BURIED OBJECT ESTIMATION SYSTEM AND EXCAVATOR

Description

CROSS-REFERENCE TO RELATED APPLICATION

The present application claims priority under 35 U.S.C. §119 to Japanese Patent Application No. 2024-225498 filed December 20, 2024, the contents of which are incorporated herein by reference.

BACKGROUND OF THE INVENTION

FIELD OF THE INVENTION

The present disclosure relates to a buried object estimation system and an excavator.

DESCRIPTION OF THE RELATED ART

In excavation work, an excavator may damage a buried object buried in an excavation target, such as a ground and the like. Therefore, work for confirming the presence of the buried object in the excavation target is performed at a work site. For example, a disclosed excavation system acquires measurement data of a ground using an underground surveying device, and estimates the position of a buried object based on the measurement data, using a trained model trained on the position of an underground buried object corresponding to measurement data for training.

When estimating the buried object using the underground surveying device in the excavation work performed by the excavator as described above, the excavation work by the excavator is frequently suspended, and the work efficiency is deteriorated. Therefore, it is required that the buried object which may possibly be buried in the excavation target can be estimated concurrently with the excavation work performed by the excavator at the work site.

SUMMARY OF THE INVENTION

The present disclosure provides a technology that can suitably estimate the probability of the presence of a buried object in an excavation target during excavation work performed by an excavator.

One embodiment of the present disclosure provides a buried object estimation system for an excavator, configured to estimate a probability of presence of a buried object exists in an excavation target. The excavator includes a lower traveling body, an upper turning body turnably provided on the lower traveling body, an attachment provided on the upper turning body and configured to excavate the excavation target, and a controller including a processor and a memory, and configured to control a movement of the attachment. The buried object estimation system includes an information processing computer including a processor and a memory, and configured to train a learning model for detecting the buried object by acquiring information related to an excavation reaction force in excavation work performed by the attachment, and transmit the learning model to the controller. The controller estimates presence or absence of the buried object in the excavation target based on an excavation reaction force acquired in an actual excavation work and the learning model.

According to one embodiment, the probability of the presence of a buried object in an excavation target can be suitably estimated during excavation work performed by an excavator.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a side view showing an excavator according to an embodiment;

FIG. 2 is a side view showing various physical quantities related with excavation attachments;

FIG. 3 is an explanatory diagram showing a basic system of an excavator;

FIG. 4 is a diagram showing an example configuration of an excavation control system mounted on the excavator of FIG. 1;

FIG. 5 is a diagram showing a cross section of ground in which a water pipe is buried;

FIG. 6 is a graph showing the relationship between excavation reaction force and approach distance;

FIG. 7 is a block diagram showing a detection system;

FIG. 8 is a block diagram showing functional parts of an information processing device for generating a learning model;

FIG. 9 is an explanatory diagram showing an excavator’s acquisition of a learning model, tuning of the learning model, and use of the learning model during excavation work;

FIG. 10 is a diagram illustrating image information displayed on a display device when a buried object estimation function is executed;

FIG. 11A is a flowchart showing a processing flow before actually performing excavation work;

FIG. 11B is a flowchart showing a buried object estimation method when actually performing excavation work; and

FIG. 12 is a schematic view showing a configuration example of an operation system.

DETAILED DESCRIPTION OF THE DISCLOSURE

Hereinafter, an embodiment for carrying out the present disclosure will be described with reference to the drawings. In the drawings, the same components are denoted by the same reference numerals, and redundant descriptions may be omitted.

FIG. 1 is a side view showing an excavator 100 according to an embodiment. The excavator 100 includes a lower traveling body 1 and an upper turning body 3 mounted on the lower traveling body 1 turnably via a turning mechanism 2.

The excavator 100 includes an attachment AT, which is an example of an attachment as a working element. The attachment AT includes a boom 4 attached to the upper turning body 3, an arm 5 attached to the tip of the boom 4, and a bucket 6 attached to the tip of the arm 5. In this specification, for convenience, a side of the upper turning body 3 on which the boom 4 is attached is referred to as the front side, and a side on which a counterweight is attached is referred to as the rear side. The boom 4 is driven by a boom cylinder 7. The arm 5 is driven by an arm cylinder 8. The bucket 6 is driven by a bucket cylinder 9.

The upper turning body 3 is mounted with a cabin 10 and a power source, such as an engine 11, an electric motor, or the like. In the cabin 10, an operation device 26, a controller 30, a display device 40, a sound output device 45, and the like are provided.

The excavator 100 includes attitude detection devices M1 for detecting the attitude of the attachment AT. The attitude detection devices M1 also serve as detection devices for detecting information about an excavation reaction force. Specifically, the attitude detection devices M1 include a boom angle sensor M1a, an arm angle sensor M1b, and a bucket angle sensor M1c. For example, a rotation angle sensor for detecting the rotation angle of a boom foot pin, a stroke sensor for detecting the stroke amount of the boom cylinder 7, and an inclination (acceleration) sensor for detecting the inclination angle of the boom 4 can be applied as the boom angle sensor M1a. Similar sensors can also be applied as the arm angle sensor M1b and the bucket angle sensor M1c.

The excavator 100 includes an object detection device 70 and the like on the upper turning body 3. The object detection device 70 detects an object existing around the excavator 100. The object is, for example, a person, an animal, a vehicle, a construction machine, a building, a hole, and the like. The object detection device 70 is composed of, for example, one of an ultrasonic sensor, a millimeter-wave radar, an imaging device, an infrared sensor, or the like, or a combination of a plurality of them. The imaging device is, for example, a monocular camera, a stereo camera, a LIDAR, a range image sensor, or the like. In the illustrated example, the object detection device 70 includes a rear camera 70B attached to the rear end of the upper surface of the upper turning body 3, a front camera 70F attached to the front end of the upper surface of the cabin 10, a left camera 70L attached to the left end of the upper surface of the upper turning body 3, and a right camera 70R attached to the right end of the upper surface of the upper turning body 3.

The object detection device 70 may be configured to detect an object (for example, a person) within a region set around the excavator 100. For example, the object detection device 70 may be configured to detect objects while distinguishing between a person and a non-human object.

FIG. 2 is a side view showing various physical quantities related with the attachment AT. The boom angle sensor M1a detects a boom angle θ1. The boom angle θ1 is an angle, with respect to a horizontal line, of a line segment P1-P2 connecting a boom foot pin position P1 and an arm connection pin position P2 in the XZ plane. The arm angle sensor M1b detects an arm angle θ2. The arm angle θ2 is an angle, with respect to a horizontal line, of a line segment P2-P3 connecting the arm connection pin position P2 and a bucket connection pin position P3 in the XZ plane. The bucket angle sensor M1c detects a bucket angle θ3. The bucket angle θ3 is an angle, with respect to a horizontal line, of a line segment P3-P4 connecting the bucket connection pin position P3 and a bucket claw tip position P4 in the XZ plane. The bucket angle θ3 may be calculated based on the content an operation on the operation device 26. For example, the bucket angle θ3 may be calculated based on outputs from pilot pressure sensors 15a and 15b (FIG. 3). In this case, the bucket angle sensor M1c may be omitted.

FIG. 3 is an explanatory diagram showing a basic system of the excavator 100. The basic system of the excavator 100 includes the engine 11, a main pump 14, a pilot pump 15, a control valve unit 17, the operation device 26, the controller 30, a display device 40, a sound output device 45, an engine control device 74, an operation mode switch 75, a buried object estimation mode switch 76, the attitude detection devices M1, excavation pressure sensors S1, and the like.

The engine 11 is a drive source for the excavator 100, and is, for example, a diesel engine operating to maintain a predetermined rotational speed. The output shaft of the engine 11 is connected to the input shafts of the main pump 14 and the pilot pump 15.

