Construction machinery with learning functions

Abstract

The construction machine (100) with a learning function of the present invention comprises: an action unit (103) having an operation unit (104); an operation unit (101); an operation status detection unit (112); an operation status detection unit (113); a reaction detection unit (114); and a learning data storage unit (115) which stores the instruction (201) output from the operation unit (101) as instruction data (211) in a time series, and stores the operation status data (212), the operation status data (213) and the reaction data (214) in a time series as prediction basic data (Pd). ; a learning unit (118), which uses the prediction basic data (PD') stored in the learning data storage unit (115) to perform mechanical learning on the instruction data (211') stored in the learning data storage unit (115), and after completing the mechanical learning, receives the input of the prediction basic data (Pd) when the action unit (103) acts, and outputs the prediction instruction (1103) of the instruction (201); and a hydraulic drive system (105), which drives the action unit (103) based on the instruction (201), the prediction instruction (1103), or the instruction (201) and the prediction instruction (1103).

Description

[0001] This application is a divisional application of the application with an international filing date of May 20, 2020, application number 202080007109.1, and invention name “Construction machinery with learning function”. Technical Field

[0002] The invention relates to a construction machine with a learning function. Background Art

[0003] As a construction machine having a learning function, for example, there is known a construction machine described in Patent Document 1. In this construction machine, a target value of an overlap length in the width direction of a rolled road is learned.

[0004] Patent Document 1: Japanese Patent No. 6360583

[0005] In the above-mentioned construction machine, it is impossible to learn the work performed by the construction machine through human operation and automatically perform the learned work.

[0006] Furthermore, when learning human-operated construction machinery, simply applying conventional learning methods requires extreme learning completion to maintain the quality of construction machinery operations at the same level as those performed by skilled operators. Furthermore, while the construction machinery's movements must be modified based on the specific operation, the infinite variety of possible operations requires extensive learning data and a debugging period to properly control the movements of the construction machinery based on the actual operation. Consequently, achieving automation of construction machinery operations in the short term is difficult.

[0007] Furthermore, in the construction industry, due to the declining birthrate and aging population, the workforce is aging and in short supply, making it imperative to pass on the skills of experienced workers. However, there is also a problem of a decrease in the number of people who can pass on these skills. Summary of the Invention

[0008] The present invention has been made to solve the above-mentioned problems, and a first object thereof is to provide a construction machine capable of learning at least the work performed by a human operator of the construction machine and automatically performing the learned work.

[0009] Furthermore, in addition to the first object, a second object is to provide a skill-transferring construction machine that can pass on the skills of skilled operators in the construction industry and can automate predetermined operations in a short period of time.

[0010] The present inventors have conducted intensive research to solve the above-mentioned problems and have obtained the following findings.

[0011] In construction machinery, multiple hydraulic actuators drive the working parts, including the working unit. For example, in a hydraulic excavator, the boom, arm, bucket, and operator's seat represent the working parts, while the hydraulic cylinders that rotate the boom, arm, and bucket, and the hydraulic motors that rotate the operator's seat, represent the hydraulic actuators.

[0012] Therefore, the operating unit is composed of a number of operating levers corresponding to the number of hydraulic actuators, and the operator operates these multiple operating levers to perform the work. In addition, the work performed by construction machinery is diverse. Therefore, it has been considered difficult to learn how to operate a construction machinery.

[0013] However, the present inventors have focused on the fact that there are relatively simple and standard tasks in construction work. If it is a relatively simple and standard task, it is relatively easy to learn.

[0014] On the other hand, in order to learn a task performed by an operator in a certain device, it is important to identify information that significantly affects the operator's operation and use it as input data for learning.

[0015] Therefore, the present inventors analyzed the operation of a skilled operator and found the following.

[0016] In short, it was found that the operator visually confirmed the working status, perceived the current operation state of the construction machine through visual observation and operation of the operating lever, and felt the reaction from the working object through the body, thereby deciding the next operation.

[0017] Taking the digging operation performed by a hydraulic excavator as an example, the operator visually confirms the digging status, visually identifies the posture of the bucket and the boom, and visually observes how the bucket, boom, and driver's seat are currently moving through the current operating position of the operating lever. In addition, if the bucket acts on the ground (digging the ground, scraping in sand, etc.), the operator feels its reaction through the body and judges whether the intended operation (action) is appropriate or not. Moreover, these are instantaneously considered to decide the next operation. Here, the so-called reaction refers to, for example, the tilt, acceleration, angular acceleration, etc. of the driver's seat. In addition, it is also found that when deciding the next operation, the operator pays attention to the rotation amount and sound of the engine as the driving source.

[0018] Here, the digging status is an example of information indicating the working status. The operating position is an example of information indicating the operating state of the hydraulic excavator. The tilt of the driver's seat is an example of a reaction from the ground. The sound of the engine is information indicating the state of the driving source and, therefore, the operating state of the hydraulic excavator.

[0019] Therefore, the inventors of the present invention have devised a method for using at least data indicating the working status, data indicating the operating state of the motion unit, and data indicating the reaction of the motion unit to the work object due to the work of the working unit as input data (prediction basis data) for learning. The present invention is based on this insight.

[0020] In order to achieve the above-mentioned purpose, a construction machine with a learning function involved in a certain aspect of the present invention comprises: an action unit, which has an operating unit, so that the above-mentioned operating unit operates in a manner of performing an operation; an operating unit, which outputs an instruction corresponding to the operator's operation; a working condition detection unit, which detects the condition of the above-mentioned operation performed by the above-mentioned operating unit and outputs the detected working condition as working condition data; an action state detection unit, which detects the action state of the above-mentioned action unit and outputs the detected action state as action state data; a reaction detection unit, which detects the reaction received by the above-mentioned action unit from the operation object due to the operation of the above-mentioned operating unit, and outputs the detected reaction as a reaction Action data output; a learning data storage unit that stores the above-mentioned instructions as instruction data in a time series, and stores prediction base data including the above-mentioned working condition data, the above-mentioned action state data, and the above-mentioned reaction data in a time series; a learning unit that uses the above-mentioned prediction base data stored in the above-mentioned learning data storage unit to perform machine learning on the above-mentioned instruction data stored in the above-mentioned learning data storage unit, and after completing the machine learning, receives input of the above-mentioned prediction base data when the above-mentioned action unit is in motion and outputs a predicted instruction for the above-mentioned instruction; and a hydraulic drive system that drives the above-mentioned action unit based on the above-mentioned instruction, the above-mentioned predicted instruction, or the above-mentioned instruction and the above-mentioned predicted instruction. Here, "instruction data" refers to data used as teacher data in the case of "teacher-based learning". "Prediction base data" refers to data input into the learning unit during machine learning so that the learning unit can predict the action instruction. The so-called "causing the working unit to operate..." means "causing the working unit to operate and move."

[0021] According to this structure, during learning, a skilled operator operates a construction machine to perform a prescribed operation. Then, instruction data corresponding to the instruction corresponding to the operation, and prediction basic data including data based on the working conditions performed by the working unit, data on the operating state of the action unit, and data on the reaction received by the action unit from the working object due to the operation of the working unit are stored in the learning data storage unit. Moreover, by inputting the prediction basic data stored in the learning data storage unit, the learning unit performs mechanical learning on the learning data to output a prediction instruction. In this way, the operation of the skilled operator can be learned and output as a prediction instruction. Moreover, after the mechanical learning is completed, if the prediction basic data is input to the learning unit, the learning unit outputs a prediction action instruction. Then, the hydraulic drive system drives the action unit based on the prediction instruction. As a result, the operation of the skilled operator is learned, and the operation is automatically performed based on the prediction instruction output from the learning unit as a result of the learning.

[0022] Therefore, it is possible to provide a construction machine that can learn the work performed by a person operating the construction machine and automatically perform the learned work.

[0023] It can also be that the above-mentioned operating unit is configured to output an action instruction corresponding to the operation of the above-mentioned operator as the above-mentioned instruction, the learning data storage unit is configured to store the above-mentioned action instruction as instruction data in a time series, and store the prediction basic data including the above-mentioned working condition data, the above-mentioned action state data and the above-mentioned reaction data in a time series, the above-mentioned learning unit is configured to use the above-mentioned prediction basic data stored in the above-mentioned learning data storage unit to mechanically learn the above-mentioned instruction data stored in the above-mentioned learning data storage unit during learning, and when the automatic control after the above-mentioned mechanical learning is completed, the input of the above-mentioned prediction basic data is received, and the predicted action instruction as the above-mentioned prediction instruction is output, and the hydraulic drive system is configured to drive the above-mentioned action unit according to the above-mentioned action instruction or the above-mentioned predicted action instruction.

[0024] According to this configuration, the hydraulic drive system is configured to drive the motion unit according to the predicted motion command, that is, the predicted motion command. Therefore, the work is automatically performed according to the learning result of the operation by the skilled operator.

[0025] Alternatively, the main body may be provided with the action unit, and the reaction detection unit may detect the reaction including at least one of the inclination, acceleration, and angular acceleration of the action unit or the main body, and output the detected reaction as the reaction data.

[0026] With this structure, the inclination, acceleration, or angular acceleration of the action unit or the main body, which a skilled operator prioritizes when deciding the next operation, is included in the prediction base data, thereby improving learning accuracy. Furthermore, since the action unit is located within the main body, any reaction to the work object affects the main body.

[0027] It can also be that the above-mentioned action state detection unit includes a drive source state detection unit, which detects the state of the drive source including the output of the drive source of the pump that drives the working oil of the above-mentioned hydraulic drive system and at least any one of the action sounds, and outputs the detected state of the drive source as drive source state data, and the above-mentioned action state data includes the above-mentioned drive source state data.

[0028] According to this configuration, the output or sound of the driving source that the skilled operator prioritizes when deciding the next operation is included in the prediction basic data, thereby improving the accuracy of learning.

[0029] Alternatively, the action state data may include the instruction data. Here, "action state data includes instruction data" means that, in the case of "teacher-based learning," instruction data stored in the learning data storage unit in a time series is used as teacher data, and instruction data at a time "earlier in the time series" than the instruction data used as teacher data is used as action state data.

[0030] According to this configuration, the motion command corresponding to the current position of the operating lever (operating portion), which is important to the skilled operator when deciding the next operation, is included in the prediction basic data, thereby improving the accuracy of learning.

[0031] The operation state detection unit may further include a posture detection unit that detects a posture of the operation unit and outputs the detected posture as posture data, and the operation state data may include the posture data.

[0032] According to this configuration, data on the posture of the operating portion that is important to the skilled operator when deciding the next operation is included in the prediction basic data, thereby improving the accuracy of learning.

[0033] Alternatively, the construction machinery with a learning function may be a skill-transferring construction machinery having a control unit, wherein the operating unit is configured to output, as the instruction, a manual action correction instruction corresponding to the operator's operation, the hydraulic drive system is configured to drive the action unit according to the basic action instruction, the automatic action correction instruction, and the manual action correction instruction, and the control unit includes: a basic action instruction unit that outputs the basic action instruction for causing the working unit to perform a basic action through the action unit; an action correction instruction generating unit that adds the manual action correction instruction to the automatic action correction instruction to generate an action correction instruction; the learning data storage unit includes an action correction instruction storage unit that stores the action correction instructions as the instruction data in a time series, and a prediction basic data storage unit that stores the prediction basic data in a time series; and the learning unit, wherein the learning unit is configured to perform machine learning on the action correction instructions stored in the action correction instruction storage unit using the prediction basic data stored in the prediction basic data storage unit, and after completing the machine learning, receives the input of the prediction basic data when the action unit is in action, and outputs the automatic action correction instruction as the prediction instruction.

[0034] According to this structure, the action unit operates the working unit via the hydraulic drive system according to basic action instructions, automatic action correction instructions, and manual action correction instructions. Therefore, when the operator does not correct the action unit's actions and the learning unit does not perform automatic action corrections, the action unit causes the working unit to perform basic actions according to the basic action instructions output by the basic action instruction unit. The operator monitors the action of the working unit while visually confirming the work being performed by the working unit. If the operator cannot perform the prescribed work with familiar movements using basic actions, the operator manually corrects the work by performing the prescribed work with familiar movements. Then, the operating unit outputs a manual action correction instruction corresponding to the manual correction, correcting the basic action, thereby performing the prescribed work with familiar movements.

[0035] On the other hand, a manual operation correction command related to the predetermined task is added to the automatic operation correction command output by the learning unit to generate an operation correction command, and the operation correction command is machine-learned by the learning unit.

[0036] When the learning unit does not perform automatic corrections as described above, it only learns manual motion correction instructions based on the operator's manual motion corrections. Since the learning unit receives input of predicted basic data corresponding to the motion of the working unit during the motion of the motion unit, when a similar motion state occurs in which the prescribed work cannot be performed with skilled motions, the learning unit outputs motion correction instructions predicted by the learning unit as automatic motion correction instructions. This automatically corrects the basic motion instructions toward performing the prescribed work with skilled motions. If the automatic correction is appropriate, the prescribed work is performed with skilled motions.

[0037] However, if learning is insufficient, or if the motion state of the motion unit differs significantly from the state predicted by learning when the prescribed task is not performed with the skilled motion, even if this correction is performed, the prescribed task will not be performed with the skilled motion. Therefore, the operator manually corrects the motion so that the prescribed task can be performed with the skilled motion, thereby allowing the motion unit to perform the prescribed task with the skilled motion. Furthermore, a manual motion correction instruction corresponding to this additional manual motion correction is added to the automatic motion correction instruction corresponding to the previous manual motion correction, and the learning unit learns the result.

[0038] This improves the learning unit's ability to correct the basic movements of the working unit. By repeating these steps, if the learning unit's ability to correct the basic movements of the working unit reaches the same level as the operator, the operator's correction of the basic movements of the working unit becomes unnecessary. In this state, the learning unit appropriately corrects the basic movements of the working unit on behalf of the operator, enabling the working unit to properly perform the prescribed work.

[0039] In this way, when the operator is an experienced person, the operator's manual motion corrections constitute the experienced person's "skills." These "skills" are accumulated in the learning unit and passed on to the learning unit, making the learning unit the "transmitter" of the experienced person's "skills." As a result, a construction machine equipped with this learning unit becomes a "skill-transmitting construction machine."

[0040] Furthermore, according to the above-described structure, the operating unit is configured to operate according to basic motion instructions, automatic motion correction instructions, and manual motion correction instructions via the hydraulic drive system. Therefore, even if the learning unit outputs an insufficient automatic motion correction instruction, the operator can manually correct the motion while observing the operation of the working unit to cause the operating unit to perform appropriate operations. Therefore, appropriate operations can be tested and corrected based on actual on-site operations. In other words, the learning unit can learn based on actual on-site operations, eliminating the need for extensive learning data and a debugging period for the learning unit. As a result, automation of specified operations can be achieved in a short period of time.

[0041] Furthermore, according to the above configuration, the basic motion instruction unit automatically executes the portions of the basic motion associated with a predetermined task that do not require correction. Therefore, the operator only needs to make the necessary corrections. This reduces the burden on the operator. Furthermore, even skilled operators experience variations in their work. Therefore, if the operator only performs a portion of the work, the accuracy of the work is improved compared to performing the entire work by the operator.

