can optimize the frac program to frac most the key gas sections and leave the most unproductive and/or water bearing sections unstimulated (unfracked). Thus, the overall size and cost of the frac job can be reduced. Further the reduced fracking results in faster drill-out and clean up operations. During drill-out of each plugged section, the use of PMT and MPST in the system 1000 provides key information about the condition and productivity potential of each section. The system 1000 also provides quantifiably improved efficiency of cleanup of the frac plug debris and in-well/free frac sand. Again, utilizing PMT, MPST, and ASMT, the final well clean up, i.e., drill-out, flowback, and testing operations provides a more productive gas well with less water production as predicted from gathered data/digital twin/AI models. In addition, the well can be placed “online”, i.e., into production, in less time, with safer automated operations, and at lower delivered cost.

As detailed in Table 1 herein, titled: “Operations and Testing During Well Drill-Out Activities”, the use of sensors including MPSTs in conjunction with advanced pressure management technologies (“PMT”), automated sand management technology (“ASMT”), and ESP methods, the new and innovative systems and methods described herein provide improved well control, safety, formation evaluation, construction, and production optimization capabilities and performance. In addition, the new and innovative systems and methods described herein provide improved safety and commercial success to well developers and owners.

TABLE 1
Operations and Testing During Well Drill-out Activities

Line	Well Construction
#	Operation/Test Type	Method	Purpose/Outcome

4	[ESP - US App	Historical and Real Time	Reduce risk from unplanned
	18/813,606] IP allowed,	data being used to plan,	events. Reduce cost and time
	but yet to issue, is to be	model, ID issues and	through performance.
	used/applied throughout	optimize performance[ESP/	Optimize activities to get
	the new innovative	PMT]. [MPST] high	optimal production from a
	capabilities listed below	quality data enabling new	well using data and AI
	for each well	planning, evaluation and	throughout all stages of well
	construction operational	automation capabilities	construction as listed in this
	APPLICATION	New and innovative	table and disclosure text.
	Historical and Real Time	method- As defined in U.S.
	data enabled Digital	patent application No.
	Twins and AI Agents and	18/813,606, also applies to
	Automation Methods	all other phases of well
	Platform also using open	construction
	and closed pressure
	management technology
	[PMT]. Adding IP
	enabling high quality
	Multiphase sensor
	technology [MPST] use
	through out.
5	Reservoir Fracture	Log pressures and flow	ID and quantify the producing
	Characterization -	data (produced Gases,	fractures, water, loss and dead
	[MPST] and [PMT]	solids and fluids) from	zones in drilled reservoir
	Pressure Management	well during drilling using	section for optimal frac, clean
	Technology Enabled	[PMT] and [MPST]. New	up, testing and production
	(MPD, UBD or open	and innovative method.	plan. Using similar methods
	hole drilling)		further characterize the
			reservoir during the drillout
			of each frac plug and section.
6	Hole cleanliness -	Accurately measure the	Identify if the cuttings are
	[MPST] enabled	volume and weight of the	being removed from the hole
		solids in the return flow	in line with drilling and ID
		from the well during	Hole cavings. Know the
		drilling/circulation.	volume of hole driller and
		Lagged to depth. New and	cave ins for a given open hole
		innovative method.	section. Avoid getting stuck
			in hole, or damaging the
			future production well bore
			and reservoir. Pre-plan the
			cement volume needed.
			Avoid the need for a hole
			caliper log.
			Allows automated
			calculation of required
			cement volumes for casing
			job.
			Allows automated
			calibration of dynamic flow
			models based on accurate
			annular volume versus depth.
			Allows accurate annular
			volume calculations for
			automated well control
			calculations.
			Allows real time feedback of
			hole sweep pills efficiency
			and ability to modify in real
			time
7	Formation Frac Test -	Increase pressure on open	ID open hole frac pressure
	[PMT and MPST]	hole until loss of fluid	limits, well fractures and loss