The main pump 14 is a hydraulic pump for supplying a hydraulic oil to the control valve unit 17 through a hydraulic oil line 16, and is, for example, a swash plate-type variable displacement hydraulic pump. In the swash plate-type variable displacement hydraulic pump, the piston stroke length defining a displacement volume changes in accordance with a change in the tilt angle of the swash plate, to change the discharge flow rate per one rotation. The tilt angle of the swash plate is controlled by a regulator 14a. The regulator 14a changes the tilt angle of the swash plate in accordance with a change in a control current from the controller 30. For example, the regulator 14a increases the tilt angle of the swash plate in accordance with an increase in the control current, to thereby increase the discharge flow rate of the main pump 14. Conversely, the regulator 14a decreases the tilt angle of the swash plate in accordance with a decrease in the control current, to thereby reduce the discharge flow rate of the main pump 14. A discharge pressure sensor 14b detects the discharge pressure of the main pump 14. An oil temperature sensor 14c detects the temperature of the hydraulic oil sucked by the main pump 14.

The pilot pump 15 is a hydraulic pump for supplying the hydraulic oil to various hydraulic control devices, such as the operation device 26 and the like, through a pilot line 25. For example, a fixed displacement hydraulic pump can be applied as the pilot pump 15.

The control valve unit 17 controls flows of the hydraulic oil related to hydraulic actuators. In the illustrated example, the control valve unit 17 includes a plurality of flow rate control valves. The control valve unit 17 selectively supplies the hydraulic oil received from the main pump 14 through the hydraulic oil line 16 to one or a plurality of hydraulic actuators in accordance with a change in a pressure (pilot pressure) corresponding to the operation direction in which the operation device 26 is operated and the operation amount by which the operation device 26 is operated. The hydraulic actuators include, for example, the boom cylinder 7, the arm cylinder 8, the bucket cylinder 9, a left traveling hydraulic motor 1A, a right traveling hydraulic motor 1B, a turning hydraulic motor 2A, and the like. In the illustrated example, the hydraulic motors (the left traveling hydraulic motor 1A, the right traveling hydraulic motor 1B, and the turning hydraulic motor 2A) are swash plate-type piston motors. However, the hydraulic motors may be replaced with electric motors.

The operation device 26 is a device used by an operator to operate the hydraulic actuators, and includes a lever 26A, a lever 26B, a pedal 26C, and the like. The operation device 26 receives supply of the hydraulic oil from the pilot pump 15 through the pilot line 25 to generate a pilot pressure. The operation device 26 applies the pilot pressure to a pilot port of a corresponding flow rate control valve through a pilot line 25a. The pilot pressure changes in accordance with both an operation direction in which the operation device 26 is operated and an operation amount by which the operation device 26 is operated. The operation device 26 may be remotely operated. In the remote operation, the operation device 26 generates a pilot pressure based on information about the operation direction and the operation amount received via wireless communication.

The operation device 26 may be an electric operation device instead of a hydraulic operation device as described above. In this case, a solenoid valve for regulating the pilot pressure may be disposed between the flow rate control valves in the control valve unit 17 and the pilot pump 15. Information about the operation direction in which the electric operation device is operated and the operation amount by which the electric operation device is operated is transmitted from the electric operation device to the controller 30 as an electric signal. The controller 30 can regulate the magnitude of the pilot pressure applied to the flow rate control valves by adjusting the opening area of the solenoid valve in accordance with the electric signal received from the electric operation device.

The controller 30 functions as a control part for driving and controlling the excavator 100. The functions of the controller 30 may be realized by any hardware or a combination of hardware and software. For example, the controller 30 is mainly composed of a microcomputer including a processor, such as a Central Processing Unit (CPU) and the like, a memory, such as a Random Access Memory (RAM), a Read Only Memory (ROM), and the like, an interface device for various inputs and outputs, and the like. The controller 30 realizes various functions by executing various programs stored in the ROM and the like on the CPU. For example, the controller 30 changes the magnitude of the control current to the regulator 14a in accordance with the pressure of the hydraulic oil in a negative control valve, and controls the discharge flow rate of the main pump 14 via the regulator 14a.

The display device 40 is a device for displaying various information, and is disposed near the driver's seat in the cabin 10. In the illustrated example, the display device 40 includes an image display part 41 and an input part 42. The image display part 41 is a liquid crystal display. The input part 42 is a membrane switch. The operator can input information and instructions into the controller 30 by using the input part 42. The operator can know the operation status and control information of the excavator 100 by viewing the image display part 41. The display device 40 is connected to the controller 30 via a communication network, such as CAN and the like. However, the display device 40 may be connected to the controller 30 via a dedicated line.

The display device 40 operates by receiving power from a storage battery 90. The storage battery 90 is charged with power generated by an alternator 11a. The electric power of the storage battery 90 is also supplied to devices other than the controller 30 and the display device 40, such as an electric component 72 and the like of the excavator 100. A starter 11b of the engine 11 can be driven by the electric power from the storage battery 90 to start the engine 11.

The sound output device 45 is a device for outputting sound information. In the illustrated example, the sound output device 45 is a loudspeaker disposed near the driver's seat in the cabin 10. The sound output device 45 may be a buzzer.

The engine control device 74 is a device for controlling the engine 11. For example, the engine control device 74 controls a fuel injection amount and the like so as to realize an engine rotational speed that is set via the input device.

The engine control device 74 transmits various data indicating the state of the engine 11 (for example, data relating to physical quantities, such as data indicating a cooling water temperature detected by a water temperature sensor 11c and the like) to the controller 30. The controller 30 stores various data in a memory 30a and transmits them to the display device 40 or the like as required. The same applies to data indicating the tilt angle of the swash plate output from the regulator 14a, data indicating the discharge pressure of the main pump 14 output from the discharge pressure sensor 14b, data indicating the hydraulic oil temperature output from the oil temperature sensor 14c, data indicating the pilot pressure output from the pilot pressure sensors 15a and 15b, and the like.

The operation mode switch 75 is a switch for switching the operation modes of the excavator 100, and is provided in the cabin 10. The operation modes are, for example, an M (manual) mode and an SA (semi-automatic) mode. The controller 30 switches the operation mode of the excavator 100 in accordance with an output from the operation mode switch 75.

The M (manual) mode is a mode in which the excavator 100 is operated in accordance with input of an operation into the operation device 26 by the operator. For example, it is a mode in which the boom cylinder 7, the arm cylinder 8, the bucket cylinder 9, and the like are operated in accordance with the content of the operation that is input into the operation device 26 by the operator. The SA (semi-automatic) mode is a mode in which, when a predetermined condition is satisfied, the excavator 100 automatically operates regardless of the content of an operation that is input into the operation device 26. For example, when the predetermined condition is satisfied, at least one of the boom cylinder 7, the arm cylinder 8, or the bucket cylinder 9 automatically operates regardless of the content of the operation that is input into the operation device 26. The operation modes may include a fully automatic mode in which all of the lower traveling body 1, the turning mechanism 2, the boom cylinder 7, the arm cylinder 8, and the bucket cylinder 9, and the like autonomously operate.

The buried object estimation mode switch 76 is a switch for starting a buried object estimation function mode, and is provided in the cabin 10. In the buried object estimation function mode, a process for estimating a buried object in an excavation-target ground is performed. In the buried object estimation function mode of the excavator 100, the presence or absence of a buried object is estimated based on an excavation reaction force. The operator switches between starting and stopping the buried object estimation function mode by operating the buried object estimation mode switch 76.

The controller 30 executes the buried object estimation function mode in accordance with a start instruction from the buried object estimation mode switch 76 and stops the buried object estimation function mode in accordance with a stop instruction from the buried object estimation mode switch 76. However, the controller 30 may activate the buried object estimation function mode when determining that an excavation operation is performed based on the attitude of the attachment AT and the like, regardless of an operation on the buried object estimation mode switch 76. For example, the controller 30 can continuously execute the buried object estimation function mode from the time when the excavation operation is started until the time when a boom raising operation is performed.

The excavation pressure sensor S1 are examples of detection devices, which are devices for detecting information about an excavation reaction force, and detect the pressure of the hydraulic oil in the hydraulic cylinders, such as the boom cylinder 7 and the like, and output the detected data to the controller 30. The excavation pressure sensors S1 according to the embodiment are composed of a combination of excavation pressure sensors S11 to S18. The excavation pressure sensor S11 detects a boom bottom pressure, which is the pressure of the hydraulic oil in a bottom-side oil chamber of the boom cylinder 7. The excavation pressure sensor S12 detects a boom rod pressure, which is the pressure of the hydraulic oil in a rod-side oil chamber of the boom cylinder 7. Similarly, the excavation pressure sensor S13 detects an arm bottom pressure, the excavation pressure sensor S14 detects an arm rod pressure, the excavation pressure sensor S15 detects a bucket bottom pressure, and the excavation pressure sensor S16 detects a bucket rod pressure. The excavation pressure sensor S17 detects a left turning pressure, which is the pressure of the hydraulic oil at a first port (left-side port) of the turning hydraulic motor 2A. The excavation pressure sensor S18 detects a right turning pressure, which is the pressure of the hydraulic oil at a second port (right-side port) of the turning hydraulic motor 2A.