[0042] In addition, although it is conceivable to pass on the skills of an experienced person by storing manual action correction instructions corresponding to the operator's manual action correction in the storage unit, since the forms of the basic action of the working unit that require correction are infinite, it is difficult to pass on the skills of an experienced person by this method in reality. On the other hand, if a learning unit is used as described above, each time a situation occurs in which the basic action of the working unit requires correction, the learning unit learns the manual action correction corresponding to that form (more precisely, the manual action correction instruction), thereby making it easy to pass on the skills of an experienced person.

[0043] Furthermore, according to the above configuration, during the operation of the motion unit, a skilled operator operates the construction machine, performing the prescribed operation while correcting the motion of the working unit according to the basic motion instructions and automatic motion correction instructions as needed. Prediction base data, including data based on the working conditions of the working unit, data on the motion unit's operating status, and data on the reaction of the working unit to the work object due to the working unit's operation, is stored in the prediction base data storage unit. Furthermore, the motion correction instructions, obtained by adding the manual motion correction instructions to the automatic motion correction instructions, are stored in the motion correction instruction storage unit. Furthermore, during subsequent learning, the learning unit uses the prediction base data stored in the prediction base data storage unit to machine-learn the motion correction instructions stored in the motion correction instruction storage unit. This allows the skilled operator's correction operations to be learned and output as automatic motion correction instructions. Furthermore, during subsequent operation of the motion unit, if the prediction base data is input to the learning unit, the learning unit outputs the automatic motion correction instructions. The hydraulic drive system then drives the motion unit while reflecting the basic motion instructions and the automatic motion correction instructions. As a result, the operation is performed according to the automatic motion correction instructions, which are the result of learning the skilled operator's correction operations. Therefore, it is possible to provide a construction machine that allows an operator to perform a correction operation on the basic operation of the working unit based on the basic operation instruction unit, thereby learning the work performed by the construction machine and automatically performing the learned work.

[0044] As a result, it is possible to provide a skill-transferring construction machine that can pass on the skills of skilled workers in the construction industry and can automate predetermined operations in a short period of time.

[0045] It can also be that the above-mentioned manual action correction instruction is an electrical instruction signal, and the above-mentioned action part includes: a hydraulic actuator, which drives the above-mentioned working part; and a control valve, which hydraulically controls the action of the above-mentioned hydraulic actuator according to the above-mentioned basic action instruction, the above-mentioned automatic action correction instruction and the above-mentioned manual action correction instruction, and the above-mentioned control valve is a solenoid valve.

[0046] According to this configuration, since the manual action correction command is an electrical command signal, it can be directly converted into numerical data. Therefore, the basic action command, the automatic action correction command, and the manual action correction command can be added together, the addition of the automatic action correction command and the manual action correction command, and the storage of the action correction command in the storage unit can be easily performed. This simplifies the configuration related to these processes compared to a case where the manual action correction command is a hydraulic signal and the control valve is a hydraulic control valve.

[0047] The present invention has the effect of providing a construction machine that can learn the work performed by a person operating the construction machine and automatically perform the learned work.

[0048] Furthermore, a specific aspect of the present invention has the effect of providing a skill-transferring construction machine that can pass on the skills of skilled workers in the construction industry and can automate predetermined operations in a short period of time. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 This is a functional block diagram showing the concept of a construction machine with a learning function according to an embodiment of the present invention.

[0050] Figure 2 This is a functional block diagram showing the configuration of a control system for a construction machine with a learning function according to the first embodiment of the present invention.

[0051] Figure 3 It is a side view showing the hardware configuration of a hydraulic excavator with a learning function, which is an example of a construction machine with a learning function.

[0052] Figure 4 Yes Figure 3 Hydraulic circuit diagram of the main hydraulic circuit of the hydraulic drive system of a hydraulic excavator with a learning function.

[0053] Figure 5 Yes Figure 3 A hydraulic circuit diagram of the operating system hydraulic circuit of the hydraulic drive system of a hydraulic excavator with a learning function.

[0054] Figure 6 Yes Figure 3 Functional block diagram of the structure of a control system for a hydraulic excavator with a learning function.

[0055] Figure 7 Yes Figure 3 Schematic diagram of predicted action instructions, learning instruction data, prediction basic data, and time series data of the learning prediction basic data of a hydraulic excavator with a learning function.

[0056] Figure 8 Yes Figure 6 A functional block diagram of the structure of the learning unit.

[0057] Figure 9 Is based on Figure 3 Schematic diagram of the construction work performed by a hydraulic excavator with learning function.

[0058] Figure 10 It is a side view showing the hardware configuration of a hydraulic excavator with a learning function as an example of a construction machine with a learning function according to a second embodiment of the present invention.

[0059] Figure 11 Yes Figure 10 Functional block diagram of the structure of a control system for a hydraulic excavator with a learning function.

[0060] Figure 12 It is a side view showing the hardware configuration of a hydraulic excavator with a learning function as an example of a construction machine with a learning function according to a third embodiment of the present invention.

[0061] Figure 13 Yes Figure 12 Functional block diagram of the structure of a control system for a hydraulic excavator with a learning function.

[0062] Figure 14 This is a functional block diagram showing the configuration of a control system for a skill-transferring construction machine according to a fourth embodiment of the present invention.

[0063] Figure 15 It is a side view showing the hardware structure of a skill-transfer hydraulic excavator as an example of a skill-transfer construction machine.

[0064] Figure 16 Yes Figure 15 The hydraulic circuit diagram of the main hydraulic circuit of the hydraulic drive system of the hydraulic excavator.

[0065] Figure 17 Yes Figure 15 The hydraulic circuit diagram of the hydraulic circuit of the operating system of the hydraulic drive system of the hydraulic excavator.

[0066] Figure 18 Yes Figure 15A functional block diagram of the structure of the control system of a skill inheritance hydraulic excavator.

[0067] Figure 19 Yes Figure 15 A schematic diagram illustrating the cycle time in the operation of a hydraulic excavator.

[0068] Figure 20 Yes Figure 15 Schematic diagram of the time series data of the hydraulic excavator's motion correction instructions, learning motion correction instructions, prediction basic data, and learning prediction basic data.

[0069] Figure 21 Yes Figure 20 A functional block diagram of the structure of the learning unit.

[0070] Figure 22 Is based on Figure 15 Schematic diagram of construction work performed by a hydraulic excavator.

[0071] Figure 23 (a) to (c) are cross-sectional views schematically showing a process in which the excavation work performed by the hydraulic excavator 20 is improved by a correction operation of the deep excavation work at the corner.

[0072] Figure 24 (a) to (c) are cross-sectional views schematically showing a process in which the excavation work performed by the hydraulic excavator 20 is improved by a correction operation corresponding to the geology of the planned excavation site.

[0073] Figure 25 It is a side view showing the hardware configuration of a skill-transfer hydraulic excavator as an example of a skill-transfer construction machine according to a fifth embodiment of the present invention.

[0074] Figure 26 Yes Figure 25 A functional block diagram of the structure of the control system of a skill inheritance hydraulic excavator.

[0075] Figure 27 It is a side view showing the hardware configuration of a skill-transfer hydraulic excavator as an example of a skill-transfer construction machine according to a sixth embodiment of the present invention.

[0076] Figure 28 Yes Figure 27 A functional block diagram of the structure of the control system of a skill inheritance hydraulic excavator.

[0077] Figure 29 This is a functional block diagram showing the configuration of an operation command generation unit of a skill-transfer hydraulic excavator as an example of a skill-transfer construction machine according to a seventh embodiment of the present invention. DETAILED DESCRIPTION

[0078] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. In addition, in all the drawings, identical or corresponding elements are denoted by the same reference numerals, and their repeated descriptions are omitted. Furthermore, the present invention is not limited to the following embodiments.

[0079] (concept)

[0080] Figure 1 This is a functional block diagram showing the concept of a construction machine with a learning function according to an embodiment of the present invention.

[0081] Reference Figure 1 The construction machine 1000 with a learning function according to an embodiment of the present invention comprises: an action unit 103 having a working unit 104 for operating the working unit 104 in a manner of performing work; an operation unit 101 for outputting an instruction 201 corresponding to an operation by an operator; a work status detection unit 112 for detecting a status of work performed by the working unit 104 and outputting the detected work status as work status data 212; an action status detection unit 113 for detecting an action status of the action unit 103 and outputting the detected action status as action status data 213; a reaction detection unit 114 for detecting a reaction received by the action unit 103 from a work object due to the work of the working unit 104 and outputting the detected reaction as reaction data 214. Output; a learning data storage unit 115, which stores the instruction 201 as instruction data 211 in a time series, and stores the prediction basic data Pd including the working status data 212, the action state data 213 and the reaction data 214 in a time series; a learning unit 118, which uses the prediction basic data Pd stored in the learning data storage unit 115 to perform mechanical learning on the instruction data 211 stored in the learning data storage unit 115, and after completing the mechanical learning, receives the input of the prediction basic data Pd when the action unit 103 is in action, and outputs the prediction instruction 1103 of the instruction 201; and a hydraulic drive system 105, which drives the action unit 103 based on the instruction 201, the prediction instruction 1103, or the instruction 201 and the prediction instruction 1103.

[0082] Reference numeral 102 denotes a main body of the construction machine with a learning function 1000 . An operating unit 103 and a hydraulic drive system 105 are provided in the main body 102 .

[0083] The command data generating unit 1101 generates command data 211 based on the command 201. Command data 211 can be generated by, for example, converting the hydraulic pressure command output from the operating unit 101 into command data (embodiments 1 to 3) or adding the predicted command output from the learning unit 118 to the command output from the operating unit 101 (embodiments 4 to 7).

[0084] The learning data storage unit 115 includes an instruction data storage unit 1102 and a prediction base data storage unit 117. The instruction data storage unit 1102 stores instruction data 211 during learning or operation, and outputs the instruction data 211' to the learning unit 118 during learning. The prediction base data storage unit 117 stores prediction base data Pd (212-214) during learning or operation, and outputs the prediction base data Pd' to the learning unit 118 during learning.

[0085] According to the construction machine 1000 with learning functionality, during learning (the previous operation in Embodiments 4 to 7), a skilled operator operates the construction machine to perform a predetermined task. Command data 211 corresponding to the command corresponding to the operation and prediction base data Pd are stored in the learning data storage unit 115. The prediction base data Pd includes data 212 based on the status of the work performed by the working unit 104, data 213 on the operational state of the motion unit 103, and data 214 on the reaction to the work object experienced by the motion unit 103 due to the operation of the working unit 104. Furthermore, by inputting the prediction base data Pd' stored in the learning data storage unit 115, the learning unit 118 performs machine learning on the learning data 211' and Pd' and outputs a prediction command 1103. This allows the skilled operator's operation to be learned and output as the prediction command 1103. After the machine learning is completed, if the prediction base data Pd is input to the learning unit 118, the learning unit 118 outputs the prediction command 1103. Then, the hydraulic drive system 105 drives the motion unit 103 based on the prediction command 1103. As a result, the operation of the skilled operator is learned, and the work is automatically performed based on the prediction command 1103 outputted from the learning unit 118 as a result of the learning.

[0086] Therefore, it is possible to provide the construction machine 1000 that can learn the work performed by a person operating the construction machine and automatically perform the learned work.

[0087] The following describes implementations of the concept of a learning-enabled construction machine 1000. These implementations are broadly divided into implementations 1 through 3, which relate to learning-enabled construction machines, and implementations 4 through 7, which relate to skill-transfer construction machines. These skills-transfer construction machines are learning-enabled construction machines that can inherit the skills of experienced operators in the construction industry and automate specific tasks in a short period of time.

[0088] (Implementation 1)

[0089] {summary}

[0090] First, an overview of the construction machine with a learning function according to the first embodiment will be described.

[0091] [structure]

[0092] Figure 2 This is a functional block diagram showing the structure of a control system for a construction machine with a learning function according to the first embodiment of the present invention. Figure 2 In the figure, arrows represent the flow of instructions, power, information, and data. Solid arrows represent the flow of instructions or data during automatic control, and dashed arrows represent the flow of instructions or data during learning. In addition, as will be described later, when past prediction basic data is used in learning, the flow of data represented by dashed lines occurs even during automatic control. Figure 6 、 11 , the same goes for 13.

[0093] Reference Figure 2 The construction machine 100 with a learning function in Embodiment 1 includes an operating unit 101, a main body 102, an operating unit 103, a hydraulic drive system 105, an operating command detector 110, a working condition detector 112, an operating state detector 113, a reaction detector 114, a learning data storage unit 115, a learning unit 118, and an operating unit driver 119. Furthermore, the construction machine 100 with a learning function includes an overall control unit and an operating mode switching operation unit (not shown). The overall control unit switches the operating mode of the construction machine 100 with a learning function between a learning mode and an automatic control mode based on an operator's operation of the operating mode switching operation unit. Hereinafter, the period when the construction machine 100 with a learning function operates in the learning mode is referred to as "learning mode," and the period when the construction machine 100 with a learning function operates in the automatic control mode is referred to as "automatic control mode."

[0094] In the first embodiment, the operating unit 101 is configured to output an action command 201 corresponding to an operator's operation as a command. The learning data storage unit 115 is configured to store the action command 201 in a time series as command data 211 and to store prediction base data Pd, including work status data 212, action state data 213, and reaction data 214, in a time series. The learning unit 118 is configured to perform machine learning on the command data 211' stored in the learning data storage unit 115 using the prediction base data Pd' stored in the learning data storage unit 115 during learning. After the machine learning is completed, the learning unit 118 receives input of the prediction base data Pd and outputs a predicted action command Pf as a predicted command 1103 during automatic control. The hydraulic drive system 105 is configured to drive the operating unit 103 according to the action command 201 or the predicted action command Pf. Furthermore, the action command 201 is converted into command data 211 by the action command detection unit 110, which serves as the command data generation unit 1101.

[0095] Hereinafter, the structure of the construction machine 100 with a learning function will be described in detail.

[0096] The operation unit 101 outputs an operation command 201 corresponding to an operation by an operator.

[0097] The operating unit 103 is connected to the main body 102 .

[0098] The operation unit 103 includes a working unit 104 for performing work, and operates the working unit 104 to perform work. Here, "operating the working unit 104" means "operating and moving the working unit 104".

[0099] "Construction machinery" refers to any work machine that can perform construction work by operating a working part through an action part according to an operator's operation. Examples of "construction machinery" include hydraulic excavators, bulldozers, tractor-type forklifts, wheel loaders, trenchers, excavators, cranes, and lift trucks.

[0100] The hydraulic drive system 105 is provided across the main body 102 and the operating unit 103. The hydraulic drive system 105 outputs a driving force 202 according to an operating instruction 201 output from the operating unit 101 or a predicted operating instruction Pf output from the learning unit 118, thereby driving the operating unit 103.

[0101] The action instruction detection unit 110 detects the action instruction 201 output from the operating unit 101 and outputs it as instruction data 211. Specifically, when the action instruction 201 is a hydraulic instruction (pilot pressure instruction), it is converted into instruction data 211 as electrical data and output. Therefore, the action instruction detection unit 110 is not an essential element. When the action instruction 201 is an electrical instruction, the action instruction detection unit 110 can also be omitted, and the action instruction 201 can be directly input into the instruction data storage unit 116 and the learning unit 118 described later. In the hydraulic excavator 10 with a learning function described later, the acceleration instruction (electrical instruction) output by the acceleration device 50, that is, the action instruction 201, is directly input into the instruction data storage unit 116 and the learning unit 118 as the instruction data 211.

[0102] The work status detection unit 112 detects the status of work performed by the working unit 104 and outputs the detected work status as work status data 212 .