	enabled	during Drilling or	zones.
		circulating.
9	Formation Pore Pres	Decrease pressure on open	ID open hole reservoir pore
	Test - [PMT and MPST]	hole until gain of reservoir	pressure limits, and detail
	enabled	fluids or gases. Drilling or	productive oil, gas or water
		circulating. New and	zones
		innovative method.
20	Well Water produced	Multiphase monitoring of	Optimize frac prog to avoid
	monitoring - BBL/hr or	water in well flow using	high water zones
	%/flowrate - [PMT and	[PMT] and [MPST] New	Optimize flowback flows and
	MPSE] enabled	and innovative method.	pressures to minimize water
			% versus hydrocarbons
22	Sand produced monitoring	[MPST] Multiphase	Measure cleanliness of well
	lbs/hr or lbs/flowrate -	monitoring of sand	Measure effectiveness of frac
	[PMT] [ASDT] and new	produced returns under	in a given well section
	sand/multiphase sensor	[PMT] well flow. New and	Manage flow pressure and
	technology enabled	innovative method.	minimize sand returns from
	[MPST]		fracked reservoir
37	Plug Drill-out Formation	Monitor and manage	Determine well section
	Evaluation and future	downhole and surface	productivity, frac
	well clean up and test	pressures and measure	effectiveness and cleanliness.
	planning [PMT] and	pumped and well bore fluid	Also detect any plug failure
	[MPST] and [ASDT]	and solid returns. PMT,	or movement from data
		MPST and ASDT. New	logged frac program. New
		and innovative method.	and innovative system and
		Detailed out in Agentic AI	method.
		and Digital Twin Text
38	Automated well annulus	Monitor and manage the	Automated pressure
	pressure management	temperature and pressure	reduction to safe design
	during well flowback	of sealed well annuli and	levels, preventing well
	operations [PMT]	relieve pressures to safe	annulus from collapse,
		limits if too high during	damage, or well fluid leakage
		flowback and testing ops.	underground or to surface.
		New and innovative
		method.
42	Automated Sand	During drill-out, clean up	Remove people from the
	Dumping to prevent	and testing, monitor sand	RED Zone. Prevent damage
	production equipment	levels in SKU's and	to production equipment
	damage [PMT] and	bypassing sand and	down stream. Optimize the
	[ASDT]and [MPST]	automatedly dump sand to	time needed to drill-out, clean
		avoid sand bypass and use	up, test and be on optimal
		PME to avoid sand bypass.	production. Eliminate or
		Alarm if sand is bypassing	minimize the risk of gas flow
		SKU's under normal	from the well through the
		operating conditions. New	automated sand dumping line
		and innovative method. -	New and innovative system
		Detail in the Agentic AI	and method.
		and Digital Twin text

Referring now to FIG. 12, there is illustrated an agentic AI and digital twin system 1200 for automatic drill-out operations. The system 1200 can utilize appropriate hardware and methods, e.g., system 1000 and method 1100, to implement an integrated approach to plug drill-out operations in both conventional and unconventional wells using data-driven digital twins and agentic AI. The objective of the agentic system 1200 is to optimize process performance, pressure management, sand handling, and formation learning during the transition from fracking to production startup. The agentic system 1200 can operate the well under pressure managed conditions and continuously learns across multiple wells.

The agentic system 1200 and methods described herein can safely and efficiently drill-out frac plugs while learning formation behavior in real time, utilizing historical data when available, controlling bottom-hole pressure (BHP) relative to pore/fracture pressure, minimizing sand carryover/erosion, managing the condition of the circulating fluid, ultimately characterizing a well's reservoir into a predefined type, and enabling efficient pre-planned transition to flowback and production after the completion of drill-out operations.

Using AI and digital twin frameworks, the system 1200 can dynamically test and characterize a well's reservoir 1010 as the frac plugs 1025 are drilled out and deliver a well-specific flowback plan with proven limits and targets. Embodiments of the system 1200 can use pressure management, optimized data gathering and machine learning and fluid circulation control while drilling out plugs 1025. Embodiments of the system 1200 can include a multi-layer digital twin wrapped and controlled by a supervisory AI twin.