A control valve E1 is a valve that operates in accordance with an instruction from the controller 30. In the illustrated example, the control valve E1 is used to forcibly operate the flow rate control valves related with predetermined hydraulic cylinders, regardless of the content of an operation that is input into the operation device 26. The control valve E1 is equivalent to the solenoid valve disposed between the flow rate control valves and the pilot pump 15 in a case of employing the above-described electric operation device.

FIG. 4 is a diagram showing a configuration example of an excavation control system mounted on the excavator 100 shown in FIG. 1. The excavation control system includes the attitude detection devices M1, the excavation pressure sensors S1, the operation mode switch 75, the buried object estimation mode switch 76, the controller 30, the control valve E1, the display device 40, the sound output device 45, and the like. The controller 30 forms an excavation reaction force calculation part 31 and a buried object estimation part 32 in the controller 30 by a processor reading out and executing a program stored in the memory.

The excavation reaction force calculation part 31 is a functional element for calculating an excavation reaction force. The excavation reaction force calculation part 31 is configured to calculate an excavation reaction force based on at least an output from the excavation pressure sensors S1. The excavation reaction force calculation part 31 according to the embodiment calculates an excavation reaction force based on an output from the excavation pressure sensors S1 and the attitude of the attachment AT detected by the attitude detection devices M1. The excavation reaction force calculation part 31 may additionally use an output from a vehicle body inclination sensor. The vehicle body inclination sensor may be composed of, for example, an acceleration sensor or a gyro sensor.

Examples of an output from the excavation pressure sensors S1 include, for example, at least one of the boom bottom pressure, the boom rod pressure, the arm bottom pressure, the arm rod pressure, the bucket bottom pressure, and the bucket rod pressure detected by the excavation pressure sensors S11 to S16.

The excavation reaction force calculation part 31 may calculate a cylinder thrust based on an output from the excavation pressure sensors S1. A cylinder thrust is calculated based on, for example, an excavation pressure and a pressure receiving area of a piston sliding in a cylinder. Examples of a cylinder thrust include, for example, a boom cylinder thrust (f1), an arm cylinder thrust (f2), and a bucket cylinder thrust (f3). Specifically, the boom cylinder thrust (f1) is expressed as the difference between a cylinder extension force, which is the product between the boom bottom pressure and the pressure receiving area of the piston in the boom bottom-side oil chamber, and a cylinder contraction force, which is the product between the boom rod pressure and the pressure receiving area of the piston in the boom rod-side oil chamber. The same applies to the arm cylinder thrust (f2) and the bucket cylinder thrust (f3).

The excavation reaction force calculation part 31 may calculate an excavation torque based on the attitude of the attachment AT and a cylinder thrust. As shown in FIG. 2, the magnitude of a bucket excavation torque (τ3) is expressed as a value obtained by multiplying the magnitude of the bucket cylinder thrust (f3) by a distance G3 between a line of action of the bucket cylinder thrust (f3) and the bucket connection pin position P3. The distance G3 is a function of the bucket angle θ3, and is an example of a link gain. The same applies to a boom excavation torque (τ1) and an arm excavation torque (τ2). A distance G1 is the distance between a line of action of the boom cylinder thrust (f1) and the boom foot pin position P1, and a distance G2 is the distance between the line of action of the arm cylinder thrust (f2) and the arm connection pin position P2.

The excavation reaction force is calculated as, for example, the product between a mechanism function in which the boom angle θ1, the arm angle θ2, and the bucket angle θ3 shown in FIG. 2 are arguments, and a function in which the boom excavation torque (τ1), the arm excavation torque (τ2), and the bucket excavation torque (τ3) are arguments. The function in which the boom excavation torque (τ1), the arm excavation torque (τ2), and the bucket excavation torque (τ3) are arguments may be a function in which the boom cylinder thrust (f1), the arm cylinder thrust (f2), and the bucket cylinder thrust (f3) are arguments. The function in which the boom angle θ1, the arm angle θ2, and the bucket angle θ3 are arguments may be a function based on a force balance equation, may be a function based on Jacobian, or may be a function based on the principle of virtual work.

As described above, the value of the excavation reaction force can be derived based on currently detected values of the various sensors. However, the detected value of the excavation pressure sensors S1 may be used as is as the information of the excavation reaction force. Alternatively, the values of the cylinder thrusts calculated based on the detected values of the excavation pressure sensors S1 may be used as the information of the excavation reaction force. As the value of the excavation reaction force, the values of the excavation torques calculated from the values of the cylinder thrusts calculated based on the detected values of the excavation pressure sensors S1, and values related to the attitude of the attachment AT derived based on the detected values of the attitude detection devices M1 may be used.

The excavation reaction force calculation part 31 may calculate an excavation reaction force acting in the turning direction based on outputs from the excavation pressure sensors S17 and S18. When the left turning pressure (P17) detected by the excavation pressure sensor S17 is higher than the right turning pressure (P18) detected by the excavation pressure sensor S18, the upper turning body 3 is going to turn in the left direction. Conversely, when the right turning pressure (P18) detected by the excavation pressure sensor S18 is higher than the left turning pressure (P17) detected by the excavation pressure sensor S17, the upper turning body 3 is going to turn in the right direction. The excavation reaction force calculation part 31 may calculate, for example, the left turning pressure (P17) in a case of the left turning pressure (P17) being higher than the right turning pressure (P18), as the excavation reaction force acting in the left turning direction. The excavation reaction force calculation part 31 may calculate, for example, the right turning pressure (P18) in a case of the right turning pressure (P18) being higher than the left turning pressure (P17), as the excavation reaction force acting in the right turning direction. When a turning electric motor is mounted instead of the turning hydraulic motor 2A, the excavation reaction force calculation part 31 may calculate the excavation reaction force acting in the turning direction based on information related to electric power, such as the direction and the magnitude of a current supplied to the turning electric motor.

The buried object estimation part 32 is configured to detect a buried object based on information related to the excavation reaction force. The buried object estimation part 32 according to the embodiment estimates the presence or absence of a buried object based on the excavation reaction force calculated by the excavation reaction force calculation part 31 and a learning model for buried object estimation, which is previously gotten hold of. This learning model will be described later in detail.

For example, the buried object estimation part 32 outputs a control instruction to the control valve E1 when estimating that there is a buried object. Upon receiving a control instruction from the buried object estimation part 32, the control valve E1 forcibly operates the flow rate control valves related to the predetermined hydraulic cylinders to forcibly extend or contract the predetermined hydraulic cylinders regardless of the content of an operation that is input into the operation device 26. For example, even when the boom operation lever is not operated, the control valve E1 forcibly extends the boom cylinder 7 by forcibly moving the flow rate control valve related to the boom cylinder 7. As a result, the excavator 100 can forcibly raise the boom 4 to reduce the excavation depth (or change courses). Alternatively, even when the arm operation lever is operated, the control valve E1 may forcibly stop the arm cylinder 8 by forcibly moving the flow rate control valve related to the arm cylinder 8. By forcibly stopping the arm 5, the excavator 100 can avoid a contact between the bucket 6 and the buried object. In this way, the control valve E1 can reduce contacts between the attachment AT and the buried object by forcibly extending or contracting, or stopping at least one of the boom cylinder 7, the arm cylinder 8, or the bucket cylinder 9 in accordance with a control instruction from the buried object estimation part 32.

The buried object estimation part 32 may output a control instruction to the display device 40 when estimating that there is a buried object. Upon receiving the control instruction from the buried object estimation part 32, the display device 40 displays an estimated position of the buried object. For example, the display device 40 may display a virtual viewpoint image representing a state of the excavator 100 when viewed from a virtual viewpoint present immediately above the excavator 100, and may display the buried object buried in the ground by superimposing it on the virtual viewpoint image. The virtual viewpoint image is generated based on images acquired by the front camera 70F, the rear camera 70B, the left camera 70L, and the right camera 70R. Alternatively, the display device 40 may display an image representing a cross-section of the ground where the excavator 100 is located, and may display the buried object buried in the ground by superimposing it on the image representing the cross-section. Further, the buried object estimation part 32 may output a control instruction to the sound output device 45 when the buried object is estimated to be present.