[0103] The operation state detection unit 113 detects the operation state of the operation unit 103 and outputs the detected operation state as operation state data 213 .

[0104] The prediction base data includes work status data 212, motion state data 213, and reaction data 214. For ease of explanation, the prediction base data used during learning is labeled Pd', while the prediction base data used during automatic control is labeled Pd. The prediction base data may also include command data 211.

[0105] The reaction detection unit 114 detects a reaction received by the operation unit 103 or the main body 102 from the work object due to the work of the working unit 104 , and outputs the detected reaction as reaction data 214 .

[0106] The learning data storage unit 115 includes an instruction data storage unit 116 and a prediction basis data storage unit 117 .

[0107] The command data storage unit 116 stores the command data 211 in time series as the command storage unit 1102. The prediction base data storage unit 117 stores the prediction base data Pd including the command data 211, work status data 212, operation state data 213, and reaction data 214 in time series.

[0108] The learning unit 118 is a learning model that performs machine learning. Examples of such learning models include neural networks, regression models, tree models, Bayesian models, time series models, clustering models, and ensemble learning models. In this embodiment, the learning model is a neural network. The learning method can be either supervised learning or unsupervised learning. Deep learning is also possible.

[0109] For example, in the case of teacher-based learning, the learning unit 118 reads the instruction data stored in the instruction data storage unit 116 as instruction data for learning Pf', and reads the prediction basic data stored in the prediction basic data storage unit 117 as prediction basic data for learning Pd'. Then, the instruction data for learning Pf' is used as teacher data, and the prediction basic data for learning Pd' is used as input data to create learning data. Then, the prediction basic data Pd' as input data is input into a machine learning model (e.g., a neural network), and the difference between the output and the teacher data Pf' is evaluated, and the evaluation is fed back to the machine learning model. Thus, the machine learning model performs machine learning on the learning data. When the machine learning is completed, the learning unit 118 outputs the output of the machine learning model as a predicted action instruction Pf to the outside. Specifically, during automatic control, in the learning unit 118, if the prediction basic data Pd is input, the machine learning model outputs the predicted action instruction Pf.

[0110] During automatic control, the operating unit driver 119 operates the operating unit 101 according to the predicted motion command Pf output from the learning unit 118. Consequently, the operating unit 101 outputs a motion command 201, and the hydraulic drive system 105 drives the operating unit 103 according to this motion command 201. Specifically, the operating unit driver 119 and the operating unit 101 function as a predicted motion command converter 120 that converts the predicted motion command Pf into the motion command 201. The operating unit driver 119 is comprised of, for example, a motor, a manipulator (robot), or the like.

[0111] Therefore, the operating unit drive unit 119 is not an essential element. For example, a predicted action instruction conversion unit that converts the predicted action instruction Pf, which is an electrical instruction signal, into a hydraulic instruction (pilot pressure instruction) may be provided instead of the operating unit drive unit 119. During automatic control, the output of the predicted action instruction conversion unit is input to the hydraulic drive system 105 instead of the action instruction from the operating unit 101. The predicted action instruction conversion unit can be composed of, for example, a pilot valve (electromagnetic proportional valve) composed of an electromagnetic valve. In addition, in the case where the construction machine 100 with a learning function is configured to output the action instruction 201 of the electrical instruction from the operating unit 101, the predicted action instruction conversion unit 120 is not required, and the predicted action instruction Pf output from the learning unit 118 is directly input to the hydraulic drive system 105.

[0112] In this manner, the hydraulic drive system 105 drives the operating unit 103 according to the predicted operation command Pf converted into an operation command by the operating unit driving unit 119 and the operating unit 101 .

[0113] [action]

[0114] During learning, a skilled operator operates a construction machine 100 equipped with a learning function and performs a predetermined task. Command data 211 corresponding to the motion command 201 corresponding to the operation and prediction base data are then stored in the learning data storage unit 115. The prediction base data includes operation status data 212 indicating the status of the operation performed by the working unit 104, motion status data 213 indicating the motion status of the motion unit, and reaction data 214 indicating the reaction to the work object received by the motion unit 103 or the main unit 102 due to the operation of the working unit 104. The predetermined task is, for example, a relatively simple, standardized task. Examples of such standardized tasks include digging, ground leveling, and compaction.

[0115] During learning, the learning unit 118 reads the instruction data stored in the instruction data storage unit 116 as instruction data for learning Pf', and reads the prediction base data stored in the prediction base data storage unit 117 as prediction base data for learning Pd'. Learning data is created using the learning instruction data Pf' as the training data and the prediction base data Pd' as the input data. The learning unit 118 then performs machine learning on this learning data. After completing machine learning, the learning unit 118, when receiving input of the prediction base data Pd, outputs a predicted action command Pf during automatic control. The hydraulic drive system 105 then drives the motion unit 103 according to the predicted action command Pf, which has been converted into an action command 201 by the operating unit drive unit 119 and the operating unit 101. As a result, the system learns the operations of a skilled operator and automatically performs the work according to the predicted action command Pf output from the learning unit 118 as a result of the learning.

[0116] As described above, according to the first embodiment, the hydraulic drive system 105 is configured to drive the motion unit 103 according to the predicted motion command Pf, which is a predicted command of the motion command 201 . Therefore, the work is automatically performed according to the learning result of the operation by the skilled operator.

[0117] {Specific structure}

[0118] Next, a specific configuration of a construction machine 100 with a learning function will be described using a hydraulic excavator 10 as an example of a construction machine.

[0119] [Hardware Structure]

[0120] Overall Structure

[0121] First, the overall structure of the hydraulic excavator 10 with a learning function will be described.

[0122] Figure 3 1 is a side view showing the hardware configuration of a hydraulic excavator 10 with a learning function, which is an example of a construction machine with a learning function.

[0123] A hydraulic excavator with a learning function (hereinafter sometimes simply referred to as a hydraulic excavator) 10 includes a main body 102. A traveling body 19 is provided on the main body 102. The traveling body 19 is constituted by a vehicle travel device including crawler tracks (caterpillars), for example.

[0124] A rotating body 15 is provided on the main body 102 so as to be rotatable about a vertical first rotation axis A1. A driver's seat (not shown) is provided on the rotating body 15, and an operating unit 101 (see FIG. Figure 6 ). In addition, although Figure 6Although the operating unit 101 is not shown, a travel operating device for operating the travel body 19 is provided at the driver's seat. A rotation motor 14 is provided on the rotating body 15 to rotate the rotating body 15. The rotation motor 14 is composed of a hydraulic motor. In addition, an engine 26 for travel is provided on the rotating body 15 (see Figure 6 The engine 26 drives the pump unit 107 of the hydraulic drive system 1 during operation (see Figure 6 ).

[0125] The base end of a boom 16 is rotatably connected to the rotating body 15 about a horizontal second rotation axis A2. The distal end and base end of a boom cylinder 11 are rotatably connected to the base end of the boom 16 and the rotating body 15, respectively. The boom 16 swings about the second rotation axis A2 as the boom cylinder 11 extends and contracts.

[0126] The base end of the boom 17 is connected to the distal end of the boom 16 so as to be rotatable about a third horizontal rotation axis A3. The distal end and base end of the arm cylinder 12 are connected to the base end of the arm 17 and the center of the boom 16 so as to be rotatable, respectively. The arm 17 swings around the third rotation axis A3 as the center through the extension and contraction of the arm cylinder 12.

[0127] The base end of a bucket 18 is connected to the distal end of the arm 17 so as to be rotatable about a fourth horizontal rotation axis A4. The distal end and base end of the bucket cylinder 13 are rotatably connected to the base end of the bucket 18 and the center of the arm 17, respectively. Bucket 18 rotates about the fourth rotation axis A4 by extending or retracting the bucket cylinder 13. Bucket 18 is an example of an attachment, and other attachments may also be attached.

[0128] The boom 16 , the arm 17 , and the bucket 18 constitute a front working mechanism. The bucket 18 constitutes a working unit 104 , and the swing structure 15 and the front working mechanism (the boom 16 , the arm 17 , and the bucket 18 ) constitute an operating unit 103 .

[0129] The hydraulic excavator 10 includes a pair of left and right hydraulic travel motors (not shown) in addition to the above-mentioned components.

[0130] The operator operates the operating unit 101 (including a travel operating device not shown) to position the hydraulic excavator at a desired location, rotate the swing body 15 , change the posture of the booms 16 and 17 , and swing the bucket 18 to perform desired work.

[0131] The hydraulic excavator 10 with a learning function further includes a first camera 311. The first camera 311 captures images of the working conditions of the bucket 18. The images captured by the first camera 311 are processed by an image processing unit 312 (see FIG. Figure 6) is subjected to image processing to obtain data representing the working status, and is output from the image processing unit 312 as the working status data 212. The optical axis 321 of the first imaging device 311 is directed toward the working object.

[0132] The first camera 311 is composed of, for example, a 3D camera, a camera with a depth sensor, etc. The first camera 311 is fixed to the main body 102 via an appropriate support member, or fixed to a fixed object (such as the ground) other than the vehicle of the hydraulic excavator 10 via an appropriate support member, or mounted on a drone.

[0133] Hydraulic drive system 1

[0134] Next, the hydraulic drive system 1 for operating the hydraulic excavator 10 will be described.

[0135] Figure 4 1 is a hydraulic circuit diagram showing a main hydraulic circuit of the hydraulic drive system 1 of the hydraulic excavator 10 with a learning function. Figure 5 1 is a hydraulic circuit diagram showing an operating system hydraulic circuit of a drive system of the hydraulic excavator with a learning function 10. The main hydraulic circuit and the operating system hydraulic circuit are provided on the revolving structure 15.

[0136] As described above, the hydraulic drive system 1 includes the boom cylinder 11 , the arm cylinder 12 , and the bucket cylinder 13 as hydraulic actuators, and also includes the swing motor 14 and a pair of left and right hydraulic travel motors (not shown).

[0137] Reference Figure 4 The hydraulic drive system 1 includes a first main pump 21 and a second main pump 23 for supplying hydraulic oil to the above-mentioned actuators. Figure 4 In order to simplify the drawing, actuators other than the rotary motor 14 are omitted.

[0138] The first main pump 21 and the second main pump 23 are driven by the engine 26. The engine 26 also drives the auxiliary pump 25. The first main pump 21, the second main pump 23 and the auxiliary pump 25 constitute the pump unit 107 (see Figure 6 The engine 26 is driven by the accelerator 50 (refer to Figure 6 ) adjusts the output. The accelerator 50 includes, for example, an accelerator pedal, and outputs an acceleration command, an electrical command corresponding to the amount of depression, to an engine control unit (not shown). The engine control unit controls the output (rotational speed) of the engine 26 according to the acceleration command.

[0139] The first main pump 21 and the second main pump 23 are, for example, variable displacement pumps that discharge hydraulic oil at a flow rate corresponding to the tilting angle. Here, the first main pump 21 and the second main pump 23 are swash plate pumps whose tilting angle is determined by the angle of the swash plate. However, the first main pump 21 and the second main pump 23 may also be swash axis pumps whose tilting angle is determined by the angle between the drive shaft and the cylinder block.

[0140] The discharge flow rate Q1 of the first main pump 21 and the discharge flow rate Q2 of the second main pump 23 are controlled by positive control. Specifically, the tilting angle of the first main pump 21 is adjusted by the first flow adjustment device 22, and the tilting angle of the second main pump 23 is adjusted by the second flow adjustment device 24. The auxiliary pump 25 is connected to the first flow adjustment device 22 and the second flow adjustment device 24 through the sub-discharge pipeline 37. The auxiliary pump 25 functions as a driving source for the first flow adjustment device 22 and the second flow adjustment device 24. The first flow adjustment device 22 and the second flow adjustment device 24 will be described in detail later.

[0141] The first center discharge line 31 extends from the first main pump 21 to the fuel tank. Multiple control valves, including a first boom control valve 41 and a rotation control valve 43, are located on the first center discharge line 31 (other than the first boom control valve 41 and the rotation control valve 43, not shown). Each control valve is connected to the first main pump 21 via a pump line 32. In other words, the control valves on the first center discharge line 31 are connected in parallel with the first main pump 21. Furthermore, each control valve is connected to the fuel tank via a tank line 33.

[0142] Similarly, a second center discharge line 34 extends from the second main pump 23 to the fuel tank. Multiple control valves, including a second arm control valve 42 and a bucket control valve 44 (other than the second arm control valve 42 and bucket control valve 44, not shown), are located on the second center discharge line 34. Each control valve is connected to the second main pump 23 via a pump line 35. In other words, the control valves on the second center discharge line 34 are connected in parallel to the second main pump 23. Furthermore, each control valve is connected to the fuel tank via a tank line 36.

[0143] The first arm control valve 41 and the second arm control valve 42 together control the supply and discharge of hydraulic oil to the arm cylinder 12. That is, hydraulic oil is supplied from the first main pump 21 to the arm cylinder 12 via the first arm control valve 41, and hydraulic oil is supplied from the second main pump 23 to the arm cylinder 12 via the second arm control valve 42. The first arm control valve 41 and the second arm control valve 42 constitute the arm control valve 40 (see Figure 5 ).

[0144] The rotation control valve 43 controls the supply and discharge of working oil relative to the rotation motor 14. That is, the working oil is supplied from the first main pump 21 to the rotation motor 14 via the rotation control valve 43. Specifically, the rotation motor 14 is connected to the rotation control valve 43 through a pair of supply and discharge pipes 61 and 62. A release pipe 63 branches out from each supply and discharge pipe 61 and 62, and the release pipe 63 is connected to the oil tank. A pressure relief valve 64 is provided in each release pipe 63. In addition, the supply and discharge pipes 61 and 62 are respectively connected to the oil tank through a pair of replenishing pipes 65. A check valve 66 is provided in each replenishing pipe 65, which allows flow from the oil tank toward the supply and discharge pipe (61 or 62) but prohibits flow in the opposite direction.

[0145] The bucket control valve 44 controls supply and discharge of hydraulic oil to and from the bucket cylinder 13 . That is, the hydraulic oil is supplied from the second main pump 23 to the bucket cylinder 13 via the bucket control valve 44 .

[0146] Although Figure 4 Not shown, but the control valve on the second center discharge line 34 includes the first control valve 45 of the boom (refer to Figure 5 ), the control valve on the first central discharge line 31 includes a boom second control valve 46 (refer to Figure 5 The boom second control valve 46 is a valve dedicated to the boom raising operation. That is, during the boom raising operation, hydraulic oil is supplied to the boom cylinder 11 via the boom first control valve 45 and the boom second control valve. During the boom lowering operation, hydraulic oil is supplied to the boom cylinder 11 only via the boom first control valve 45.

[0147] like Figure 5 As shown, the boom control valve 47 (the boom first control valve 45 and the boom second control valve) is operated by the boom operating device 71. The arm control valve 40 (the arm first control valve 41 and the arm second control valve 42) is operated by the arm operating device 51. The rotation control valve 43 is operated by the rotation operating device 54. The bucket control valve 44 is operated by the bucket operating device 57. The boom operating device 71, the arm operating device 51, the rotation operating device 54 and the bucket operating device 57 each include an operating lever and output an operating signal (instruction) corresponding to the tilt angle of the operating lever.