Described below is a representative example of the agentic system 1200 using a multi-layer digital-twin (DT) stack for fracked well plug drill-out operations. The system 1200 can be enabled and supported through the use of additional high quality sensor/data, active pressure management (PM), sand separation, and fluids circulation management. For purposes of this example, the “Pressure Management” (PM) operating modes of the example embodiment can be defined as follows:

• • a) For Underbalanced/UBD mode, the bottom hole pressure (BHP)<Pore pressure.
  • b) For Balanced MPD mode, the BHP≈Pore pressure.
  • c) For Overbalanced MPD mode, the BHP>Pore pressure.
    UBD mode maximizes inflow (i.e., of returns) and measurement sensitivity/diagnostics, but has a higher risk of well control and sand incidents than other modes. Balanced MPD mode provides little or no inflow compared to UBD, maintains good measurement sensitivity, with lower to zero risk of well control and sand incidents than UBD. Overbalanced MPD mode provides less inflow than Balanced MPD, but with the lowest risk of well control incidents.

The system 1200 learns continuously from downhole and surface sensors as plugs 1025 are drilled out and each stage is cleaned up. In operations one or more DT may be in use at a given time, but it not necessary for each or all of them to be used in any particular wells' operations.

The digital-twin ecosystem of the system 1200 mirrors the physical system hardware 1000 and the reservoir 1010 in real time, integrating physics-based and AI-driven models across the surface, wellbore, and reservoir domains. These are orchestrated by an AI agent that continuously learns from data, adjusts control parameters, generates predictive insights and allows for safe and clear oversight by system operators.

In some embodiments, during system operation each digital twin (DT) in the DT stack 1201 executes an ongoing loop of simulations. In every cycle, the DT ingests a fresh set of input values, runs the simulation, and generates a corresponding set of output values. The paired input-output data from each cycle is then supplied to an AI module, such as AIAM 1222. Using this large corpus of input-output pairs produced across many DT iterations, the AI module applies machine learning to learn and model the relationships that map inputs to outputs for the underlying system. Following this pattern, each respective DT in the DT stack 1201 can simultaneously run a different ongoing loop of simulations regarding different aspects of the system, and all of the data from the different respective DT simulations is provided to the AI module for learning and improving the model of the underlying system. In some embodiments, the fresh input values provided to a DT for subsequent iterations can be provided from the AI module based on real-time data or output data from a different DT in the DT stack 1201. In some embodiments, the AI module can use the data from the DT iterations along with machine learning to make estimates and projections regarding system behavior and/or to modify control settings or values in accordance with system objectives.

For example, a representative example of the system 1200 using a multi-layer digital-twin (DT) stack 1201 for frack plug drill-out operations may comprise six DTs. A first DT in this example DT stack 1201 can be a wellbore hydraulics and transient twin 1202, which simulates multiphase hydraulics, transient flow, cuttings transport, and pressure wave propagation. The physics models used in the first DT 1202 can include transient pressure, temperature, viscosity and flow rate; cuttings slip/settling, bed height; foam/gas fraction; surge/swab. The data from the iterations of the first DT 1202 can be used, as detailed herein in FIG. 2A and the associated description, to support real-time analysis and prediction 208, integration of the digital twin outputs 210 and supporting decision making and adjustment outputs 212. Further the outputs can contribute to training the AI module for, e.g., improved BHP estimate/forecast, standpipe pressure (SPP), frictional losses, cuttings load and risk of pack-off, hydrate/foam flags and optimal pump/choke ramps.