The controller 30 starts the buried object estimation function mode in accordance with a start instruction from the buried object estimation mode switch 76. When the buried object estimation function mode is started, the buried object estimation part 32 estimates the presence or absence of a buried object based on an excavation reaction force calculated by the excavation reaction force calculation part 31. On the other hand, the controller 30 stops the buried object estimation function mode in accordance with a stop instruction from the buried object estimation mode switch 76. This can inhibit the excavator 100 from erroneously estimating that there is a buried object in response to a fluctuation in the excavation reaction force, and from outputting a control instruction to the control valve E1, the display device 40, or the sound output device 45, even though it is clear that there is no buried object. When the buried object estimation function mode is stopped, the excavation reaction force calculation part 31 does not need to calculate an excavation reaction force, so that the calculation load can be reduced.

The buried object estimation function mode can be executed both when the operation mode of the excavator 100 is the M (manual) mode and when it is the SA (semi-automatic) mode. However, the buried object estimation function mode may be executed only when the SA (semi-automatic) mode is selected. When the SA (semi-automatic) mode is selected, the operator can improve the buried object detection accuracy by moving the attachment AT along a previously set target course.

Next, with reference to FIG. 5, a movement of the excavator 100 when the excavator 100 estimates a water pipe U1 as a buried object will be described as a representative case. FIG. 5 is a diagram showing a cross-section of a ground in which the water pipe U1 is buried. In FIG. 5, the ground ES before being excavated is indicated by a broken line.

First, the operator of the excavator 100 operates the operation mode switch 75 to switch the operation mode of the excavator 100 to the SA (semi-automatic) mode. The operator manually operates the operation device 26 to move the claw tip of the bucket 6 to a desired position (first position PS1). After moving the claw tip of the bucket 6 to the desired position, the operator operates the buried object estimation mode switch 76 to start the buried object estimation function mode.

In the SA (semi-automatic) mode, the controller 30 autonomously move the attachment AT. Specifically, the controller 30 automatically extend or contract at least one of the boom cylinder 7, the arm cylinder 8, or the bucket cylinder 9 in a manner to move a predetermined part of the attachment AT along a previously set target course TP. However, even in the buried object estimation function mode, the controller 30 does not need to autonomously move the attachment AT, and may move the attachment AT in accordance with the content of an operation on the operation device 26 by the operator.

The controller 30 automatically moves the attachment AT such that the claw tip of the bucket 6 moves along a previously set first target course TP1 (dash-dotted line) in the excavation work. When the claw tip of the bucket 6 moves along the first target course TP1, the excavation reaction force calculation part 31 repeats calculation of the excavation reaction force based on outputs from the attitude detection devices M1 and outputs from the excavation pressure sensors S1. The buried object estimation part 32 repeats estimation of the presence or absence of a buried object based on the excavation reaction force calculated by the excavation reaction force calculation part 31.

When the claw tip of the bucket 6 reaches the end of the first target course TP1, the controller 30 stops autonomously moving the attachment AT. This means that the buried object estimation part 32 has not detected any buried object right up to the last point at which the claw tip of the bucket 6 reaches the end of the first target course TP1.

Thereafter, the operator operates the buried object estimation mode switch 76 to stop the buried object estimation function. The controller 30 may automatically stop the buried object estimation function when the operator performs a boom raising operation or a boom raising and turning operation by manually operating the operation device 26. After performing an earth dumping operation and a boom lowering and turning operation, the operator moves the claw tip of the bucket 6 to the next desired position (second position PS2) in order to perform the next excavation operation. For example, the second position PS2 is located at a depth D1 from the ground ES before being excavated, and is located almost directly under the first position PS1.

After moving the claw tip of the bucket 6 to the second position PS2, the operator activates the buried object estimation function mode. When the buried object estimation function mode is activated, the controller 30 automatically moves the attachment AT such that the claw tip of the bucket 6 moves along a previously set second target course TP2 (dash-dotted line). When the claw tip of the bucket 6 moves along the second target course TP2, the excavation reaction force calculation part 31 repeats calculation of the excavation reaction force based on outputs from the attitude detection devices M1 and outputs from the excavation pressure sensors S1. The buried object estimation part 32 repeats estimation of the presence or absence of a buried object based on the excavation reaction force calculated by the excavation reaction force calculation part 31.

When the claw tip of the bucket 6 reaches the end of the second target course TP2, the controller 30 stops autonomously moving the attachment AT. This means that the buried object estimation part 32 has not detected any buried object right up to the last point at which the claw tip of the bucket 6 reaches the end of the second target course TP2.

Thereafter, the operator performs an earth dumping operation and a boom lowering and turning operation in the same manner as described above, and then moves the claw tip of the bucket 6 again to the next desired position (third position PS3) in order to perform the next excavation operation. The third position PS3 is located at a depth D2 from a first exposed surface and almost directly under the second position PS2.

After moving the claw tip of the bucket 6 to the third position PS3, the operator activates the buried object estimation function. When the buried object estimation function is activated, the controller 30 automatically moves the attachment AT such that the claw tip of the bucket 6 moves along a previously set third target course TP3 (dash-dotted line).

When the claw tip of the bucket 6 moves along the third target course TP3, the excavation reaction force calculation part 31 repeats calculation of the excavation reaction force based on outputs from the attitude detection devices M1 and outputs from the excavation pressure sensors S1. The buried object estimation part 32 repeats estimation of the presence or absence of a buried object based on the excavation reaction force calculated by the excavation reaction force calculation part 31.

For example, the buried object estimation part 32 estimates that there is a buried object when the claw tip of the bucket 6 reaches a fourth position PS4. The fourth position PS4 is located at a depth D3 from a second exposed surface and on the third target course TP3. The fourth position PS4 is a position at which the distance between the water pipe U1, which is a buried object existing in a direction along the third target course TP3, and the claw tip of the bucket 6 becomes a value AD1.

Referring to FIG. 6, a process for estimating the presence or absence of a buried object based on the excavation reaction force will be described. FIG. 6 is a graph showing the relationship between the excavation reaction force F and the approach distance AD. The vertical axis of FIG. 6 is the excavation reaction force F calculated by the excavation reaction force calculation part 31, and the horizontal axis of FIG. 6 is the approach distance AD. The approach distance AD is the distance between the current position of the claw tip of the bucket 6 and the buried object (water pipe U1) in the direction along the target course TP. FIG. 6 shows that the approach distance AD decreases from the left to the right on the horizontal axis until the approach distance AD becomes zero. That is, the claw tip of the bucket 6 is located farther from the water pipe U1 when the approach distance AD is the value AD0 than when the approach distance AD is the value AD1.

Specifically, the dashed double-dotted line in FIG. 6 exemplifies a learning model serving as a reference of the excavation reaction force. That is, the learning model is represented by a function or the like representing the relationship between the change in the approach distance AD and the excavation reaction force in the excavation work. The function representing the learning model may be any of a linear function, a polynomial function, a logarithmic function, or the like. This learning model has learned the excavation reaction force that the attachment AT receives from the excavation target (in-ground) when there is no buried object. In the learning model, the excavation reaction force F increases as the approach distance AD decreases. This is because the excavation reaction force increases as the amount of earth and sand loaded into the bucket 6 increases as the bucket 6 approaches the machine body (upper turning body 3).

The buried object estimation part 32 compares this learning model with the excavation reaction force repeatedly calculated in the excavation work to estimate the presence or absence of a buried objects. For example, the excavation reaction force when the claw tip of the bucket 6 moves along the third target course TP3 in FIG. 5 is shown by a solid line in FIG. 6.

In this case, in the approach distance AD range of AD0 to AD1, the excavation reaction force fluctuates substantially in accordance with the learning model. However, when the approach distance AD exceeds AD1, the excavation reaction force diverges from the learning model and greatly increases. This is because the water pipe U1 exists on the third target course TP3 of the bucket 6, and as the claw tip of the bucket 6 approaches the water pipe U1, the earth and sand between the bucket 6 and the water pipe U1 is compressed between them.