[0148] In this embodiment, the boom control device 71, the arm control device 51, the swing control device 54, and the bucket control device 57 are pilot control valves that output pilot pressure commands corresponding to the tilt angle of the control lever. Therefore, the arm control device 51 is connected to the pair of pilot ports of the arm first control valve 41 via a pair of pilot lines 52 and 53. The swing control device 54 is connected to the pair of pilot ports of the swing control valve 43 via a pair of pilot lines 55 and 56. The bucket control device 57 is connected to the pair of pilot ports of the bucket control valve 44 via a pair of pilot lines 58 and 59. The boom control device 71 is connected to the pair of pilot ports of the boom first control valve 45 via a pair of pilot lines 72 and 73. Furthermore, the pair of pilot ports of the arm second control valve 42 are connected to the pilot lines 52 and 53 via a pair of pilot lines 52a and 53a. In the boom second control valve 46, only the pilot port for boom raising operation is connected to the pilot line 73 via the pilot line 73a, and the other pilot port is not connected to the pilot line 72. Therefore, the boom second control valve 46 does not operate when the boom operating device 71 performs a boom lowering operation.

[0149] However, each operating device may be an electric joystick that outputs an electric signal (command) corresponding to the tilt angle of the operating lever, and a pair of electromagnetic proportional valves may be connected to the pilot port of each control valve.

[0150] Pressure sensors 81 to 86, 91, and 92 for detecting the pressure of the pilot pressure command are provided in the pilot lines 52, 53, 55, 56, 58, 59, 72, and 73, respectively. Pressure sensors 81 and 82 for detecting the pressure of the pilot pressure command output from the boom operating device 51 may also be provided in the pilot lines 52a and 53a. These pressure sensors 81 to 86, 91, and 92 constitute the operation command detection unit 110 (see FIG. Figure 6 ).

[0151] The first flow regulating device 22 and the second flow regulating device 24 are electrically controlled by the flow control device 8. For example, the flow control device 8 has a memory such as a ROM, a RAM, and a CPU, and the CPU executes a program stored in the ROM. The flow control device 8 controls the first flow regulating device 22 and the second flow regulating device 24 in such a manner that the larger the pilot pressure instruction (operation signal) detected by the pressure sensors 81 to 86, 91, and 92, the larger the tilting angle of the first main pump 21 and / or the second main pump 23. For example, when a rotation operation is performed alone, the flow control device 8 controls the first flow regulating device 22 in such a manner that the larger the pilot pressure instruction output from the rotation operation device 54, the larger the tilting angle of the first main pump 21.

[0152] [Structure of the control system]

[0153] Next, the configuration of the control system of the hydraulic excavator 10 with a learning function will be described.

[0154] Overall Structure

[0155] Figure 6 1 is a functional block diagram showing the structure of a control system of a hydraulic excavator 10 with a learning function. Figure 6 The control system of the hydraulic excavator 10 related to the learning function in this embodiment is shown in the figure. Therefore, the control system of the traveling body 19 not related to the learning function in this embodiment is omitted. The hydraulic excavator 10 with a learning function also includes an overall control unit and an operation mode switching operation unit (not shown). The overall control unit switches the operation mode of the hydraulic excavator 10 with a learning function to a learning mode or an automatic control mode in response to an operator's operation of the operation mode switching operation unit.

[0156] Reference Figure 6 The acceleration device 50 , the arm operating device 51 , the swing operating device 54 , the bucket operating device 57 , and the boom operating device 71 constitute an operating unit 101 . The operating unit 101 is provided at a driver's seat of the swing structure 15 .

[0157] When an operator steps on the accelerator pedal of the accelerator device 50, the accelerator device 50 outputs an acceleration command corresponding to the amount the accelerator pedal was stepped on, i.e., an actuation command 201. The engine 26 then drives the pump unit 107 with an output corresponding to the actuation command 201. The pump unit 107 then discharges hydraulic fluid into the hydraulic circuit 106 at a discharge rate corresponding to the output of the pump unit 107.

[0158] When the operator operates the operating lever of the rotation operating device 54, the rotation operating device 54 outputs a pilot pressure command (rotation command) corresponding to the tilt angle of the operating lever, namely, an operation command 201. The rotation control valve 43 then supplies or discharges hydraulic oil to or from the rotation motor 14 in accordance with the operation command 201. The rotation motor 14 then rotates the rotating body 15 in response to the supply and discharge of hydraulic oil.

[0159] When the operator operates the operating lever of the boom operating device 71, the boom operating device 71 outputs a pilot pressure command (boom operation command) corresponding to the tilt angle of the operating lever, namely, an operation command 201. The boom control valve 47 then supplies or discharges hydraulic oil to the boom cylinder 11 in accordance with the operation command 201. The boom cylinder 11 then raises or lowers the boom 16, and consequently, the boom 16 and the arm 17, in response to the supply or discharge of hydraulic oil.

[0160] When the operator operates the operating lever of the arm operating device 51, the arm operating device 51 outputs a pilot pressure command (arm operation command) corresponding to the tilt angle of the operating lever, namely, an operation command 201. The arm control valve 44 then supplies or discharges hydraulic oil to the arm cylinder 12 in accordance with the operation command 201. The arm cylinder 12 then swings the arm 17 in response to the supply and discharge of hydraulic oil.

[0161] When the operator operates the operating lever of the bucket operating device 57, the bucket operating device 57 outputs an operation command 201, which is a pilot pressure command (bucket operation command) corresponding to the tilt angle of the operating lever. The bucket control valve 44 then supplies or discharges hydraulic oil to or from the bucket 18 in accordance with the operation command 201. The bucket cylinder 13 then rotates the bucket 18 in response to the supply or discharge of hydraulic oil.

[0162] Through the above operations, the work intended by the operator is performed.

[0163] On the other hand, pressure sensors 81 to 86, 91, and 92 respectively detect pilot pressure commands, or motion commands 201, output from arm operating device 51, swing operating device 54, bucket operating device 57, and boom operating device 71, and output these commands as command data 211. Pressure sensors 81 to 86, 91, and 92 constitute motion command detection unit 110.

[0164] The command data storage unit 116 of the learning data storage unit 115 stores these command data 211 in a time series. Furthermore, the command data storage unit 116 stores the acceleration command, or motion command 201, output from the acceleration device 50 in a time series as command data 211. The command data 211 includes acceleration commands, arm motion commands, rotation commands, bucket motion commands, and boom motion commands.

[0165] As described above, the first camera 311 captures the status of the work being performed by the bucket 18. The image processing unit 312 then processes the image captured by the first camera 311 to generate data representing a predetermined work status, and outputs the data as the work status data 212. Thus, the first camera 311 and the image processing unit 312 constitute the work status detection unit 112.

[0166] The hydraulic excavator 10 is equipped with a microphone 313. For example, the microphone 313 is located near the engine 26 disposed on the rotating structure 15. The microphone 313 receives the operating sound of the engine 26, converts it into operating sound data, and outputs the operating sound data as the operating state data 213. The greater the output of the engine 26, the louder the operating sound of the engine 26; the smaller the output of the engine 26, the quieter the operating sound of the engine 26. Therefore, the operating sound data represents the driving source state of the pump unit 107, and thus represents the operating state data of the hydraulic excavator 10. Therefore, the microphone 313 constitutes the driving source state detection unit, and thus the operating state detection unit 113.

[0167] A gyro sensor 314 is provided on the rotating body 15 or the main body 102 of the hydraulic excavator 10. The gyro sensor 314 detects the inclination and vibration (including excitation force, acceleration, and angular acceleration, etc.) of the rotating body 15 or the main body 102, converts it into inclination and vibration data, and outputs the inclination and vibration data as reaction data 214. For example, when the boom causes the bucket 18 to penetrate the ground, the rotating body 15 and the main body 102 tilt due to the reaction force from the ground, and the rotating body 15 and the main body 102 vibrate due to the excitation force via the boom (action part 103). Therefore, the inclination and vibration data output by the gyro sensor 314 represent the reaction received by the action part 103 and the main body 102. Therefore, the gyro sensor 314 constitutes the reaction detection part 114.

[0168] The prediction basic data storage unit 117 of the learning data storage unit 115 stores the work status data 212 output by the image processing unit 312, the action status data 213 output by the microphone 313, the reaction data 214 output by the gyro sensor 314, the instruction data 211 output by the action instruction detection unit 110, and the action instruction output by the acceleration device 50, i.e., the instruction data 211, as prediction basic data in time series.

[0169] As described above, the learning unit 118 performs machine learning on the learning data during learning, and after the machine learning is completed, upon receiving input of the prediction basic data Pd during automatic control, outputs the prediction operation command Pf.

[0170] The learning data storage unit 115, the learning unit 118, the image processing unit 312, and the above-mentioned overall control unit (not shown) are composed of, for example, an arithmetic unit having a processor and a memory. The learning unit 118, the overall control unit, and the image processing unit 312 are functional modules that are realized by the processor executing a prescribed program stored in the memory of the arithmetic unit. The learning data storage unit 115 is composed of the memory. Specifically, the arithmetic unit is composed of, for example, a microcontroller, an MPU, an FPGA (Field Programmable Gate Array), a PLC (Programmable Logic Controller), etc. The learning data storage unit 115, the learning unit 118, the image processing unit 312, and the above-mentioned overall control unit (not shown) can also be composed of a single arithmetic unit that performs centralized control, or can be composed of a plurality of arithmetic units that perform distributed control. These single arithmetic units or a plurality of arithmetic units are, for example, provided on the rotating body 15 of the hydraulic excavator 10.

[0171] The operating unit drive unit 119 is comprised of first to fifth servomotors M1 to M5. These first to fifth servomotors M1 to M5 rotate the operating lever of the boom operating device 71, the operating lever of the bucket operating device 57, the operating lever of the swing operating device 54, the operating lever of the arm operating device 51, and the accelerator pedal of the accelerator 50, respectively, according to the predicted motion command Pf output by the learning unit 118. Consequently, during automatic control, the boom control valve 47, bucket control valve 44, swing control valve 43, and arm control valve 40 of the hydraulic drive system 1 drive the boom 16, bucket 18, swing structure 15, and arm 17 of the operating unit 103 according to the predicted motion command Pf. Furthermore, the engine 26 drives the pump unit 107 of the hydraulic drive system 1 according to the predicted motion command Pf.

[0172] Next, the configuration of the learning unit 118 will be described in detail. Figure 7 Schematic diagram showing time series data of the predicted operation command Pf, the learning command data Pf′, the prediction basic data Pd, and the learning prediction basic data Pd′ in the hydraulic excavator 10 with a learning function. Figure 8 It is a functional block diagram showing the configuration of the learning unit 118 .

[0173] <Temporal relationship between each time series data>

[0174] First, the temporal relationship between each time series data will be described.

[0175] Reference Figure 7The hydraulic excavator 10 performs a predetermined operation through automatic control. To teach the hydraulic excavator 10 this predetermined operation, a skilled operator performs the predetermined operation as a learning exercise. Time t0' indicates the start time of this learning exercise. This learning exercise ends at time tu'. During this learning exercise, command data 211 and prediction base data 212-214 are acquired at predetermined sampling intervals and stored in a time series in the command data storage unit 116 and the prediction base data storage unit 117, respectively.

[0176] When the learning action is completed, the instruction data and the prediction base data are read out from the instruction data storage unit 116 and the prediction base data storage unit 117 respectively as the learning instruction data Pf' and the learning prediction base data Pd'. The learning instruction data Pf' is the time series data Pf0', Pf1', Pf3'...Pfu' (hereinafter referred to as Pf0'~Pfu'). In the following, the subscript numbers in each time series data represent the order of the sampling time (interval time). Therefore, the time series data with the same subscript number refers to the data obtained at the same sampling time. The learning prediction base data Pd' is also the time series data Pd0'~Pdu'.

[0177] Machine learning is performed using these learning command data Pf' and learning prediction base data Pd'. When machine learning is completed, the hydraulic excavator 10 is automatically controlled to perform a predetermined operation. The current time t0 represents the start time of the predetermined operation during automatic control. During automatic control, command data 211 and prediction base data 212-214 are acquired at predetermined sampling intervals and sequentially input into the learning unit 118 as prediction base data Pd. The prediction base data Pd is time series data Pd0-Pdu.

[0178] The learning unit 118 outputs a predicted operation command Pf based on the prediction basic data Pd. The predicted operation command Pf is time series data Pf0 to Pfu.

[0179] The hydraulic excavator 10 operates according to the predicted operation command Pf, whereby the hydraulic excavator 10 is automatically controlled.

[0180] <Structure of the learning unit 118>

[0181] Reference Figure 8 The learning unit 118 includes, for example, a neural network 1400 , a learning data and teacher data generating unit 1401 , a data input unit 1402 , and a learning evaluation unit 1403 .

[0182] Neural network 1400 includes an input layer, intermediate layers, and an output layer. The number of neurons in each layer is appropriately set. Known learning methods can be applied to learning neural network 1400. Therefore, a brief description is provided here. Here, neural network 1400 is, for example, a recursive neural network. The learning method is, for example, supervised learning.

[0183] The learning data and teacher data generating unit 1401 generates time series data pn1 to pnu of teacher data pn based on time series data Pf0' to Pfu' of learning instruction data Pf' and generates time series data Pd0' to Pdu-1' of learning data based on time series data Pd0' to Pdu' of learning prediction basic data Pd'.

[0184] The data input unit 1402 sequentially inputs the time series data Pd0' to Pdu-1' of the learning data to each neuron of the input layer. At this time, if the data input unit 1402 inputs the time series data Pdi of the learning data at a certain sampling time ti, the neural network 1400 calculates the predicted action instruction Pni+1 at the next sampling time ti+1 through forward operation. Then, the learning evaluation unit 1403 obtains the time series data pni+1 at the next sampling time ti+1 based on the time series data pn1 to pnu of the teacher data pn, and calculates the sum of the square errors e of the acceleration instruction, the arm action instruction, the rotation instruction, the bucket action instruction, and the boom action instruction, for example, with respect to the predicted action instruction Pni+1 and the time series data pni+1 of the teacher data pn. 2 Then, the learning evaluation unit 1403 updates the weights of the neural network 1400 through backward operation. The data input unit 1402 and the learning evaluation unit 1403 perform this process on all the time series data Pd0' to Pdu-1' of the learning data. For example, in all the processes, when the sum of the squared errors e 2 When the value falls below a predetermined threshold, the learning is terminated.

[0185] After learning is complete, during automatic control, data input unit 1402 inputs prediction base data Pd0 for the current sampling time t0, for example. Then, neural network 1400 outputs predicted action command Pn1 for the next sampling time t1 as predicted action command Pf1. Furthermore, it outputs an appropriate initial action command as predicted action command Pf0.

[0186] As a result, the hydraulic excavator 10 is automatically controlled by the predicted operation command Pf based on the learning result of the neural network 1400 (learning unit 118 ).

[0187] Furthermore, during learning, the data input unit 1402 may input the previous time series data Pdi-1 to Pdi-n (n is a predetermined positive number) when inputting the time series data Pdi of the learning data at each sampling time ti. However, in this case, during automatic control, the data input unit 1402 must similarly input the past prediction base data Pdj-1 to Pdj-n at each sampling time tj along with the current prediction base data Pdj. During automatic control, the prediction base data storage unit 117 stores the prediction base data Pd0 to Pdj-1 in a time series format at each sampling time tj. Therefore, the data input unit 1402 reads this data from the prediction base data storage unit 117 to generate the past prediction base data Pdj-1 to Pdj-n. This improves the learning efficiency of the neural network 1400. The reason is that when the operator predicts the movement of the bucket 18, he not only considers the instantaneous working conditions, the operating state of the hydraulic excavator 10 and the reaction from the working object at the current moment, but also considers a series of previous working conditions, the operating state of the hydraulic excavator 10 and the reaction from the working object, and predicts the next movement of the bucket 18, thereby accurately predicting the movement of the bucket 18.