A second DT in this example DT stack 1201 can be a fluids and chemistry twin 1204, which simulates mud/oil/water/gas composition; viscosity vs T/P; friction reducer; breaker; emulsion/foam stability; H₂S/CO₂ handling. The physics models used in the second DT 1204 can include PVT+rheology; mixing/aging; carryover effects on separation; gas-cut mud density. The data from the iterations of the second DT 1204 can be used, as detailed herein in FIG. 2A and the associated description, to support real-time analysis and prediction 208, integration of the digital twin outputs 210 and supporting decision making and adjustment outputs 212. Further the outputs contribute to training the AI module for, e.g., improved recommended rheology targets for transport, chemical dosing, sweeps, fluid routing (recirculate vs treat vs dispose), contamination alerts.

A third DT in this example DT stack 1201 can be a pressure management (pm)/rig and surface process twin 1206, which models choke, pumps, sand traps, and recirculation loops for optimal pressure and solids control. The physics models used in the third DT 1206 can include compressible multiphase system network; choke AP/Cv models; separator flows/efficiency, system erosion. The data from the iterations of the third DT 1206 can be used by the AI module for improved control decisions, e.g., closed-loop BHP regulation (via pump rate and+choke position+fluid density), tank/level control, automated sand-dump logic, flare minimization. The data from the iterations of the third DT 1206 can be used, as detailed herein in FIG. 2A and the associated description, to support real-time analysis and prediction 208, integration of the digital twin outputs 210 and supporting decision making and adjustment outputs 212. Further the outputs contribute to training the AI module for, e.g., improved ECD/BHP targets vs time, choke setpoints, separator/sand-trap loading forecasts, erosion alarms, fluid routing (to recirc vs to sales/flare).

A fourth DT in this example DT stack 1201 can be a bit/bha mechanics twin 1208, i.e., “AutoDriller”, which models torque, drag, WOB, dP, ROP, vibration, and plug milling dynamics. The physics models used in the fourth DT 1208 can include Torque and Drag string model; bit-rock response; vibration. The data from the iterations of the fourth TD 1208 can be used, as detailed herein in FIG. 2A and the associated description, to support real-time analysis and prediction 208, integration of the digital twin outputs 210 and supporting decision making and adjustment outputs 212. Further the outputs contribute to training the AI module for, e.g., improved WOB/RPM/flow for plug materials vs formation; advisory to avoid stick-slip; predicted ROP; plug transition detection.

A fifth DT in this example DT stack 1201 can be a reservoir-near-wellbore/fracture interaction twin 1210, which estimates fracture connectivity, pore pressure, and formation response to varying BHP. The physics models used in the fifth DT 1210 can include reduced-order reservoir/frac connectivity. (light for automation). The data from the iterations of the fifth DT 1210 can be used, as detailed herein in FIG. 2A and the associated description, to support real-time analysis and prediction 208, integration of the digital twin outputs 210 and supporting decision making and adjustment outputs 212. Further the outputs contribute to training the AI model for learning stage contribution while drilling out, improved connectivity index, fines/proppant mobilization risk vs drawdown, recommended BHP envelopes per stage.

A sixth DT in this example DT stack 1201 can be a sand production and separation twin 1212, which tracks sand generation, transport, and separation performance in real time. The physics modeled by the sixth DT 1212 include sand separator volume thresholds; Sand vol tracking through return flow line, PM chokes, separators, SKU dump lines; cyclone/trap efficiency curves; erosion wear maps. The data from the iterations of the sixth DT 1212 can be used, as detailed herein in FIG. 2A and the associated description, to support real-time analysis and prediction 208, integration of the digital twin outputs 210 and supporting decision making and adjustment outputs 212. Further the outputs contribute to training the AI module for, e.g., improved sand-rate now/next, sand-rate per stage, dump timing, allowable velocity limits, predicted wear hotspots, recommended choke step size.