The buried object estimation part 32 calculates, for example, the difference between the learning model and the actual excavation reaction force, and determines whether or not the difference between the learning model and the actual excavation reaction force is equal to or greater than a threshold value, thereby estimating the presence or absence of a buried object (water pipe U1). FIG. 6 shows an example in which the difference between the learning model and the actual excavation reaction force becomes equal to or greater than the threshold value at a position at which the approach distance AD is AD2. That is, the position at which the approach distance is AD2 corresponds to the timing at which the claw tip of the bucket 6 reaches the fourth position PS4 on the third target course TP3 in FIG. 5. This position AD2 is, for example, a position that is approximately 20 cm away from the buried object. The position at which the approach distance AD is AD3 is a position at which the buried object (water pipe U1) is present.

However, the comparison method between the learning model and the actual excavation reaction force is not limited to the above, and various methods may be employed. For example, the buried object estimation part 32 may compare an average increasing rate of the actual excavation reaction force with respect to the approach distance AD with an average increasing rate of the learning model. When the average increasing rate of the actual excavation reaction force exceeds the average increasing rate of the learning model by more than a predetermined value, it can be estimated that there is a buried object.

Next, a buried object estimation system 200 for generating the above-described learning model for buried object detection and for providing the learning model will be described with reference to FIG. 7. FIG. 7 is a block diagram showing the buried object estimation system 200.

The buried object estimation system 200 includes an information processing device 210, a communication network 220, a cloud server 230, and a controller 30 of the excavator 100.

The information processing device 210 is an information processing part for generating a learning model to be used for buried object detection by the excavator 100. A well-known computer including a processor 211, a memory 212, an input/output interface 213, and a communication interface 214 may be applied as the information processing device 210. The information processing device 210 may be composed of one computer or a plurality of computers.

The information processing device 210 is provided, for example, in a company or an organization that manufactures, manages, or uses a work machine, such as the excavator 100 and the like. The information processing device 210 manages information on the use statuses and states of a plurality of excavators 100 or other work machines, and provides information to the excavators 100 or other work machines. The information processing device 210 is connected to the cloud server 230 via the communication network 220, and provides and acquires information via the cloud server 230. Alternatively, the information processing device 210 may be configured to perform information communication with a portable terminal, such as a computer, a smartphone, a tablet, and the like available for use by a worker via the cloud server 230 (or directly).

The communication network 220 of the buried object estimation system 200 may be the Internet, or a dedicated line, such as an Ethernet and the like. As described above, in the buried object estimation system 200 according to the embodiment, the information processing device 210 and each excavator 100 exchange information by accessing the cloud server 230 via the communication network 220. For example, when the excavator 100 uploads to the cloud server 230 various types of excavator information including an excavation reaction force when performing excavation work at a work site, the information processing device 210 acquires this information from the cloud server 230 at an appropriate timing.

Conversely, when the information processing device 210 uploads work information (not shown) to the cloud server 230, the excavator 100 downloads the work information from the cloud server 230 at an appropriate timing prior to the work. The work information includes position information of the work site, information on a 3D model or a 2D model of the work site, information on work machines used at the work site, a learning model for buried object detection, and the like.

In the buried object estimation system 200, the information processing device 210 and the excavator 100 may be configured to perform information communication directly without the cloud server 230 (or via another computer). Alternatively, a worker may provide information about the excavator 100 to the information processing device 210 by storing the information about the excavator 100 in an external storage device and connecting the external storage device to the information processing device 210. The same applies to the information to be provided from the information processing device 210 to the excavator 100.

Next, a learning model for buried object detection generated by the information processing device 210 will be described with reference to FIG. 8. FIG. 8 is a block diagram showing functional parts of the information processing device 210 for generating a learning model. The information processing device 210 generates a learning model for buried object detection by, for example, an unsupervised learning method. As an example, by the processor 211 reading and executing a program stored in the memory 212, the information processing device 210 internally constructs a sensor data storage part 215, a training data extraction part 216, a reaction force model generation part 217, an optimization part 218, an evaluation part 219, and the like.

The sensor data storage part 215 stores information acquired by various sensors of the excavator 100. Here, the information processing device 210 stores in the sensor data storage part 215, pieces of information about excavators 100 of the same type in association with each other, and sensor values and time information in association with each other. Information stored in the sensor data storage part 215 includes information related to the excavation reaction force (sensor data from the excavation pressure sensors S1 and attitude detection devices M1). Further, the information stored in the sensor data storage part 215 may include information on a field external to the excavator 100 detected by the object detection device 70 (a camera, LiDAR, and the like) of the excavator 100. Further, in a case of implementing a driving assistance for the excavator 100, the sensor data storage part 215 may store simulation data obtained by simulating an operation of the excavator 100 including excavation work by the attachment AT. The simulation data may be one that has been actually used by an excavator 100 in excavation work, or may be one that is obtained from a simulation previously performed by the information processing device 210 so as to be provided to the excavator 100.

The training data extraction part 216 is the input layer of machine learning, for extracting data to be used for generating a learning model for buried object detection from various types of information stored in the sensor data storage part 215 and providing the data to the reaction force model generation part 217. The data used for generating the learning model includes above-described information related to the excavation reaction force (the pressures from the excavation pressure sensors S1 and the like), information related to the attitude of the excavator 100, information from the object detection device 70.

The reaction force model generation part 217 generates a learning model for buried object detection based on the data provided from the training data extraction part 216. Here, the excavation target excavated by the excavator 100 has different soil textures, such as earth, sand, earth and sand, hard bedrock, weak rock mass, and the like. Since different soil textures produce different excavation reaction forces, the reaction force model generation part 217 generates a learning model for each different soil texture. Therefore, learning models are generated, including an earth learning model used in earth excavation, a sand learning model used in sand excavation, a gravel learning model used in gravel excavation, and the like.

For example, the reaction force model generation part 217 recognizes a pattern of sequential data of the excavation reaction force by a time series analysis neutral net, using an extracted feature quantity of the excavation reaction force. The reaction force model generation part 217 extracts a feature quantity of topographic data based on information (captured image information) from a camera or a LiDAR, representing actual excavation work performed by the excavator 100, or based on construction data. Then, the reaction force model generation part 217 generates a reaction force prediction model by combining the pattern of the excavation reaction force and the feature quantity of the topographic data in the fully-connected layer. The reaction force prediction model is, for example, a function or table information indicating a change in the excavation reaction force with respect to the elapse of time or a change in the approach distance AD. The reaction force prediction model is calculated for each soil texture type (earth, sand, earth and sand, and the like) based on the feature quantity of the topographic data.

The optimization part 218 optimizes reaction force prediction models and generates learning models (an earth learning model, a sand learning model, an earth and sand learning model, and the like) based on one or more reaction force prediction models generated by the reaction force model generation part 217 being input. For example, the optimization part 218 executes the Bayesian optimization process based on the reaction force prediction models, and the feature quantity of the topographic data or any other excavation work condition. The optimization part 218 can obtain a learning model with high accuracy by using a plurality of reaction force prediction models generated by the reaction force model generation part 217. The optimization process of the optimization part 218 is not limited to the Bayesian optimization process, and various techniques, such as grid search and the like, may be employed.

The evaluation part 219 evaluates how well a generated learning model is and outputs the evaluation result to the reaction force model generation part 217. For example, the evaluation part 219 may evaluate the learning model based on information on actual presence or absence of a buried object confirmed by a revealing work described later. Alternatively, learning model evaluation may be performed by optimization of a hyperparameter that can be set or adjusted by a user of the information processing device 210. Thus, a learning model can be generated in a manner to satisfy a desired requirement, such as improving the performance of the learning model, smoothly obtaining a learning model, and the like.

When the information processing device 210 generates one or more learning models (an earth learning model, a sand learning model, and an earth and sand learning model) for buried object detection by the above-described process, the information processing device 210 uploads the learning models to the cloud server 230. Thus, the excavator 100 at the work site can download the learning models by accessing the cloud server 230.