[0188] Furthermore, information other than the working conditions, the operating state of the hydraulic excavator 10 , and the reaction from the working object may be used as learning data and input data for automatic control of the hydraulic excavator 10 .

[0189] [action]

[0190] Next, the operation of the hydraulic excavator 10 with a learning function configured as described above will be described. Hereinafter, the operation of the hydraulic excavator 10 will be described by taking as an example a case where the hydraulic excavator 10 performs a pit digging operation as a predetermined operation.

[0191] Figure 9 Schematic diagram showing the status of construction work performed by the hydraulic excavator 10 with a learning function.

[0192] Reference Figure 9 First, a first camera 311 is installed above the planned location of the pit 131. The first camera 311 is mounted on, for example, a suitable support member installed on the ground. The first camera 311 is connected to the aforementioned computing unit (learning unit 118) installed on the rotating body 15 of the hydraulic excavator 10, for example, via wireless communication, so that data can be communicated. The first camera 311 is located, for example, above the center of the pit 131 and is installed so that the optical axis 321 is directed toward the center of the pit 131.

[0193] Next, the hydraulic excavator 10 is switched to the learning mode. First, the hydraulic excavator 10 is set to an initial state. As the initial state, for example, the hydraulic excavator 10 is located at a setting location suitable for digging the pit 131 and takes an initial posture (for example, Figure 9 The installation position is set, for example, to a location where the bucket 18 can shovel soil to dig the pit 131 without moving the hydraulic excavator 10, and the shoveled soil in the bucket 18 can be discarded to the soil storage area. Here, the installation position is assumed to be midway between the planned location of the pit 131 and the soil storage area. Furthermore, the first imaging device 311 is assumed to be a three-dimensional camera.

[0194] <Digging work>

[0195] Next, the skilled operator performs a digging operation by operating the hydraulic excavator 10. The operator performs the digging operation roughly as follows, for example.

[0196] First, the bucket 18 is lowered from the lowered position above the pit 131 and penetrates the ground at the predetermined location of the pit 131 (including the ground inside the pit 131) (penetrating action). During this penetrating action, the operator operates the bucket 18 so that the claw at the end faces downward before lowering the bucket 18. Then, while pressing the penetrated bucket 18 against the ground, the bucket 18 is rotated forward to scoop up sand and soil (scooping action). Next, the bucket 18 is moved to the lowered position and lifted from the pit 131 (lifting action). Next, the rotating body 15 is rotated until the booms 16 and 17 are facing the location where the sand and soil are placed (forward rotation action). Next, the bucket 18 is rotated to the opposite side to discharge the sand and soil in the bucket 18 to the location where the sand and soil are placed (sand discharging action). Next, the rotating body 15 is rotated in the opposite direction so that the bucket 18 is in the lowered position (reverse rotation action). Thereafter, the above series of operations are repeated, and when a predetermined pit 131 having a predetermined planar shape and a predetermined depth is formed, the pit excavation work is completed.

[0197] <Judgment information for the next operation>

[0198] During this series of actions, the operator visually confirms the status of the digging, visually identifies the posture of the bucket 18 and the booms 16 and 17, and visually observes how the bucket 18, the boom and the driver's seat (rotating body 15) are currently moving through the current operating positions of the operating levers of the operating devices 51, 54, 57, and 71. In addition, when the bucket 18 acts on the ground (digs the ground, scrapes into the soil, etc.), the operator feels its reaction by the body to judge whether the intended operation (action) is appropriate or not. Then, these are instantaneously considered to decide the next operation. Here, the so-called reaction refers to, for example, the tilt and vibration of the driver's seat (including excitation force, acceleration, angular acceleration, etc.). In addition, when deciding the next operation, the operator pays attention to the state of the engine 26 as the drive source (rotation amount, sound, etc.).

[0199] Here, the digging status is an example of information indicating the working status. The operating position is an example of information indicating the operating state of the hydraulic excavator 10. The tilt and vibration of the operator's seat (rotating structure 15) are examples of reaction from the ground. The status of the engine 26 (rotational speed, sound, etc.) is information indicating the state of the drive source and, therefore, the operating state of the hydraulic excavator 10.

[0200] <Acquiring data for learning>

[0201] On the other hand, in the above series of operations, data for learning is acquired as follows.

[0202] The first camera 311 captures the status of the digging operation performed by the bucket 18, and the image processing unit 312 performs image processing on the captured image to generate data representing the status of the digging operation as operation status data 212, and the data is stored by the learning data storage unit 115 (more precisely, the prediction basic data storage unit 117).

[0203] Specifically, the first camera 311 primarily captures images of the arm 17, bucket 18, and pit 131 during the thrusting to lifting motion, and primarily captures images of pit 131 during the forward rotation to reverse rotation motion. The image processing unit 312 applies known image processing, such as edge processing, to the captured images to distinguish the areas of the arm 17 and bucket 18 from those of pit 131, thereby generating data on the planar shape and central depth of pit 131. In particular, since the images captured during the soil discharge motion do not include the arm 17 and bucket 18 areas, and only the pit 131 area, data on the planar shape and central depth of pit 131 can be accurately determined in the work status data 212 during the soil discharge motion.

[0204] When the operator operates the accelerator 50, arm operating device 51, rotation operating device 54, bucket operating device 57, and boom operating device 71, the learning data storage unit 115 (command data storage unit 116 and prediction base data storage unit 117) stores command data 211 of the motion command 201 corresponding to each operation. During the above series of operations, the combination of command data 211 corresponding to the operations of the accelerator 50, arm operating device 51, rotation operating device 54, bucket operating device 57, and boom operating device 71 determines how the bucket 18, boom, and operator's seat (swinging structure 15) will move at each moment.

[0205] During the above series of operations, the microphone 313 receives the sound of the engine 26. The learning data storage unit 115 (more precisely, the prediction base data storage unit 117) stores the sound data (operation state data 213).

[0206] In particular, during the shoveling operation, for example, the operator increases the engine output and rotates the bucket 18. This allows for appropriate shoveling of sand and soil on the ground surface of the pit 131. Therefore, during the shoveling operation, the magnitude of the increased engine output is determined by the magnitude of the sound data.

[0207] During the above series of operations, the gyro sensor 314 detects the tilt and vibration (including excitation force, acceleration, and angular acceleration) of the rotating body 15 (driver's seat) or the main body 102 as a reaction from the floor of the pit 131. The learning data storage unit 115 (more precisely, the prediction base data storage unit 117) stores this reaction data 214.

[0208] In particular, during the piercing action, the reaction from the ground when the bucket 18 penetrates the ground indicates the hardness of the ground. Therefore, during the piercing action, the hardness of the ground that the bucket 18 penetrates is determined based on the magnitude of the reaction data.

[0209] In this manner, in the present embodiment, data that can identify information suitable for determining the next operation is acquired as learning data.

[0210] Machine Learning

[0211] After the learning operation is completed, the neural network performs machine learning using the learning data stored in the storage unit. As described above, when the time series data Pdi of the learning data at each sampling time ti is input, the previous time series data Pdi-1 to Pdi-n (n is a predetermined positive number) are also input.

[0212] Automatic Control

[0213] The hydraulic excavator 10 that has completed the mechanical learning is set to the above-mentioned initial state. Next, the hydraulic excavator 10 is switched to the automatic control mode. Then, the automatic control of the hydraulic excavator 10 is started. During this automatic control, the prediction basic data Pd (work status data 212, instruction data 211, action state data (sound data) 213 and reaction data 214) are input to the learning unit 118 (to be precise, the data input unit 402 of the neural network) and stored in the prediction basic data storage unit 117. In addition, at each sampling time tj, the data input unit 402 inputs the past prediction basic data Pdj-1 to Pdj-n in the time series data Pd0 to Pdj-1 stored in the prediction basic data storage unit 117 together with the current prediction basic data Pdj. As a result, the learning unit 118 outputs the prediction action instruction Pf.

[0214] In this manner, the hydraulic excavator 10 is automatically controlled, and the excavation work is performed in the same manner as when it is operated by a skilled operator.

[0215] As described above, according to the first embodiment, during learning, when a skilled operator operates the construction machine 100 to perform a predetermined operation, command data 211 corresponding to the motion command 201 corresponding to the operation and prediction base data are stored in the learning data storage unit 115. The prediction base data includes data 212 based on the working conditions of the working unit 104, data 213 on the operating state of the motion unit 103, and data 214 on the reaction to the working object received by the motion unit 103 or the main body 102 due to the operation of the working unit 104. Then, by inputting the prediction base data stored in the learning data storage unit 115, the learning unit 118 performs machine learning on the learning data and outputs a predicted motion command. This allows the skilled operator's operation to be learned and output as a predicted motion command Pf. Furthermore, during automatic control, when the prediction base data Pd is input to the learning unit 118, the learning unit 118 outputs the predicted motion command Pf. The hydraulic drive system 1 then drives the motion unit 103 according to the predicted motion command Pf. As a result, the operation performed by a skilled operator can be learned and the work can be automatically performed according to the predicted motion command Pf outputted from the learning unit 118 as a result of the learning. Therefore, it is possible to provide a construction machine 100 that can learn the work performed by a human operator and automatically perform the learned work.

[0216] (Implementation Method 2)

[0217] The second embodiment of the present invention illustrates an embodiment in which, in the hydraulic excavator 10 according to the first embodiment, the operating state detection unit 113 further includes a second imaging device 331 and an image processing unit 333 .

[0218] Figure 10It is a side view showing the hardware configuration of the hydraulic excavator 10 with a learning function according to the second embodiment. Figure 11 Yes Figure 10 Functional block diagram of the control system structure of the hydraulic excavator 10 with a learning function.

[0219] Reference Figure 10 as well as Figure 11 The operating state detection unit 113 of the hydraulic excavator 10 of this embodiment further includes a second imaging device 331 and an image processing unit 333. The remaining configuration is the same as that of the hydraulic excavator 10 of the first embodiment.

[0220] The second camera 331 captures the entire hydraulic excavator 10. The image captured by the second camera 331 is processed by the image processing unit 333 to obtain data representing the posture of the hydraulic excavator 10. The image processing unit 333 then outputs the image processing unit 333 as the operating state data 213. The optical axis 332 of the second camera 331 is directed toward the hydraulic excavator 10.

[0221] The second imaging device 331 is constituted by, for example, a common digital camera and is fixed to a fixed object (such as the ground) different from the vehicle of the hydraulic excavator 10 via an appropriate supporting member, or is mounted on a drone.

[0222] The image processing unit 333 processes the image captured by the second imaging device 331, extracts the external shape of the hydraulic excavator 10, and outputs the external shape data as the operation state data 213. The external shape data is posture data that can identify the posture of the operation unit 103 of the hydraulic excavator 10.

[0223] According to embodiment 2, in the series of actions described in embodiment 1, the posture of the hydraulic excavator 10 is determined by the outer shape of the hydraulic excavator 10 in the action state data 213, so the efficiency of the mechanical learning of the learning unit 118 is improved, and the predicted action instruction Pf output by the learning unit 118 during automatic control becomes more appropriate.

[0224] (Implementation 3)

[0225] In the third embodiment of the present invention, an example is given in which the operating state detection unit 113 further includes a sensor unit 341 in the hydraulic excavator 10 of the first embodiment.

[0226] Figure 12 It is a side view showing the hardware configuration of the hydraulic excavator 10 with a learning function according to the third embodiment. Figure 13 Yes Figure 12 Functional block diagram of the control system structure of the hydraulic excavator 10 with a learning function.

[0227] Reference Figure 13 In the hydraulic excavator 10 of this embodiment, the operating state detection unit 113 further includes a sensor unit 341. The configuration other than this is the same as that of the hydraulic excavator 10 of Embodiment 1. The sensor unit 341 is composed of sensors S1 to S4.

[0228] Reference Figure 12 The hydraulic excavator 10 is provided with sensors S1 to S4. Specifically, the rotating body 15 is provided with a sensor S1 that detects the rotation angle of the rotating body 15 about the rotation axis A1. The base end of the boom 16 is provided with a sensor S2 that detects the rotation angle of the boom 16 about the rotation axis A2. The base end of the arm 17 is provided with a sensor S3 that detects the rotation angle of the arm 17 about the rotation axis A3. The distal end of the arm 17 is provided with a sensor S4 that detects the rotation angle of the bucket 18 about the rotation axis A4.

[0229] The sensor unit 341 outputs data on the rotation angles detected by the sensors S1 to S4 as the operating state data 213. The combination of the data on the rotation angles detected by the sensors S1 to S4 is posture data capable of specifying the posture of the operating unit 103 of the hydraulic excavator 10.

[0230] According to embodiment 3, in the series of actions described in embodiment 1, the posture of the hydraulic excavator 10 is determined by the combination of the rotation angles respectively detected by sensors S1 to S4 in the action state data 213, so the efficiency of the mechanical learning of the learning unit 118 is improved, and during automatic control, the predicted action instruction Pf output by the learning unit 118 becomes more appropriate.

[0231] (Implementation 4)

[0232] {summary}

[0233] First, an overview of the skill-transferring construction machine according to the fourth embodiment will be described.

[0234] [structure]

[0235] Figure 14 This is a functional block diagram showing the control system structure of the skill transfer construction machine according to the fourth embodiment of the present invention. Figure 14 In FIG, arrows represent the flow of instructions, power, information, and data. Solid arrows represent the flow of instructions or data when the action unit 103 is in operation, and dashed arrows represent the flow of instructions or data when learning. In addition, as will be described later, when past prediction basic data is used in learning, the flow of data represented by dashed lines occurs even when the action unit is in operation. This is Figure 18 、 26 , the same goes for No. 28.

[0236] Reference Figure 14 The skill inheritance construction machine 200 of embodiment 4 further includes an operating unit 101, a main body 102, an action unit 103, a hydraulic drive system 105, an operation status detection unit 112, an action state detection unit 113, a reaction detection unit 114 and a control unit 401.

[0237] The control unit 401 includes a basic motion instruction unit 2119, a motion instruction generator 2120, a motion correction instruction generator 2110, a learning data storage unit 115, and a learning unit 118. The learning data storage unit 115 includes a motion correction instruction storage unit 2116 and a prediction base data storage unit 117.

[0238] In embodiment 4, the construction machinery with a learning function is a skill-transferring construction machinery 200 having a control unit 401, the operating unit 101 is configured to output, as an instruction 201, a manual action correction instruction 403 corresponding to the operator's operation, the hydraulic drive system 105 is configured to drive the action unit 103 according to the basic action instruction 402, the automatic action correction instruction 404 and the manual action correction instruction 403, and the control unit 401 has: a basic action instruction unit 2119, which outputs the basic action instruction 402 that causes the working unit to perform a basic action through the action unit 103; an action correction instruction generation unit 2110, which serves as an instruction data generation unit 1101, adds the manual action correction instruction 403 to the automatic action correction instruction 404, and generates an action correction instruction Pm; an action correction instruction storage unit 2116, which is the instruction data storage unit 1102, stores the action correction instruction Pm as instruction data in a time series; a prediction basic data storage unit 117; and a learning unit 118. Moreover, the learning unit 118 is configured to use the prediction basic data Pd' stored in the prediction basic data storage unit 117 to perform machine learning on the action correction instruction Pm' stored in the action correction instruction storage unit 2116. After completing the machine learning, the learning unit 118 receives the input of the prediction basic data Pd when the action unit 103 is in action, and outputs the automatic action correction instruction 404 as the prediction instruction 1103.