In some embodiments, the system 1200 can comprise an AIAM-Supervisory Control 1216 for real-time operational control optimization, outputs and learning, which integrates all layers of the plug drill-out system. The AIAM-Supervisory Control 1216 is a subset of an AI Agent Module (“AIAM”) 1222, which integrates the digital twin outputs and supports decision making and adjustment outputs learning patterns and optimizing plug drill-out operations across wells using similar methods as described herein in connection with FIG. 2A and FIG. 2B for autonomous drilling. The models and methods used by the AIAM-Superviory Control 1216 can include hybrid physics and machine learning, again as described herein in greater detail in connection with FIGS. 2A, 2B, 3, 4 and 5 for autonomous drilling. The digital twin outputs may use data fusion as detailed herein in connection with FIG. 4, block 406, and the outputs are used by the trained AIAM 1222 (and also used for machine learning by the AI System as described in connection with FIG. 2B, block 236) to make decisions regarding adjusting the system's controls, sending notifications, or making recommendations to the user, again using similar methods as described in FIGS. 2A, 2B, 3, 4 and 5 for autonomous drilling. The necessary control adjustment commands are then sent by the AIAM-Supervisory Control 1216 to the integrated control system 1218 for implementation via the choke 1024, pump 1016, drawworks 1009, separators 1028, 1030, 1032, automated valves 1065, etc. The commanded changes to the control system 1218 are also fed back from the AIAM-Supervisory Control 1216 to the DT stack 1201 to update the models, as denoted by the circular arrows 1224 in FIG. 12.

To define its desired performance, the agentic drill-out system 1200 can be configured with a well program and a set of one or more operational control objectives (and some cases boundary system operational limits), collectively denoted 1214. The set of operational limits 1214 can include, but is not limited to the following: First, to track BHP setpoint trajectory (per stage) to achieve learning without risking unplanned influx/losses; second, to maintain transport margin (avoid cuttings/sand beds, annular velocity management); third, to minimize sand to surface and erosion while preserving diagnostic inflow; fourth, to manage returns separation and recirculation fluid quality; and fifth, to gather quality data for real time machine learning/AI control and future ML/AI learning. To achieve these operational control objectives 1214, the agentic drill-out system 1200 can be configured to perform one or more system control actions 1218 including, but not limited to, controlling pump speed/flow, controlling drill string feed off/WOB, controlling choke opening/PD back-pressure, controlling RCD/BOPs, controlling surface system flow automation, sand separation management, etc.

The AI aspects of the agentic drill-out system 1200 can be configured to provide real-time outputs and learning in conjunction with the AI Agent Module 1222. As illustrated in FIG. 12, in some embodiments the AIAM 1222 is operationally disposed between the “Historical Data” layer and the “Supervising Current Well” layer of the agentic system 1200, whereby inputs and outputs of the AI Agent Module can be retrieved from, and output to, each module during operation. In some embodiments, the AIAM 1222 outputs can include pre-job planning, e.g., pressure ramps, choke schedules, stage order, expected sand and fluids profile. In some embodiments, the AI outputs can include real-time adjustments, e.g., control actions within safety envelopes. In some embodiments, the AI outputs can include formation pressure indicators, e.g., controlled underbalance periods and stage classification. In some embodiments, the AI outputs can include optimized parameters for drilling each plug (e.g., WOB, RPM, flow). In some embodiments, the AI outputs can include mud/fluid recipe adjustments (e.g., viscosity/FR/foam) to maintain transport without over-shear. In some embodiments, the AI outputs can include fines/proppant mobilization thresholds versus drawdown (safe BHP window for later flowback). In some embodiments, the AI outputs can include plug failure or movement detection. In some embodiments, the AI learning can include post-job learning, e.g., updated priors per basin/landing zone/completion design. In some embodiments, the AI learning can include fracture conductivity/connectivity index per stage inferred from rate-pressure response while opening plugs. In some embodiments, the AI outputs and learning can include sand-risk map versus choke Ap and annular velocity (avoid system erosion).

The agentic drill-out system 1200 utilizes hardware, supporting sensors, and a data backbone to provide real-time data to the system various electromechanical control unit, digital twin framework, and AI/machine learning aspects. While some of these real-time sensor elements and their associated values have been previously described in connection with the hardware drill-out system 1000, they are now further described (non-exhaustively).