The controller 30 of the excavator 100 estimates the presence or absence of a buried object during the excavation work as described above by using the learning models downloaded during the excavation work. Here, the excavator 100 can perform buried object estimation based on the soil texture, by selecting a learning model corresponding to the soil texture of the excavation target from among the plurality of types of learning models (the earth learning model, the sand learning model, and the earth and soil learning model). For example, when the operator of the excavator 100 recognizes the soil texture of the excavation target, the operator operates the input part 42 of the display device 40 and inputs the soil texture of the excavation target. Thus, the controller 30 can select the learning model to be used from among the plurality of types of learning models.

Alternatively, the controller 30 may be configured to automatically select the soil texture of the excavation target based on captured image information acquired by a camera and the like of the object detection device 70 of the excavator 100. This makes it possible to immediately switch to an appropriate learning model even when, for example, the soil texture changes partway during excavation and the operator does not notice it.

However, the hardness and the viscosity of the excavation target excavated by the excavator 100 in an actual operation phase change due to the impact of air temperature and humidity, or moisture content in the soil associated with these conditions, and the like. Therefore, the excavator 100 may not be able to detect a buried object by simply using the learning model that the excavator 100 gets hold of, which may fail to serve as the excavation reaction force matching the actual work site. Therefore, the controller 30 of the excavator 100 performs a process for tuning (adjusting) the learning model to match the actual excavation target, through excavation of the excavation target at the actual work site for confirmation of the state of the excavation target.

FIG. 9 is an explanatory diagram showing the excavator 100’s acquisition of a learning model, tuning of the learning model, and use of the learning model during excavation work. As shown in FIG. 9, the excavator 100 performs a previous excavation work in an area where it has become clear that no buried object is present at the work site. In the previous excavation work, the excavator 100 detects an actual excavation reaction force by various sensors (the attitude detection devices M1, and the excavation pressure sensors S1).

The controller 30 can determine whether or not the learning model needs tuning, by comparing the degree of change in the excavation reaction force at the actual work site with the learning model that the controller 30 gets hold of. When determining that the learning model needs tuning, the controller 30 tunes the learning model by using the actual excavation reaction force at the work site. For example, in the tuning of the learning model, the learning model (predicted excavation reaction force) represented by a linear function, a polynomial function, a logarithmic function, or the like is fitted with plotted data of the actual excavation reaction force at the work site. As a result, the learning model is appropriately corrected to a model that would indicate a change in the excavation reaction force matching the actual work site. Hereinafter, the learning model that is corrected is also referred to as a corrected learning model.

The tuning at the work site may be performed only for a previously selected learning model (for example, an earth learning model), or may be performed for all of a plurality of types of learning models (for example, an earth learning model, a sand learning model, and a gravel learning model) that are gotten hold of. At the work site, the soil texture may change in the process of excavation work. Therefore, when the controller 30 has calculated corrected learning models for a plurality of types of learning models, model switching can be smoothly performed.

After performing the previous excavation work, the controller 30 performs the excavation work shown in FIG. 5 using the corrected learning model. That is, by performing the excavation work in the SA (semi-automatic) mode, the excavator 100 autonomously moves the attachment AT, while the excavation pressure sensors S1 and the attitude detection devices M1 acquire the respective detection values. The excavation reaction force calculation part 31 of the controller 30 calculates the excavation reaction force based on these sensors.

The buried object estimation part 32 of the controller 30 compares the calculated excavation reaction force with the corrected learning model to estimate the presence or absence of a buried object in the excavation target during the excavation work by the attachment AT. As described above, since the corrected learning model has been tuned to match the actual work site, the controller 30 can accurately estimate that a buried object is present in the excavation target.

When it is estimated that a buried object is present in the excavation target, the excavator 100 may autonomously control the movement of the attachment AT such that the attachment AT does not contact the buried object. Specifically, the controller 30 disables the arm closing operation based on the detection of the buried object to stop the attachment AT from moving. Alternatively, the controller 30 may change courses such that the claw tip of the bucket 6 does not contact the buried object, by automatically extending or contracting the boom cylinder 7 to raise the boom 4.

When it is estimated that a buried object is present in the SA (semi-automatic) mode or the M (manual) mode, it is preferable that the controller 30 calls the attention of the operator such that the attachment AT does not approach the buried object. For example, the controller 30 may inform the operator of the level of the distance between the claw tip of the bucket 6 and the buried object via the sound output device 45. In this case, the controller 30 may make the interval between intermittent sounds shorter as the distance between the bucket 6 and the buried object becomes smaller. When the claw tip of the bucket 6 approaches the buried object very much, the controller 30 may issue a loud alarm to the operator via the sound output device 45.

When the controller 30 of the excavator 100 estimates that a buried object is present, a buried object revealing work may be performed at the work site to confirm the presence or absence of a buried object. For example, in the revealing work, the worker at the work site digs the location where an object is estimated to be buried by using a tool, thereby revealing any buried object from the excavation target. In the revealing work, the operator of the excavator 100 may confirm the presence or absence of a buried object by excavating the location where an object is estimated to be buried by carefully operating the attachment AT. When the presence or absence of a buried object is actually confirmed by the revealing work at the location where a buried object is estimated to be present, the worker may transmit this information to the controller 30 of the excavator 100 and/or the cloud server 230 by using an information terminal device available for use by the worker. As a result, the controller 30 of the excavator 100 and/or the cloud server 230 stores the data of the excavation reaction force at which it is estimated that an object is buried, and the result indicating the presence or the absence of a buried object in the revealing work in association with each other.

In a case of receiving information indicating that a buried object is actually present in the revealing work, the controller 30 or the cloud server 230 can recognize that the buried object detection using the learning model (corrected learning model) has functioned correctly. On the other hand, in a case of receiving information indicating that there is actually no buried object in the revealing work, the controller 30 or the cloud server 230 can recognize that the buried object detection using the learning model (corrected learning model) was incorrect. Here, the controller 30 can newly correct the learning model (or the threshold value for determining a buried object) based on the obtained information. Thus, the excavator 100 can execute the buried object estimation function mode using the corrected learning model or the threshold value that is newly corrected.

Furthermore, the controller 30 may display information related to buried object estimation on the display device 40 serving as a user interface when the buried object estimation function mode is executed. FIG. 10 is a diagram illustrating image information 400 displayed on the display device 40 when the buried object estimation function mode is executed.

The image information 400 may be automatically displayed along with execution of the buried object estimation function mode regardless of the SA (semi-automatic) mode or the M (manual) mode. The image information 400 includes a mode selection display section 410, a selected model display section 420, an abnormality degree indicator 430, a model parameter 440, a model accuracy display section 450, a reaction force display section 460, an environment display section 470, an external connection setting display section 480, and a detail setting display section 490.

In the mode selection display section 410, a plurality of types of learning modes, a tuning mode for tuning the learning mode, or the buried object estimation function mode in which an actual excavation work is performed can be selected by an operation by the operator. For example, the operator operates a switch provided on the lever 26A or the lever 26B of the operation device 26 to perform an operation for turning on the mode selection display section 410. In response to being turned on, the mode selection display section 410 displays a selection window including various modes, for the operator to select a desired mode.

The selected model display section 420 displays various learning models (an earth learning model, a sand learning model, a gravel learning model, and the like) selected by the operator. After the tuning mode is executed, the selected model display section 420 may display an indication representing that the model has been tuned (a pictogram representing a corrected learning model, and the like).

When the buried object estimation function is executed in actual excavation work, the abnormality degree indicator 430 converts the possibility of presence of a buried object into the abnormality degree and notifies it to the operator. For example, the abnormality degree is calculated as the ratio between the difference of the excavation reaction force from the learning model, and the threshold value retained. In the illustrated example, the abnormality degree is notified by a circular ring graph and a numerical value. The color of the circular ring graph may change in accordance with the abnormality degree, by, for example, changing to green when the abnormality degree is in the range of 0% to 50%, to yellow when the abnormality degree is in the range of 50% to 80%, and to red when the abnormality degree is in the range of 80% to 100%.

When the abnormality degree is high (in other words, when the difference from the threshold is large), the controller 30 can notify the operator of the buried object via the abnormality degree indicator 430. When the abnormality degree is high, the controller 30 may automatically stop moving the attachment AT or change courses as described above. Alternatively, the operator may monitor the abnormality degree and determine, for example, to stop moving the attachment AT when the abnormality degree is 70%, to stop moving the attachment AT when the abnormality degree is 90%, or the like.