[0239] Hereinafter, the structure of the skill-transferring construction machine 200 will be described in detail.

[0240] "Construction machinery" refers to any work machine that can perform construction work by operating a working part through an action part according to an operator's operation. Examples of "construction machinery" include hydraulic excavators, bulldozers, tractor-type forklifts, wheel loaders, trenchers, excavators, cranes, and lift trucks.

[0241] The operation unit 101 outputs a manual operation correction command 403 corresponding to the operator's operation.

[0242] The operating unit 103 is connected to the main body 102 .

[0243] The operation unit 103 includes a working unit 104 for performing work, and operates the working unit 104 to perform work. Here, "operating the working unit 104" means "operating and moving the working unit 104".

[0244] The hydraulic drive system 105 is provided across the main body 102 and the operating unit 103. The hydraulic drive system 105 outputs a driving force 202 according to an operating command 201 output from the operating command generating unit 2120, thereby driving the operating unit 103.

[0245] The work status detection unit 112 detects the status of work performed by the working unit 104 and outputs the detected work status as work status data 212 .

[0246] The operation state detection unit 113 detects the operation state of the operation unit 103 and outputs the detected operation state as operation state data 213 .

[0247] In the machine learning of the learning unit 118, which will be described later, the data used to predict the motion correction instructions to be learned is referred to as prediction base data. The prediction base data includes work status data 212, motion state data 213, and reaction data 214. For ease of explanation, the prediction base data used for learning is denoted by the reference symbol Pd', while the prediction base data used during the operation of the motion unit 103 is denoted by the reference symbol Pd. Furthermore, the prediction base data Pd and Pd' may also include data other than the work status data 212, motion state data 213, and reaction data 214.

[0248] The reaction detection unit 114 detects a reaction received by the operation unit 103 or the main body 102 from the work object due to the work of the working unit 104 , and outputs the detected reaction as reaction data 214 .

[0249] The basic operation instruction unit 2119 outputs a basic operation instruction 402 for causing the operation unit 103 to cause the working unit 104 to perform a basic operation.

[0250] The motion correction command generation unit 2110 adds the manual motion correction command 403 to the automatic motion correction command 404 to generate the motion correction command Pm.

[0251] The motion correction command storage unit 2116 stores the motion correction command Pm in time series. The prediction base data storage unit 117 stores the prediction base data including the work status data 212, the motion state data 213, and the reaction data 214 in time series.

[0252] The learning unit 118 is a learning model that performs machine learning. Examples of such learning models include neural networks, regression models, tree models, Bayesian models, time series models, clustering models, and ensemble learning models. In this embodiment, the learning model is a neural network. The learning method can be either supervised or unsupervised.

[0253] For example, in the case of teacher-based learning, the learning unit 118 reads the motion correction instruction Pm stored in the motion correction instruction storage unit 2116 as the motion correction instruction Pm' for learning, and reads the prediction base data Pd stored in the prediction base data storage unit 117 as the prediction base data Pd' for learning. Furthermore, the learning motion correction instruction Pm' is used as the teacher data pn, and the learning prediction base data Pd' is used as input data to create learning data. Furthermore, the prediction base data Pd', which serves as input data, is input into a machine learning model (e.g., a neural network), and the difference between its output Pn and the teacher data pn is evaluated, and the evaluation is fed back to the machine learning model. Thus, the machine learning model performs machine learning on the learning data. In other words, the machine learning model learns the motion correction instruction Pm' using the prediction base data Pd'.

[0254] When machine learning is complete, learning unit 118 outputs the output of the machine learning model as automatic motion correction command 404. Specifically, when motion unit 103 is operating, upon receiving input of prediction base data Pd, learning unit 118 outputs automatic motion correction command 404, which is a prediction of the motion correction command Pm' to be learned.

[0255] The motion command generation unit 2120 adds the automatic motion correction command 404 and the manual motion correction command 403 output from the operating unit 101 to the basic motion command 402 to generate the motion command 201. Here, the basic motion command, the automatic motion correction command, and the manual motion correction command are numerical data representing the opening degree of the hydraulic control valve provided in the hydraulic drive system 105, and thus can be added to and subtracted from each other.

[0256] The hydraulic drive system 105 drives the actuator 103 according to the motion command 201 .

[0257] [action]

[0258] The above-mentioned prescribed operation is, for example, a relatively simple, standardized operation. Examples of such standardized operations include digging, ground leveling, and rolling operations. Assume that such a prescribed operation is repeated multiple times by the skill-transferring construction machine 200 while changing the work location. In this case, between two consecutive prescribed operations, the learning unit 118 learns the motion correction command Pm' from the previous prescribed operation. In the subsequent prescribed operation, if the prediction basic data Pd is input to the learning unit 118, the learning unit 118 outputs an automatic motion correction command 404 reflecting the learning result.

[0259] Specifically, in the previous predetermined operation, the skilled operator operates the skill-transferring construction machine 200 and performs the predetermined operation while correcting the operation of the working unit 104 according to the basic operation command 402 and the automatic operation correction command 404 as needed.

[0260] Prediction base data Pd, including work status data 212 indicating the work status performed by the work unit 104, motion state data 213 indicating the motion state of the motion unit 103, and reaction data 214 indicating the reaction the motion unit receives from the work object due to the work performed by the work unit, is stored in a time-series manner in the prediction base data storage unit 117. Furthermore, motion correction instructions Pm, which are the addition of manual motion correction instructions 403 to automatic motion correction instructions 404, are stored in a time-series manner in the motion correction instruction storage unit 2116. Furthermore, during subsequent learning, the learning unit 118 reads the prediction base data Pd' from the prediction base data storage unit 117 and, using this data, performs machine learning on the motion correction instructions Pm' read from the motion correction instruction storage unit 2116. This allows the correction operation performed by a skilled operator to be learned and output as the automatic motion correction instructions 404. Furthermore, during the next predetermined operation, when the prediction base data Pd is input to the learning unit 118, the learning unit 118 outputs the automatic motion correction instructions 404. The hydraulic drive system 105 then drives the working unit 103 while reflecting the basic motion command 402 and the automatic motion correction command 404. As a result, the automatic motion correction command 404, reflecting the results of learning the correction operation of the skilled operator, is executed. Therefore, a construction machine 200 can be provided that allows the operator to perform correction operations on the basic motion of the working unit 104 based on the basic motion command unit 2119, thereby learning the skills of the construction machine 200 and automatically performing the learned operations.

[0261] Thus, it is possible to provide the skill-transferring construction machine 200 that can transfer the skills of skilled workers in the construction industry and can automate predetermined work in a short period of time.

[0262] {Specific structure}

[0263] Next, a specific structure of the skill-transferring construction machine 200 will be described by taking a hydraulic excavator 20 as an example of a construction machine.

[0264] [Hardware Structure]

[0265] Overall Structure

[0266] First, the overall structure of the skill-transfer hydraulic excavator 20 will be described. The overall structure of the skill-transfer hydraulic excavator 20 is basically the same as that of the hydraulic excavator with a learning function 10 according to the first embodiment, but differs in some parts.

[0267] Figure 15 1 is a side view showing the hardware configuration of a skill-transfer hydraulic excavator 20 as an example of a skill-transfer construction machine.

[0268] The skill-transfer hydraulic excavator (hereinafter, sometimes simply referred to as a hydraulic excavator) 20 includes a main body 102. A traveling body 19 is provided on the main body 102. The traveling body 19 is constituted by a vehicle travel device including a crawler track (caterpillar), for example.

[0269] A rotating body 15 is provided on the main body 102 so as to be rotatable about a vertical first rotation axis A1. A driver's seat (not shown) is provided on the rotating body 15, and an operating unit 101 (see FIG. Figure 18 ). In addition, although Figure 18 Although the operating unit 101 is not shown, a travel operating device for operating the travel body 19 is provided at the driver's seat. A rotation motor 14 is provided on the rotating body 15 to rotate the rotating body 15. The rotation motor 14 is composed of a hydraulic motor. In addition, an engine 26 for travel is provided on the rotating body 15 (see Figure 18 The engine 26 drives the pump unit 107 of the hydraulic drive system 1 during operation (see Figure 18 ).

[0270] The base end of a boom 16 is rotatably connected to the rotating body 15 about a horizontal second rotation axis A2. The distal end and base end of a boom cylinder 11 are rotatably connected to the base end of the boom 16 and the rotating body 15, respectively. The boom 16 swings about the second rotation axis A2 as the boom cylinder 11 extends and contracts.

[0271] The base end of the boom 17 is connected to the distal end of the boom 16 so as to be rotatable about a third horizontal rotation axis A3. The distal end and base end of the arm cylinder 12 are connected to the base end of the arm 17 and the center of the boom 16 so as to be rotatable, respectively. The arm 17 swings around the third rotation axis A3 as the center through the extension and contraction of the arm cylinder 12.

[0272] The base end of a bucket 18 is connected to the distal end of the arm 17 so as to be rotatable about a fourth horizontal rotation axis A4. The distal end and base end of the bucket cylinder 13 are rotatably connected to the base end of the bucket 18 and the distal end of the arm 17, respectively. Bucket 18 rotates about the fourth rotation axis A4 by extending or retracting the bucket cylinder 13. Bucket 18 is an example of an attachment, and other attachments may also be attached.

[0273] The boom 16 , the arm 17 , and the bucket 18 constitute a front working mechanism. The bucket 18 constitutes a working unit 104 , and the swing structure 15 and the front working mechanism (the boom 16 , the arm 17 , and the bucket 18 ) constitute an operating unit 103 .

[0274] The hydraulic excavator 20 includes a pair of left and right hydraulic travel motors (not shown) in addition to the above-mentioned components.

[0275] As will be described later, the operator's seat of the skill transfer hydraulic excavator 20 is provided with an operation mode switching operation unit (not shown) for switching the operation mode of the hydraulic excavator 20 between manual mode and semi-automatic mode. The operator operates the operation mode switching operation unit to set the operation mode to manual mode. Thereafter, the operator operates the operation unit 101 (including the travel operation device, not shown) to position the hydraulic excavator at the work site for the predetermined operation. Thereafter, the operator operates the operation mode switching operation unit to switch the operation mode to semi-automatic mode. Thereafter, the operator rotates the swing body 15, changes the posture of the booms 16 and 17, and rotates the bucket 18 to perform the predetermined operation.

[0276] The skill transfer hydraulic excavator 20 further includes a first camera 311. The first camera 311 captures images of the working conditions performed by the bucket 18. The images captured by the first camera 311 are processed by an image processing unit 312 (see FIG. 1 ) described later. Figure 18 ) is subjected to image processing to obtain data representing the working status, and is output from the image processing unit 312 as the working status data 212. The optical axis 321 of the first imaging device 311 is directed toward the working object.

[0277] The first camera 311 is composed of, for example, a 3D camera, a camera with a depth sensor, etc. The first camera 311 is fixed to the main body 102 via an appropriate support member, or fixed to a fixed object (such as the ground) other than the vehicle of the hydraulic excavator 20 via an appropriate support member, or mounted on a drone.

[0278] Hydraulic drive system 1

[0279] Next, the hydraulic drive system 1 for operating the hydraulic excavator 20 will be described.

[0280] Figure 161 is a hydraulic circuit diagram showing a main hydraulic circuit of the hydraulic drive system 1 of the skill transfer hydraulic excavator 20 . Figure 17 1 is a hydraulic circuit diagram showing an operating system hydraulic circuit of a drive system of the skill transfer hydraulic excavator 20. The main hydraulic circuit and the operating system hydraulic circuit are provided in the revolving body 15.

[0281] As described above, the hydraulic drive system 1 includes the boom cylinder 11 , the arm cylinder 12 , and the bucket cylinder 13 as hydraulic actuators, and also includes the swing motor 14 and a pair of left and right hydraulic travel motors (not shown).

[0282] Reference Figure 16 The hydraulic drive system 1 includes a first main pump 21 and a second main pump 23 for supplying hydraulic oil to the above-mentioned actuators. Figure 16 In order to simplify the drawing, actuators other than the rotary motor 14 are omitted.

[0283] The first main pump 21 and the second main pump 23 are driven by the engine 26. The engine 26 also drives the auxiliary pump 25. The first main pump 21, the second main pump 23 and the auxiliary pump 25 constitute the pump unit 107 (see Figure 18 The engine 26 is accelerated by the acceleration device 50 (refer to Figure 18 ) to adjust the output. The accelerator 50 includes, for example, an accelerator pedal, and outputs an acceleration correction command 75, an electrical command corresponding to the amount of depression. The engine control device (not shown) then controls the output (rotational speed) of the engine 26 according to a separate operation command 75' corresponding to the acceleration correction command 75.

[0284] The first main pump 21 and the second main pump 23 are, for example, variable displacement pumps that discharge hydraulic oil at a flow rate corresponding to the tilting angle. Here, the first main pump 21 and the second main pump 23 are swash plate pumps whose tilting angle is determined by the angle of the swash plate. However, the first main pump 21 and the second main pump 23 may also be swash axis pumps whose tilting angle is determined by the angle between the drive shaft and the cylinder block.

[0285] The discharge flow rate Q1 of the first main pump 21 and the discharge flow rate Q2 of the second main pump 23 are controlled by an electric positive control method. Specifically, the tilting angle of the first main pump 21 is adjusted by the first flow adjustment device 22, and the tilting angle of the second main pump 23 is adjusted by the second flow adjustment device 24. The auxiliary pump 25 is connected to the first flow adjustment device 22 and the second flow adjustment device 24 through the sub-discharge pipeline 37. The auxiliary pump 25 functions as a driving source for the first flow adjustment device 22 and the second flow adjustment device 24. The first flow adjustment device 22 and the second flow adjustment device 24 will be described in detail later.

[0286] The first center discharge line 31 extends from the first main pump 21 to the fuel tank. Multiple control valves, including a first boom control valve 41 and a rotary control valve 43, are located on the first center discharge line 31 (other than the first boom control valve 41 and rotary control valve 43, not shown). Each control valve is connected to the first main pump 21 via a pump line 32. In other words, the control valves on the first center discharge line 31 are connected in parallel with the first main pump 21. Furthermore, each control valve is connected to the fuel tank via a tank line 33.

[0287] Similarly, a second center discharge line 34 extends from the second main pump 23 to the fuel tank. Multiple control valves, including a second arm control valve 42 and a bucket control valve 44 (other than the second arm control valve 42 and bucket control valve 44, not shown), are located on the second center discharge line 34. Each control valve is connected to the second main pump 23 via a pump line 35. In other words, the control valves on the second center discharge line 34 are connected in parallel to the second main pump 23. Furthermore, each control valve is connected to the fuel tank via a tank line 36.

[0288] The first arm control valve 41 and the second arm control valve 42 together control the supply and discharge of hydraulic oil to the arm cylinder 12. That is, hydraulic oil is supplied from the first main pump 21 to the arm cylinder 12 via the first arm control valve 41, and hydraulic oil is supplied from the second main pump 23 to the arm cylinder 12 via the second arm control valve 42. The first arm control valve 41 and the second arm control valve 42 constitute the arm control valve 40 (see Figure 17 ).