Surface equipment sensor locations and measured parameters/values for the drill-out system 1000 include: standpipe pressure 1050, casing pressure 1051, choke inlet pressure 1052 and outlet 1053 pressure; plug/junk catcher dP 1054; flowrate in/out 1042 and/or 1044 (volume, mass and multiphase); choke position (0-100% open) 1024; mud return temperature 1055; separator pressures 1056/levels 1057; sand flow rates 1048 (acoustic/UT/Radar/cyclone dP); gas flow 1058/flare metering 1058; tank levels 1059; and chemical dosing 1060.

Downhole and well string sensor locations and their associated measured parameters/values for the drill-out system 1000 can include BHA sensors located on or within the bottom hole assembly 1012, namely: a) Measurement While Drilling (“MWD”) sensors, e.g., real-time downhole telemetry (usually mud-pulse or EM) that sends directional and drilling parameters to surface; typical MWD measurements/data include: inclination, azimuth, gamma ray, shock/vibration, toolface, etc.; b) Logging While Drilling (“LWD”) sensors, e.g., formation-evaluation sensors placed in the drill string and run while drilling; typical LWD measurements/data include: resistivity, density, neutron, sonic, imaging, etc. (often combined with MWD in the same BHA); c) Pressure While Drilling (“PWD”) sensors, e.g., downhole annular and internal pressure sensors that record and transmit real-time data; typical PWD measurements/data include ECD, standpipe pressure, surge/swab, and transient pressure behavior, which can be used for wellbore stability and kick/loss monitoring. Downhole and well string sensor locations and measured parameters/values for the drill-out system 1000 can also include near-bit sensors located on or within the downhole motor unit 1014, namely: a) RPM (rotary speed) measurements; b) weight on bit (“WOB”) measurements; and c) Vibration measurements. Downhole and string sensor locations and measured parameters/values for the drill-out system 1000 can further include distributed fiber sensors 1015 having a fiber optic cable extending along all or part of the wellbore, namely: a) Distributed Temperature Sensing (DTS) sensors; and; b) Distributed Acoustic Sensing (DAS) sensors. DTS sensors can use the fiber optic cable 1015 to measure temperature along all or part of the wellbore in real time. DTS sensors can be used to identify fluid entry points, circulation losses, flow behind casing, and well integrity issues. DAS sensors can use the fiber optic cable 1015 to record acoustic energy along the wellbore. DAS sensors can be used to detect flow, drill bit vibration, perforation events, frac stages, and general downhole activity. In preferred embodiments, the DAS sensor uses the same fiber optic cable as the DTS sensor, thus providing combined DTS/DAS sensors.

Fluid sensor locations and measured parameters/values for the drill-out system 1000 include: Inlet flowmeter 1061; Inline densitometer/viscometer 1062; gas-oil-water cut 1063 (MPFM or VFM). In addition, fluid/mud engineer report data (from human or AI engineers) can be digitally provided to the system.

Further, operational metadata can be provided to the DT and AI aspects of the drill-out system 1000 to supplement the real-time data. This ops metadata can include: Frac data, e.g., plug type/material, stage spacing, proppant, fluid recipes, etc; drilling data, e.g., MWD/LWD, fracture zone, pressure profile, etc.

Operation of the agentic drill-out system 1200 across multiple wells and pads can facilitate continuous AI agent development. In particular, by use of the digital twin, ML, and AI architecture as presented on multiple wells, the cumulative collection of real-time and historical well construction and production data increases the training materials available to AI/ML aspects. This training enhances the system's effectiveness to utilize the emerging technology of higher performance quantitative sensors in a cost effective manner. Further, the training enhances the system's effectiveness in providing automated sand management technology. Further, the training enhances the system's effectiveness in providing advanced pressure management during all phases of well construction. Thus, the agentic drill-out system 1200 provides several new innovative methods and system capabilities made possible in the planning and execution of plug drill-out operations, without the need for time consuming and expensive post drill-out logging.