The model parameter 440 displays the hyperparameters in the learning model (corrected learning model) in a manner that the hyperparameters can be set by the operator. The hyperparameters include, for example, the soil texture at the work site, the detection accuracy of the buried object estimation function, and the operation mode of the excavator 100.

The model accuracy display section 450 displays the result of evaluation of a currently applied learning model (corrected learning model). For example, in the evaluation of the learning model, the score is increased in a case where there is actually no buried object while absence of a buried object is being detected, and the score is decreased in a case where there is a buried object while absence of a buried object is being detected. Alternatively, in the evaluation of the learning model, the score is increased in a case where there is actually a buried object when presence of a buried object is detected, and the score is decreased in a case where there is actually no buried object when presence of a buried object is detected.

The reaction force display section 460 displays the learning model (corrected learning model) actually used in the excavation work and the excavation reaction force detected by the sensors (the attitude detection devices M1 and the excavation pressure sensors S1) of the excavator 100 in the form of a graph. For example, the graph is information indicating a change in the excavation reaction force with respect to a change in the excavation distance by representing the excavation distance on the horizontal axis and the excavation reaction force on the vertical axis. Alternatively, the graph is information showing a change in the excavation reaction force with respect to the elapse of time by representing the excavation time on the horizontal axis and the excavation reaction force on the vertical axis. Thus, the operator can operate the excavator 100 while actually recognizing the learning model and change in the actual excavation reaction force in real-time.

The environment display section 470 displays environmental factor variables. The environmental factor variables include, for example, the date and time when the excavator 100 performs excavation work, weather, operator ID, and the like.

The external connection setting display section 480 displays the communication environment between the controller 30 and the external communication network. For example, when connecting to the cloud server 230, the external connection setting display section 480 can be operated by the operator, such that a state of being connected with the cloud server 230 can be set.

The detail setting display section 490 enables the operator to change settings other than those described above associated with the execution of the buried object estimation function.

With the image information 400 described above displayed on the display device 40, the operator of the excavator 100 can make settings related to the learning model in the buried object estimation function and can well recognize the relationship between the learning model and the excavation reaction force when actually performing the excavation work. In particular, by displaying the learning model, the excavator 100 according to the embodiment can inform the operator that the excavator 100 is determining that there is a buried object because the excavation reaction force during the excavation work is diverging from the learning model.

The excavator 100 according to the embodiment is basically configured as described above, and this operation will be described below with reference to the flowcharts of FIGS. 11A and 11B. FIG. 11A is a flowchart showing a process flow before actually performing the excavation work. FIG. 11B is a flowchart showing a buried object estimation method when actually performing the excavation work.

In the buried object estimation method, the controller 30 of the excavator 100 performs steps S101 to S109 of FIGS. 11A and 11B.

The controller 30 accesses the cloud server 230 to acquire various learning models stored in the cloud server 230 (step S101). The acquisition pattern of the learning models is not limited to this, and a memory device storing the learning models may be connected to the controller 30, such that the controller 30 may acquire the learning models. Alternatively, the controller 30 may store the learning models in advance. In this case, step S101 may be omitted.

The controller 30 displays a display prompting tuning of the learning model at the actual work site, and prompts the operator of the excavator 100 to perform a previous excavation work, thereby tuning the learning model (step S102). By performing this tuning, the controller 30 can acquire a corrected learning model matching the actual work site. However, the operator may select not to perform tuning, and in this case, the learning model that the controller 30 gets hold of is used as is. Further, for example, when a plurality of excavators 100 are used at the same work site, tuning may be performed by one excavator 100 and information on the tuning result (corrected learning model) may be stored in the cloud server 230. Then, the other excavators 100 acquire the tuning result from the one excavator 100 from the cloud server 230. Thus, tuning by a plurality of excavators 100 can be omitted. Alternatively, tuning may be performed by a plurality of excavators 100, and the tuning results may be stored in the cloud server 230, such that the learning model may be corrected using the plurality of tuning results in the cloud server 230 and transmitted to each excavator 100. Thus, the tuning accuracy of the learning model of each excavator 100 can be improved.

Then, the excavator 100 shifts to the excavation work at the actual work site. The controller 30 determines whether the operator has performed selection of the operation mode, selection of the buried object estimation function mode, and the like (step S103). When the buried object estimation function mode has been selected (step S103: YES), the process proceeds to step S104.

In step S104, the controller 30 reads out the stored learning model (corrected learning model). Here, the controller 30 displays the image information 400 described above such that the operator can view the learning model to be used.

Then, the excavator 100 moves the attachment AT based on the control from the controller 30 or an operation of the operator, to perform the excavation work on the excavation target (step S105).

In this excavation work, the controller 30 acquires sensor data of the attitude detection devices M1 and the excavation pressure sensors S1, calculates an excavation reaction force from these data, compares the calculated excavation reaction force with the learning model, and determines the presence or absence of a buried object (step S106). When it is determined that no buried object is present (step S106: NO), the process proceeds to step S107.

In step S107, the controller 30 determines whether to terminate the excavation work. When not terminating the excavation work (step S107: NO), the process returns to step S105 to continue the excavation work. On the other hand, when terminating the excavation work (step S107: YES), the current process flow is ended.

When it is determined in step S106 that a buried object is present (step S106: YES), the process proceeds to step S108. In step S108, the controller 30 performs stopping of the movement of the attachment AT, course change, and the like as described above, and displays information indicating the presence of a buried object on the display device 40. Thus, the operator can recognize that the excavator 100 has stopped operating because a buried object is present during the excavation on the excavation target.

Here, the operator of the excavator 100 may inform the nearby workers at the work site that a buried object has been detected, thereby prompting the workers to perform the revealing work of any buried object. Then, the workers provides feedback regarding the information as to whether or not there actually is a buried object to the controller 30 or the cloud server 230 (step S109).

As a result, the controller 30 can evaluate the learning model (corrected learning model) that is used, based on the information from the workers about the actual presence or absence of a buried object, thereby correcting the learning model or the threshold value to be used in the next excavation work. For example, when there is no buried object, the controller 30 can correct the function representing the learning model (corrected learning model), taking into consideration, the changes in the excavation reaction force that have occurred this time.

As described above, the excavator 100 and the buried object estimation method can suitably detect a buried object buried in the excavation target by performing the excavation work using the learning model trained by the information processing device 210. Moreover, the excavator 100 uses a corrected learning model matching the actual work site, by tuning the learning model acquired from the cloud server 230 at the actual work site. As a result, the excavator 100 can more accurately estimate the presence or absence of a buried object.

Next, a configuration example of an operation system SYS according to another embodiment will be described with reference to FIG. 12. FIG. 12 is a schematic diagram showing the configuration example of the operation system SYS. The operation system SYS includes an excavator 100, a remote control office RC, and a management center MC. The excavator 100 shown in FIG. 12 has the same configuration as that of the excavator 100 shown in FIG. 1.

The excavator 100, the remote control office RC, and the management center MC are connected to each other to be able to exchange data via a communication network NW. The excavator 100, the remote control office RC, and the management center MC may be connected to each other so as to be able to exchange data with each other directly without using the communication network NW. In the illustrated example, the excavator 100 transmits information about the work site to the remote control office RC. Thus, the remote operator RO in the remote control office RC can understand the state of the work site based on the information from the excavator 100.

The excavator 100 is equipped with sensors capable of three-dimensionally recognizing the position and shape of an object existing at the work site. For example, the excavator 100 is equipped with a spatial recognition device. Therefore, the excavator 100 can transmit the result of three-dimensionally measuring the work site to the remote control office RC.

The spatial recognition device is a device for recognizing the space around the excavator 100. In the illustrated example, the spatial recognition device is a LiDAR. The LiDAR measures, for example, the distance between each of one million or more points present in the monitoring range and the LiDAR. The spatial recognition device may be any device that can measure the distance to an object. For example, the spatial recognition device may be a stereo camera or a combination of an imaging device and a range finder, such as a millimeter-wave radar.

One or a plurality of excavators 100 may be included in the operation system SYS. When a plurality of excavators 100 are included, the remote operator RO of one specific excavator 100 can acquire information about the work site acquired by that one specific excavator 100, and information about the work site acquired by the other one or a plurality of excavators 100 as well.