[0289] The rotation control valve 43 controls the supply and discharge of working oil relative to the rotation motor 14. That is, the working oil is supplied from the first main pump 21 to the rotation motor 14 via the rotation control valve 43. Specifically, the rotation motor 14 is connected to the rotation control valve 43 through a pair of supply and discharge pipes 61 and 62. A release pipe 63 branches out from each supply and discharge pipe 61 and 62, and the release pipe 63 is connected to the oil tank. A pressure relief valve 64 is provided in each release pipe 63. In addition, the supply and discharge pipes 61 and 62 are connected to the oil tank through a pair of replenishing pipes 65. A check valve 66 is provided in each replenishing pipe 65, which allows flow from the oil tank toward the supply and discharge pipe (61 or 62) but prohibits flow in the opposite direction.

[0290] The bucket control valve 44 controls supply and discharge of hydraulic oil to and from the bucket cylinder 13 . That is, the hydraulic oil is supplied from the second main pump 23 to the bucket cylinder 13 via the bucket control valve 44 .

[0291] Although Figure 16 Not shown, but the control valve on the second center discharge line 34 includes the first control valve 45 of the boom (refer to Figure 17 ), the control valve on the first central discharge line 31 includes a boom second control valve 46 (refer to Figure 17 The boom second control valve 46 is a valve dedicated to the boom raising operation. That is, during the boom raising operation, hydraulic oil is supplied to the boom cylinder 11 via the boom first control valve 45 and the boom second control valve. During the boom lowering operation, hydraulic oil is supplied to the boom cylinder 11 only via the boom first control valve 45.

[0292] In addition, in the present embodiment, each of the control valves 41 to 46 is constituted by an electromagnetic control valve.

[0293] Reference Figure 17 The boom control valve 47 (the first boom control valve 45 and the second boom control valve) is operated by the boom operating device 71. The arm control valve 40 (the first arm control valve 41 and the second arm control valve 42) is operated by the arm operating device 51. The swing control valve 43 is operated by the swing operating device 54. The bucket control valve 44 is operated by the bucket operating device 57.

[0294] Specifically, the arm operating device 51, the rotation operating device 54, the bucket operating device 57, and the boom operating device 71 each include an operating lever and output an electrical command signal corresponding to the tilt angle of the operating lever, namely, a pair of individual manual operation correction commands (52, 53), (55, 56), (58, 59), and (72, 73). These individual manual operation correction commands (52, 53), (55, 56), (58, 59), and (72, 73) and the acceleration correction command 75 output by the acceleration device 50 (described later) constitute the manual operation correction command 403.

[0295] A pair of separate manual action correction instructions 52 and 53 output from the boom operating device 51 are added to a pair of separate basic action instructions (not shown) corresponding to the boom operating device 51 in the basic action instruction 402 and a pair of separate automatic action correction instructions (not shown) corresponding to the boom operating device 51 in the automatic action correction instruction 404 in the action instruction generating unit 2120 to become a pair of separate action instructions 52' and 53' corresponding to the boom operating device 51, and are respectively input to a pair of electromagnetic parts (solenoids) of the first boom control valve 41 and a pair of electromagnetic parts of the second boom control valve 42.

[0296] A pair of separate manual action correction instructions 55 and 56 output from the rotary operating device 54 are added to a pair of separate basic action instructions (not shown) corresponding to the rotary operating device 54 in the basic action instruction 402 and a pair of separate automatic action correction instructions (not shown) corresponding to the rotary operating device 54 in the automatic action correction instruction 404 in the action instruction generating unit 2120 to become a pair of separate action instructions 55' and 56' corresponding to the rotary operating device 54, and are input to a pair of electromagnetic parts of the rotary control valve 43.

[0297] A pair of separate manual action correction instructions 58 and 59 output from the bucket operating device 57 are added to a pair of separate basic action instructions (not shown) corresponding to the bucket operating device 57 in the basic action instruction 402 and a pair of separate automatic action correction instructions (not shown) corresponding to the bucket operating device 57 in the automatic action correction instruction 404 in the action instruction generation unit 2120 to become a pair of separate action instructions 58' and 59' corresponding to the bucket operating device 57, and are input to a pair of electromagnetic parts of the bucket control valve 44.

[0298] The pair of individual manual motion correction commands 72 and 73 output from the boom operating device 71 are added to the pair of individual basic motion commands (not shown) corresponding to the boom operating device 71 in the basic motion commands 402 and the pair of individual automatic motion correction commands (not shown) corresponding to the boom operating device 71 in the automatic motion correction commands 404 in the motion command generation unit 2120. These commands are then generated as a pair of individual motion commands 72' and 73' corresponding to the boom operating device 71 and input to the pair of solenoids of the first boom control valve 45. Furthermore, the individual motion command 73' is input to the boom raising solenoid of the second boom control valve 46, while no individual motion command is input to the boom lowering solenoid of the second boom control valve 46. Therefore, the second boom control valve 46 does not operate when the boom operating device 71 is operating to lower the boom.

[0299] In addition, the acceleration correction instruction 75 output from the acceleration device 50 is added to the separate basic action instruction (not shown) corresponding to the acceleration device 50 in the basic action instruction 402 and the separate automatic action correction instruction (not shown) corresponding to the acceleration device 50 in the automatic action correction instruction 404 in the action instruction generating unit 2120 to become a separate action instruction 75' corresponding to the acceleration device 50, and is input to the engine 26.

[0300] A pair of individual motion instructions ( 52 ′, 53 ′), ( 55 ′, 56 ′), ( 58 ′, 59 ′), ( 72 ′, 73 ′), and the individual motion instruction 75 ′ constitute the motion instruction 201 .

[0301] The first flow regulating device 22 and the second flow regulating device 24 are electrically controlled by the flow control device 8. For example, the flow control device 8 includes a memory such as a ROM or RAM and a CPU, and the CPU executes a program stored in the ROM. The flow control device 8 controls the first flow regulating device 22 and the second flow regulating device 24 so that the larger the pair of individual action instructions (52', 53'), (55', 56'), (58', 59'), (72', 73') corresponding to each operating device 51, 54, 57, 71, the larger the tilt angle of the first main pump 21 and / or the second main pump 23. For example, when a rotation operation is performed alone, the flow control device 8 controls the first flow regulating device 22 so that the larger the pair of individual action instructions (55', 56') corresponding to the rotation operating device 54, the larger the tilt angle of the first main pump 21.

[0302] [Structure of the control system]

[0303] Next, the configuration of the control system of the skill-transfer hydraulic excavator 20 will be described.

[0304] Overall Structure

[0305] Figure 18 2 is a functional block diagram showing the structure of the control system of the skill transfer hydraulic excavator 20. Figure 18 The control system of the hydraulic excavator 20 related to the learning function in this embodiment is shown in the figure. Therefore, the control system of the traveling body 19 not related to the learning function in this embodiment is omitted. In addition, the skill-transfer hydraulic excavator 20 includes an overall control unit and an operation mode switching operation unit (not shown). The overall control unit switches the operation mode of the skill-transfer hydraulic excavator 20 to manual mode or semi-automatic mode based on the operator's operation of the operation mode switching operation unit.

[0306] <Operation Department>

[0307] Reference Figure 18 The acceleration device 50 , the arm operating device 51 , the swing operating device 54 , the bucket operating device 57 , and the boom operating device 71 constitute an operating unit 101 . The operating unit 101 is provided at a driver's seat of the swing structure 15 .

[0308] Hydraulic control

[0309] Manual Mode

[0310] In manual mode, when the operator depresses the accelerator pedal of the accelerator device 50, the accelerator device 50 outputs an acceleration correction command 75 corresponding to the amount the accelerator pedal is depressed. The engine 26 then drives the pump unit 107 with an output corresponding to the individual operation command 75' corresponding to the acceleration correction command 75. The pump unit 107 then discharges hydraulic fluid into the hydraulic circuit 106 at a discharge rate corresponding to the output of the pump unit 107.

[0311] When the operator operates the operating lever of the rotation operating device 54, the device 54 outputs a pair of individual manual operation correction commands 55 and 56 (rotation manual operation correction commands) corresponding to the tilt angle of the operating lever. The rotation control valve 43 then supplies and discharges hydraulic oil to and from the rotation motor 14 in accordance with a pair of individual operation commands 55' and 56' corresponding to the individual manual operation correction commands 55 and 56. The rotation motor 14 then rotates the rotating body 15 in response to the supply and discharge of hydraulic oil.

[0312] When the operator operates the operating lever of the boom operating device 71, the boom operating device 71 outputs a pair of individual manual operation correction commands 72 and 73' (boom manual operation correction commands) corresponding to the tilt angle of the operating lever. The boom control valve 47 then supplies and discharges hydraulic oil to the boom cylinder 11 in accordance with the pair of individual operation commands 72' and 73' corresponding to the individual manual operation correction commands 72 and 73. The boom cylinder 11 then raises and lowers the boom 16 in response to the supply and discharge of hydraulic oil.

[0313] When the operator operates the control lever of the arm operating device 51, the arm operating device 51 outputs a pair of individual manual operation correction commands 52 and 53 (arm manual operation correction commands) corresponding to the tilt angle of the control lever. The arm control valve 40 then supplies and discharges hydraulic oil to the arm cylinder 12 in accordance with a pair of individual operation commands 52' and 53' corresponding to the individual manual operation correction commands 52 and 53. The arm cylinder 12 then swings the arm 17 in response to the supply and discharge of hydraulic oil.

[0314] When the operator operates the operating lever of the bucket operating device 57, the bucket operating device 57 outputs a pair of individual manual operation correction commands 58 and 59 (bucket manual operation correction commands) corresponding to the tilt angle of the operating lever. The bucket control valve 44 then supplies and discharges hydraulic oil to the bucket cylinder 13 in accordance with a pair of individual operation commands 58' and 59' corresponding to the individual manual operation correction commands 58 and 59. The bucket cylinder 13 then rotates the bucket 18 in response to the supply and discharge of hydraulic oil.

[0315] Through the above operations, the work intended by the operator is performed.

[0316] Semi-automatic mode

[0317] In the semi-automatic mode, the action unit 103 (15-18) acts according to the basic action instruction 402 output from the basic action instruction unit 2119 and the automatic action correction instruction 404 output from the learning unit 118. If the operator operates the operation unit 101 (50, 51, 54, 57, 71), the action of the action unit 103 is corrected according to the operation.

[0318] <Forecast Basic Data Testing Department>

[0319] As described above, the first camera 311 captures the status of the work being performed by the bucket 18. The image processing unit 312 then processes the image captured by the first camera 311 to generate data representing a predetermined work status, and outputs the data as the work status data 212. Thus, the first camera 311 and the image processing unit 312 constitute the work status detection unit 112.

[0320] The hydraulic excavator 20 is equipped with a microphone 313. For example, the microphone 313 is located near the engine 26 disposed on the rotating structure 15. The microphone 313 receives the operating sound of the engine 26, converts it into operating sound data, and outputs the operating sound data as the operating state data 213. The greater the output of the engine 26, the louder the operating sound of the engine 26; the smaller the output of the engine 26, the quieter the operating sound of the engine 26. Therefore, the operating sound data represents the driving source state of the pump unit 107, and thus represents the operating state data of the hydraulic excavator 20. Therefore, the microphone 313 constitutes the driving source state detection unit, and thus the operating state detection unit 113.

[0321] A gyro sensor 314 is provided on the rotating body 15 or the main body 102 of the hydraulic excavator 20. The gyro sensor 314 detects the inclination and vibration (including excitation force, acceleration, and angular acceleration, etc.) of the rotating body 15 or the main body 102, converts it into inclination and vibration data, and outputs the inclination and vibration data as reaction data 214. For example, if the boom 16 and the arm 7 cause the bucket 18 to penetrate the ground, the rotating body 15 and the main body 102 tilt due to the reaction force from the ground, and the rotating body 15 and the main body 102 vibrate due to the excitation force via the arm 17 and the boom 16 (the action part 103). Therefore, the inclination and vibration data output by the gyro sensor 314 represent the reaction received by the action part 103 and the main body 102. Therefore, the gyro sensor 314 constitutes the reaction detection unit 114.

[0322] <Control Department>

[0323] The control unit 401, the image processing unit 312, and the aforementioned overall control unit (not shown) are composed of, for example, an arithmetic unit having a processor and a memory. The motion correction instruction generation unit 2110, the learning unit 118, the basic motion instruction unit 2119, the motion instruction generation unit 2120, the overall control unit, and the image processing unit 312 are functional modules implemented by the processor executing a prescribed program stored in the memory of the arithmetic unit. The learning data storage unit 115 is composed of the memory. Specifically, the arithmetic unit is composed of, for example, a microcontroller, an MPU, an FPGA (Field Programmable Gate Array), a PLC (Programmable Logic Controller), etc. The control unit 401, the image processing unit 312, and the aforementioned overall control unit (not shown) can also be composed of a single arithmetic unit that performs centralized control, or can be composed of multiple arithmetic units that perform distributed control. These single arithmetic units or multiple arithmetic units are, for example, provided on the rotating body 15 of the hydraulic excavator 20.

[0324] <Basic motion command unit 2119>

[0325] The basic motion instruction unit 2119 outputs the basic motion instruction 402 in accordance with the control program for a predetermined operation stored in the memory of the arithmetic unit.

[0326] <Action Command Generator 2120>

[0327] Here, as described above, the action instruction generation unit 2120 adds a pair of separate manual action correction instructions 52, 53 output from the boom operating device 51, a pair of separate basic action instructions (not shown) corresponding to the boom operating device 51 in the basic action instruction 402 output from the basic action instruction unit 2119, and a pair of separate automatic action correction instructions (not shown) corresponding to the boom operating device 51 in the automatic action correction instruction 404 output from the learning unit 118, to generate a pair of separate action instructions 52', 53' corresponding to the boom operating device 51.

[0328] In addition, a pair of separate manual action correction instructions 55 and 56 output from the rotational operation device 54, a pair of separate basic action instructions (not shown) corresponding to the rotational operation device 54 in the basic action instruction 402, and a pair of separate automatic action correction instructions (not shown) corresponding to the rotational operation device 54 in the automatic action correction instruction 404 are added together to generate a pair of separate action instructions 55' and 56' corresponding to the rotational operation device 54.

[0329] In addition, a pair of separate manual action correction instructions 58, 59 output from the bucket operating device 57, a pair of separate basic action instructions (not shown) corresponding to the bucket operating device 57 in the basic action instruction 402, and a pair of separate automatic action correction instructions (not shown) corresponding to the bucket operating device 57 in the automatic action correction instruction 404 are added to generate a pair of separate action instructions 58', 59' corresponding to the bucket operating device 57.

[0330] In addition, a pair of separate manual action correction instructions 72, 73 output from the boom operating device 71, a pair of separate basic action instructions (not shown) corresponding to the boom operating device 71 in the basic action instruction 402, and a pair of separate automatic action correction instructions (not shown) corresponding to the boom operating device 71 in the automatic action correction instruction 404 are added to generate a pair of separate action instructions 72', 73' corresponding to the boom operating device 71.

[0331] In addition, the acceleration correction instruction 75 output from the acceleration device 50, the separate basic action instruction (not shown) corresponding to the acceleration device 50 in the basic action instruction 402, and the separate automatic action correction instruction (not shown) corresponding to the acceleration device 50 in the automatic action correction instruction 404 are added together to generate a separate action instruction 75' corresponding to the acceleration device 50.