As described herein, the automated drill-out system 1200 can include a continuously-learning AI agent that generates new processes providing real-time control and value add information. These new processes can include, but are not limited to, the following: pre-job planning processes; operational system safety barrier processes; teal-time supervisory control and automated reservoir testing processes; wellbore and surface hydraulics behavior processes; post-drill-out handoff processes; digital twin/AI agent optimization processes.

In some embodiments, the pre-job planning processes created by the agentic system 1200 are implemented before the first frac plug is drilled. The pre-job planning processes can include a process to generate a stage-by-stage pressure-management plan specifying underbalanced/balanced/overbalanced (“UB/B/OB”) targets, ramp rates, hold times for diagnostics, and safe envelopes. The pre-job planning processes can include a process to predict expected sand and fluid profiles per stage; size separator(s), size flare capacity, and circulation routing. The pre-job planning processes can include a process to recommend bottom hole assembly (“BHA”) configuration, recommend drilling fluid recipes, recommend weight on bit (“WOB”), RPM, and flow parameters to meet predicted transport and measurement objectives.

In some embodiments, the operational system safety barriers processes created by the agentic system 1200 are implemented in a non-negotiables manner. The operational safety processes can include a process to first learn well and/or system specifics and then maintain operations inside pressure limits, burst/collapse limits, and erosion limits. The operational safety processes can include a process to apply uncertainty risk policies, e.g., when operational data confidence is relatively low, the process defaults to a conservative balance/overbalance mode of operation, whereas when operational data confidence is relatively high, the process can allow an aggressive underbalanced mode of operation. The operational safety processes can include a process to generate a customized user interface 1220 (“UI”) wherein advisor modes and autonomy modes are cleanly separated. One such UI 1220 can maintain human-in-the-loop requirements for pressure management (“PM”) mode changes, relatively higher risk operations, and abnormal conditions, whereas during relatively lower risk operations human-in-the-loop supervision can be optional.

In some embodiments, the real-time (“RT”) supervisory control and automated reservoir testing processes created by the agentic drill-out system 1200 can include a process to maintain bottom hole pressure (“BHP”) to setpoint using digital twins (“DTs”) and coordinated pump and choke moves. The RT control and automated testing processes can include a process to generate circulation fluid management recommendations to hold rheology within specifications. The RT control and automated testing processes can include a process to learn cause-effect relations between control actions (e.g., choke steps, pump ramps, chemistry tweaks) and outcomes (e.g., rate of penetration (“ROP”), sand rate, stability) during drill-out. The RT control and automated testing processes can include a process to perform mini-tests (i.e., short, steady underbalanced holds) as each plug opens to estimate or determine stage contribution and fracture quality. The RT control and automated testing processes can include a process to learn-forward/plan a series of micro-tests during drill-out comprising first switching to underbalanced mode to estimate or determine pore pressure and connectivity, then returning to balanced/overbalanced mode for safety/transport. The RT control and automated testing processes can include a process to generate stage-level connectivity and effective fracture conductivity via pressure-rate deconvolution during plug break-throughs. The RT control and automated testing processes can include a process to determine pore pressure and drawdown limits by analyzing controlled underbalance windows and step-tests and flag stages that warrant special treatment in flowback. The RT control and automated testing processes can include a process to generate confidence-bounded estimates for values of well characteristics including pore pressure, skin, connectivity indices, fines/proppant mobilization. The RT control and automated testing processes can include a process to generate gas/oil/water cleanup signatures and their link to the well completion design executed (i.e., stage spacing, perforations, proppant type, volumes pumped).