A communication device T2, a remote controller RCC, an operation device 26E, an operation sensor 43, a display device D1E, an internal sound output device SP2E, and an internal sound collecting device M2E are installed in the remote control office RC. An operation seat DS to be seated by the remote operator RO who remotely operates the excavator 100 is installed in the remote control office RC.

The communication device T2 is configured to communicate with a communication device attached to the excavator 100.

The remote controller RCC is an arithmetic device for executing various arithmetic operations. In the present embodiment, the remote controller RCC is composed of a microcomputer including a CPU and a memory. Various functions of the remote controller RCC are realized by the CPU executing a program stored in the memory.

The display device D1E is a device capable of displaying various information. The display device D1E displays an image that is based on information transmitted from the excavator 100, such that the remote operator RO in the remote control office RC can visually recognize the surrounding state of the excavator 100. In the illustrated example, the display device D1E is a liquid crystal display for displaying an image captured by the imaging device mounted on the excavator 100. The display device D1E may be a display or a projector for realizing naked-eye stereopsis, and may be VR goggles and the like.

The internal sound output device SP2E is a device capable of outputting sound information. The internal sound output device SP2E outputs a sound based on information transmitted from the excavator 100 such that the remote operator RO in the remote control office RC can hear the sound occurring at the work site.

The operation device 26E is equipped with the operation sensor 43 for detecting the content of an operation on the operation device 26E. The operation sensor 43 is, for example, an inclination sensor for detecting the inclination angle of an operation lever, an angle sensor for detecting the swing angle of the operation lever about a swing shaft, and the like. The operation sensor 43 may be composed of any other sensor, such as a pressure sensor, a current sensor, a voltage sensor, a distance sensor, and the like. The operation sensor 43 outputs information about a detected content of an operation on the operation device 26E to the remote controller RCC. The remote controller RCC generates an operation signal based on the received information and transmits the generated operation signal to the excavator 100. The operation sensor 43 may be configured to generate the operation signal. In this case, the operation sensor 43 may output the operation signal to the communication device T2 by bypassing the remote controller RCC. With such a configuration, the remote operator RO can remotely operate the excavator 100 from the remote control office RC.

The management center MC is a facility equipped with various devices for managing the excavator 100 at the work site, or the remote operation of the excavator 100 by the remote operator RO in the remote control office RC and the like. In the illustrated example, the management center MC is installed at a location separated from both the work site of the excavator 100 and the remote control office RC. A management device 300, an internal sound output device SP2C, and an internal sound collecting device M2C are installed in the management center MC.

The management device 300 is an example of the control part, and is, for example, a server computer (generally referred to as a cloud server), or an edge server. The management device 300 is typically a fixed terminal device, but may be a portable terminal device (for example, a laptop computer, a tablet, a smartphone, or the like).

In the above-described operation system SYS as well, the excavator 100 can perform buried object estimation using the learning model. When the remote operator RO executes the buried object estimation function when performing excavation work at the work site, the controller 30 of the excavator 100 or the remote controller RCC reads out the learning model described above and compares it with the excavation reaction force. Thus, in the excavation work at the work site, it is possible to perform the excavation work by suitably determining the presence or absence of a buried object.

In the buried object estimation system 200 according to the embodiment, the information processing part for generating a learning model is provided in the information processing device 210 external to the excavator 100. However, the information processing part may be provided in the controller 30 of the excavator 100. In other words, the buried object estimation system 200 may be realized only by the configuration of the excavator 100.

The technical idea and effects of the present disclosure described in the above embodiments will be described below.

A first aspect of the present disclosure is the buried object estimation system 200 for estimating a possibility that a buried object is present in an excavation target, for the excavator 100 including: the lower traveling body 1; the upper turning body 3 turnably provided on the lower traveling body 1; the attachment AT provided on the upper turning body 3 and configured to excavate the excavation target; and the control part (controller 30) configured to control a movement of the attachment AT, wherein the buried object estimation system 200 includes an information processing part (information processing device 210, or information processing computer) configured to: train a learning model for detecting a buried object by acquiring information related to an excavation reaction force in excavation work by the attachment (AT); and transmit the learning model to the control part, and wherein the control part estimates presence or absence of a buried object in the excavation target based on an excavation reaction force acquired in actual excavation work and the learning model.

According to the foregoing description, the buried object estimation system 200 can suitably estimate a buried object in the excavation target during the excavation work by the excavator 100, by using the learning model trained on information related to the excavation reaction force. That is, the buried object estimation system 200 can estimate the presence or absence of a buried object from changes in the excavation reaction force with respect to the learning model, at a higher probability than can a configuration for simply monitoring the excavation reaction force acquired from the sensors of the excavator 100. In addition, the buried object estimation system 200, in which sensor data is accumulated every time excavation work is performed, can improve the learning model. As a result, a buried object can be estimated more accurately during excavation work performed by the excavator 100.

Further, the control part (controller 30) acquires information related to the state of the excavation target on which actual excavation work is performed, and tunes the learning model that the control part (controller 30) gets hold of, based on the information related to the state of the excavation target. Thus, according to the buried object estimation system 200, it is possible to tune the learning model to match the state of the excavation target (soil texture, moisture content, viscosity, weather, and the like) at the actual work site. As a result, the presence or absence of a buried object can be estimated in accordance with the state of the excavation target, and a buried object can be estimated more accurately.

Further, the information related to the state of the excavation target is information on the excavation reaction force acquired when the excavation target on which actual excavation work is to be performed is previously excavated. Thus, the buried object estimation system 200 can tune the learning model to the one matching the actual work site based on the information on the excavation reaction force acquired when the excavation target is previously excavated.

The information processing part (information processing device 210) generates a plurality of learning models for respective types of soil textures of the excavation target, and the control part (controller 30) selects the learning model to be used, based on the soil texture of the excavation target on which the actual excavation work is performed. Thus, the excavator 100 can suitably monitor the excavation reaction force in the actual excavation work, using the learning model matching the soil texture of the excavation target.

After determination that a buried object is present in the actual excavation work is made, the control part (controller 30) or the information processing part (information processing device 210) acquires information confirming the presence or absence of a buried object, and evaluates the learning model based on the information confirming the presence or absence of a buried object. Thus, the buried object estimation system 200 can improve the learning model based on the presence or absence of a buried object confirmed by revealing work, and can better improve the accuracy in the next buried object estimation.

The control part (controller 30) calculates the difference between the excavation reaction force in the actual excavation work and the learning model, and determines that a buried object is absent when the difference is less than a threshold value, and determines that a buried object is present when the difference is greater than or equal to the threshold value. Thus, the buried object estimation system can easily and accurately estimate the presence or absence of a buried object.

When determining that a buried object is present, the control part (controller 30) stops moving the attachment AT or changes courses. Thus, the excavator 100 can avoid the attachment AT contacting the estimated buried object.

The display device 40 for displaying the learning model used in actual excavation work is also provided. Thus, the operator of the excavator 100 can well recognize the learning model used in buried object estimation.

The display device 40 also displays the image information 400 including information related to the result of estimating the presence or absence of a buried object. Thus, the operator of the excavator 100 can immediately recognize the result of estimating the presence or absence of a buried object and take necessary measures.

A second aspect of the present disclosure is the excavator 100 including the lower traveling body 1; the upper turning body 3 turnably provided on the lower traveling body 1; the attachment AT provided on the upper turning body 3, and configured to excavate an excavation target; and a control part (controller 30) configured to control a movement of the attachment AT, wherein the control part is configured to: acquire a learning model for detecting a buried object by acquiring information related to an excavation reaction force in excavation work by the attachment AT; and estimate presence or absence of the buried object in the excavation target based on an excavation reaction force acquired in actual excavation work and the learning model. Also in this case, a buried object in the excavation target can be suitably estimated during the excavation work by the excavator 100.

The buried object estimation system 200 and the excavator 100 according to the embodiments disclosed herein are exemplary and not limiting in any respect. The embodiments can be modified and improved in various forms without departing from the scope and the spirit of the appended claims. The particulars described in the above plurality of embodiments can be configured in other forms to the extent that no contradiction occurs, and can be combined to the extent that no contradiction occurs.