In some embodiments, the wellbore and surface hydraulics behavior processes created by the drill-out system 1200 can include a process to model BHP from choke position and pump-rate transfer function including lag and nonlinearity under multiphase conditions. In some embodiments, the wellbore and surface hydraulics behavior processes can include a process to model the cuttings transport operating window (critical annular velocity versus rheology and inclination). In some embodiments, the wellbore and surface hydraulics behavior processes can include a process to model sand production onset, erosion risk envelope, and to estimate surface fluid and sand flow rates. In some embodiments, the wellbore and surface hydraulics behavior processes can include a process to model separator performance control including, e.g., implementing proactive flow path and pressure adjustments ahead of known surge patterns. In some embodiments, the wellbore and surface hydraulics behavior processes can include a process to model sand separation management with predictive dump scheduling. In some embodiments, the wellbore and surface hydraulics behavior processes can include a process to generate separator performance curves through gas/oil ratios

In some embodiments, the post-drill-out handoff processes created by the drill-out system 1200 can encompass handoff to cleanup, well-testing, or commissioning operations. In some embodiments, the post-drill-out handoff processes include a process to generate a well-specific flowback plan, e.g., a plan including initial choke sizing and settings, drawdown schedule, sand limits, separator sizing and setpoints, and expected cleanup timeline. In some embodiments, the post-drill-out handoff processes include a process to generate “no-go” boundaries (i.e., values) for subsequent operations, including values for erosion, separator carryover, pressure integrity, etc. and early-warning predictors with predefined time horizons (e.g., 15-30 minute horizon). In some embodiments, the post-drill-out handoff processes include a process to identify benchmarks against other pad wells, i.e., highlighting deviations needing additional engineering review. In some embodiments, the post-drill-out handoff processes include a process to categorize wells according to behavioral archetypes (e.g., fast-cleanup, gas-early, water-heavy, sand-prone) defined by clustering wells according to statistically significant similarities in their measured performance values.

In some embodiments, the digital twin/AI agent optimization processes created by the agentic system 1200 (again using similar methods as described herein in connection with FIGS. 2A, 2B, 3, 4 and 5 for autonomous drilling) can generate deliverables such as reduced non-productive time (“NPT”), e.g., quantified reductions in NPT characterized by reduced actual well control incidents, pack-off incidents, or sand incidents versus predicted or expected incidents. In some embodiments, the digital twin/AI agent optimization processes can generate deliverable such as improved equipment life, e.g., quantified control of actual erosion and quantified reduction of actual separator overloads versus predicted or expected values.

In some embodiments, the digital twin/AI agent optimization processes can generate deliverables such as optimized stage drill-out, e.g., quantified improvement in actual quality and accuracy of subsurface pore-pressure targets versus predicted or expected values. In some embodiments, the digital twin/AI agent optimization processes can generate deliverables such as optimized stage cleanup, e.g., providing an informed choke strategy to quantifiably accelerate transition to stable flowback versus predicted or expected transition values. In some embodiments, the digital twin/AI agent optimization processes can generate deliverables such as enhanced reservoir understanding, e.g., early identification of stage connectivity and pressure profiles. In some embodiments, the digital twin/AI agent optimization processes can generate deliverables such as reduced time to stable well commissioning, e.g., delivering a tighter, pre-validated flowback plan. In some embodiments, the digital twin/AI agent optimization processes can generate deliverables such as improved early and future production, e.g., using drawdown discipline and stage-balanced cleanup versus predicted or expected values. In some embodiments, the digital twin/AI agent optimization processes can generate deliverables such as scalable intelligence, e.g., present learning is transferred cross-well to improve new-well performance versus predicted or expected values.

It will be appreciated by those skilled in the art having the benefit of this disclosure that this automated well drilling, flowback, drill-out, and production system provides an improved solution to the challenge of well flowback and production after fracking. It should be understood that the drawings and detailed description herein are to be regarded in an illustrative rather than a restrictive manner and are not intended to be limiting to the particular forms and examples disclosed. On the contrary, included are any further modifications, changes, rearrangements, substitutions, alternatives, design choices, and embodiments apparent to those of ordinary skill in the art, without departing from the spirit and scope hereof, as defined by the following claims. Thus, it is intended that the following claims be interpreted to embrace all such further modifications, changes, rearrangements, substitutions, alternatives, design choices, and embodiments.