[0332] <Motion Correction Command Generation Unit 2110>

[0333] Here, the action correction instruction generating unit 2110 adds a pair of separate manual action correction instructions 52, 53 output from the boom operating device 51 and a pair of separate automatic action correction instructions (not shown) corresponding to the boom operating device 51 in the automatic action correction instruction 404 output from the learning unit 118, and generates a pair of separate action correction instructions (not shown. Hereinafter referred to as boom action correction instructions) corresponding to the boom operating device 51.

[0334] In addition, a pair of separate manual action correction instructions 55, 56 output from the rotational operation device 54 and a pair of separate automatic action correction instructions (not shown) corresponding to the rotational operation device 54 in the automatic action correction instruction 404 are added to generate a pair of separate action correction instructions (not shown. Hereinafter, referred to as rotational action correction instructions) corresponding to the rotational operation device 54.

[0335] In addition, a pair of separate manual action correction instructions 58, 59 output from the bucket operating device 57 and a pair of separate automatic action correction instructions (not shown) corresponding to the bucket operating device 57 in the automatic action correction instruction 404 are added to generate a pair of separate action correction instructions (not shown. Hereinafter, referred to as bucket action correction instructions) corresponding to the bucket operating device 57.

[0336] In addition, a pair of separate manual action correction instructions 72, 73 output from the boom operating device 71 and a pair of separate automatic action correction instructions (not shown) corresponding to the boom operating device 71 in the automatic action correction instruction 404 are added to generate a pair of separate action correction instructions (not shown. Hereinafter, referred to as boom action correction instructions) corresponding to the boom operating device 71.

[0337] In addition, the acceleration correction instruction 75 output from the acceleration device 50 and the separate automatic action correction instruction (not shown) corresponding to the acceleration device 50 in the automatic action correction instruction 404 are added to generate a separate action correction instruction (not shown. hereinafter referred to as the acceleration action correction instruction) corresponding to the acceleration device 50.

[0338] These individual motion correction commands constitute the motion correction command Pm.

[0339] <Learning Data Storage Unit 115>

[0340] The motion correction command storage unit 2116 of the learning data storage unit 115 stores the motion correction command Pm in time series.

[0341] The prediction basic data storage unit 117 of the learning data storage unit 115 stores prediction basic data Pd including the work status data 212 output by the image processing unit 312 , the action state data 213 output by the microphone 313 , and the reaction data 214 output by the gyro sensor 314 in time series.

[0342] <Study Section 118>

[0343] As described above, when learning, the learning unit 118 performs machine learning on the learning data composed of the action correction instruction (hereinafter referred to as the learning action correction instruction) Pm' read from the action correction instruction storage unit 2116 and the prediction basic data (hereinafter referred to as the learning prediction basic data) Pd' read from the prediction basic data storage unit 17. After completing the machine learning, when the action unit 103 is in action, if the input of the prediction basic data Pd is received, the automatic action correction instruction 404 is output.

[0344] Next, the configuration of the learning unit 118 will be described in detail. Figure 19 Yes Figure 15 Schematic diagram illustrating the cycle time in the operation of the skill inheritance hydraulic excavator 20. Figure 20 Schematic diagram showing time series data of the motion correction command Pm, the learning motion correction command Pm′, the prediction base data Pd, and the learning prediction base data Pd′ in the skill-transferring hydraulic excavator 20 . Figure 21 It is a functional block diagram showing the configuration of the learning unit 118 .

[0345] <Time relationship between each time series data>

[0346] First, the temporal relationship between each time series data will be described.

[0347] Reference Figure 19 , the hydraulic excavator 20 repeatedly performs the prescribed operation multiple times. Although the time interval between the previous and next prescribed operations is not constant, in this embodiment, for the sake of convenience, the time from the start of the previous prescribed operation to the time when the next prescribed operation is about to start is referred to as the "cycle time". In addition, the time between the previous prescribed operation and the next prescribed operation serves as the time for the learning unit 118 to learn using the learning data in the "previous prescribed operation" (learning action correction instructions Pm' and learning prediction basic data Pd'). Hereinafter, this "time in between" will be referred to as the "learning time", and the time when the hydraulic excavator 20 performs the prescribed operation will be referred to as the "action time". In addition, the action of the hydraulic excavator 20 in the "action time" of each cycle time will be referred to as "~ action".

[0348] exist Figure 20 In the example, the "action" currently in progress is referred to as the "current action", and the "action" before that is referred to as the "previous action". In the currently ongoing action, the time series data Pm0, Pm1, Pm2, Pm3...Pmu (hereinafter referred to as Pm0~Pmu) of the action correction instruction Pm are obtained at a specified sampling interval. In addition, the time series data Pd0~Pdu of the prediction basic data Pd are obtained in the same manner. Moreover, the time series data Pm0~Pmu of the action correction instruction Pm obtained in the previous action become the time series data Pm0'~Pmu' of the learning action correction instruction Pm' in the current action. In addition, the time series data Pd0~Pdu of the prediction basic data Pd obtained in the previous action become the time series data Pd0'~Pdu' of the learning prediction basic data Pd' in the current action. In the following, the subscript numbers in each time series data indicate the order of the sampling time (interval time). Therefore, the time series data with the same subscript number refer to the data obtained at the same sampling time.

[0349] <Structure of the learning unit 118>

[0350] Reference Figure 21 The learning unit 118 includes, for example, a neural network 500 , a learning data and teacher data generating unit 501 , a data input unit 502 , and a learning evaluation unit 503 .

[0351] Neural network 500 includes an input layer, intermediate layers, and an output layer. The number of neurons in each layer is appropriately set. Known learning methods can be applied to learning neural network 500. Therefore, a brief description is provided here. Here, neural network 500 is, for example, a recursive neural network. The learning method is, for example, supervised learning.

[0352] The learning data and teacher data generating unit 501 generates time series data pn1 to pnu of teacher data pn based on time series data Pm0' to Pmu' of learning motion correction instructions Pm' and generates time series data Pd0' to Pdu-1' of learning data based on time series data Pd0' to Pdu' of learning prediction basic data Pd'.

[0353] The data input unit 502 sequentially inputs the time series data Pd0' to Pdu-1' of the learning data to each neuron of the input layer. At this time, if the data input unit 502 outputs the time series data Pdi of the learning data at a certain sampling time ti, the neural network 500 calculates the predicted action correction instruction Pni+1 at the next sampling time ti+1 through forward operation. Then, the learning evaluation unit 503 obtains the time series data pni+1 at the next sampling time ti+1 based on the time series data pn1 to pnu of the teacher data pn, and calculates the sum of the square errors e of the acceleration action correction instruction, the arm action correction instruction, the rotation action correction instruction, the bucket action correction instruction, and the boom action correction instruction for the predicted action correction instruction Pni+1 and the time series data pni+1 of the teacher data pn, for example. 2 Then, the learning evaluation unit 503 updates the weights of the neural network 500 through backward operation. The data input unit 502 and the learning evaluation unit 503 perform this process on all the time series data Pd0' to Pdu-1' of the learning data. For example, in all the processes, when the sum of the squared errors e 2 When the value falls below a predetermined threshold, the learning is terminated.

[0354] When this learning is complete, during the next operation of the hydraulic excavator 20, the data input unit 502 inputs the predicted basic data Pd0 at the current sampling time t0. The neural network 500 then outputs the predicted motion correction command Pn1 for the next sampling time t1 as the automatic motion correction command 404. Furthermore, an appropriate initial motion command is output as the initial value of the automatic motion correction command 404.

[0355] Thus, the learning result of the neural network 500 (learning unit 118 ) is reflected in the operation of the hydraulic excavator 20 .

[0356] Furthermore, when the input unit 502 inputs the time series data Pdi of the learning prediction base data Pd' at a certain sampling time ti, it can also input the previous time series data Pdi-1 to Pdi-n (n is a predetermined positive number). However, in this case, during the operation time, the data input unit 502 must input the past prediction base data Pdj-1 to Pdj-n at each sampling time tj, along with the current prediction base data Pdj. During the operation time, at each sampling time tj, the prediction base data Pd0 to Pdj-1 are stored in the prediction base data storage unit 117 in a time series. Therefore, the data input unit 502 reads this data from the prediction base data storage unit 117 and generates the past prediction base data Pdj-1 to Pdj-n. This improves the learning efficiency of the neural network 500. The reason is that when the operator predicts the movement of the bucket 18, he not only considers the instantaneous working conditions, the operating state of the hydraulic excavator 20 and the reaction from the working object at the current moment, but also considers a series of working conditions, the operating state of the hydraulic excavator 20 and the reaction from the working object before that, and predicts the next movement of the bucket 18, thereby accurately predicting the movement of the bucket 18.

[0357] Furthermore, information other than the working conditions, the operating state of the hydraulic excavator 20 , and the reaction from the working object may be used as learning data and input data of the operating time of the hydraulic excavator 20 .

[0358] [action]

[0359] Next, the operation of the skill transfer hydraulic excavator 20 configured as described above will be described. Hereinafter, the operation of the hydraulic excavator 20 will be described using as an example a case where the hydraulic excavator 20 performs a pit excavation operation as a predetermined operation.

[0360] Figure 22 Schematic diagram showing a situation of construction work performed by the skill-transfer hydraulic excavator 20 .

[0361] Reference Figure 22 First, a first camera 311 is installed above the planned location of the pit 131. The first camera 311 is mounted on, for example, a suitable support member installed on the ground. The first camera 311 is connected to the above-mentioned computing unit (learning unit 118) installed on the rotating body 15 of the hydraulic excavator 20, for example, via wireless communication, so that data can be communicated. The first camera 311 is located, for example, above the center of the pit 131 and is installed so that the optical axis 321 is directed toward the center of the pit 131.

[0362] Next, the hydraulic excavator 20 is set to an initial state. As the initial state, for example, the hydraulic excavator 20 is located at a setting location suitable for digging the pit 131 and takes an initial posture (for example, Figure 22 The installation position is set, for example, to a location where the hydraulic excavator 20 can dig pit 131 by shoveling soil with the bucket 18 without moving, and then discard the shoveled soil in the bucket 18 to a sand storage area. Here, the installation position is assumed to be midway between the planned location of pit 131 and the sand storage area. Furthermore, the first camera 311 is assumed to be a three-dimensional camera.

[0363] Then, the hydraulic excavator 20 is switched to the semi-autonomous mode.

[0364] <Digging work>

[0365] *No corrective action*

[0366] Next, the skilled operator performs the digging operation while performing corrective operations on the hydraulic excavator 20. Initially, the operator confirms the basic operation of the hydraulic excavator 20 without performing any corrective operations. This operation is hereinafter referred to as the initial operation. In this initial operation, the digging operation is performed by operating the operation unit 103 by outputting the basic operation instruction 402 from the basic operation instruction unit 2119. The control program for prescribing (implementing) the prescribed operation of the basic operation instruction unit 2119 assumes that the geology of the site where the digging is to take place (such as the composition and hardness of the foundation) is a prescribed geology. This digging operation is performed, for example, as follows.

[0367] First, the basic action instruction unit 2119 causes the bucket 18 to descend from the lowered position above the pit 131 and penetrate the ground at the predetermined location of the pit 131 (including the ground inside the pit 131) (penetration action). In this penetration action, before lowering the bucket 18, the bucket 18 is operated so that the claw at the end faces downward. Then, while pressing the penetrated bucket 18 against the ground, the bucket 18 is rotated forward to scoop up sand and soil (scooping action). Then, the bucket 18 is moved to the above-mentioned lowered position and lifted from the pit 131 (lifting action). Then, the rotating body 15 is rotated until the boom 16 faces the sand and soil storage location (forward rotation action). Then, the bucket 18 is rotated to the opposite side to discharge the sand and soil in the bucket 18 to the sand and soil storage location (sand discharge action). Then, the rotating body 15 is rotated in the opposite direction so that the bucket 18 is located at the above-mentioned lowered position (reverse rotation action). After that, when this series of actions is repeated a predetermined number of times, the digging operation is completed.

[0368] Meanwhile, during this period, data for learning is acquired, and after a predetermined task is completed, machine learning is performed by the learning unit 118 , which will be described in detail later.

[0369] *The first form has a correction operation*

[0370] Next, the operator performs a first type of correction operation as an example of the correction operation. Figure 23 (a) to (c) are cross-sectional views schematically showing a process in which the excavation work performed by the hydraulic excavator 20 is improved by a correction operation of the deep excavation work at the corner. Figure 23 In (a) to (c), the solid line represents the cross-sectional shape of the pit 131 actually formed, and the two-dot chain line represents the predetermined cross-sectional shape of the pit 131 envisioned in the control program for predetermined operation.

[0371] By performing the operation without correction in the initial operation, Figure 23 The cross-sectional pits left at the corners as shown in (a) are because deep excavation work at corners is difficult and cannot be adequately addressed by the control program used for the prescribed work.

[0372] In this case, the operator switches the operation mode to manual mode by operating the operation mode switching control, and manually operates the hydraulic excavator 20 to form the pit 131 into the specified cross-sectional shape. The hydraulic excavator 20 is then moved to the next planned pit location. The operation from this point onward is hereinafter referred to as the second operation.

[0373] Reference Figure 18 The operator switches the hydraulic excavator 20 to semi-automatic mode. The hydraulic excavator 20 then begins the basic motion described above. The operator manipulates the operating unit 101 as needed to modify the motion of the hydraulic excavator 20's motion unit 103. In particular, when deep-digging a corner of the pit 131, the basic motion is substantially fully modified.

[0374] In this manner, when the operator operates the operating unit 101, namely, the accelerator 50, the arm operating device 51, the rotation operating device 54, the bucket operating device 57, and the boom operating device 71, a manual motion correction command 403 is output, and the basic motion command 402 is corrected by this manual motion correction command 403. As a result, the basic motion of the operating unit 103 is corrected according to this correction. Meanwhile, this manual motion correction command 403 is added to the automatic motion correction command 404 output by the learning unit 118 to generate the motion correction command Pm.

[0375] <Judgment information for the next operation>

[0376] During this correction, the operator visually confirms the status of the digging, and visually identifies the postures of the bucket 18, the dipper arm 17, and the boom 16, thereby intuitively observing how the bucket 18, the dipper arm 17, the boom 16, and the driver's seat (rotating body 15) are currently moving. In addition, if the bucket 18 acts on the ground (digs the ground, scrapes into the soil, etc.), the reaction is felt with the body, thereby judging whether the current operation (action) is appropriate or not. Moreover, these are instantaneously considered to determine the next operation. Here, the so-called reaction refers to, for example, the tilt and vibration of the driver's seat (including excitation force, acceleration, angular acceleration, etc.). In addition, when deciding the next operation, the operator pays attention to the state of the engine 26 as the drive source (rotation amount, sound, etc.).

[0377] Here, the digging status is an example of information indicating the working status. The posture of the bucket 18, arm 17, and boom 16 is an example of information indicating the operating state of the hydraulic excavator 20. The tilt and vibration of the driver's seat (rotating body 15) are examples of reaction from the ground. The status of the engine 26 (rotation amount, sound, etc.) is information indicating the state of the drive source and, therefore, the operating state of the hydraulic excavator 20. Therefore, it is preferable to use this information as prediction base data Pd' based on the machine learning performed by the learning unit 118.

[0378] <Acquiring data for learning>

[0379] The motion correction command storage unit 2116 of the learning data storage unit 115 stores the motion correction command Pm corresponding to each of the above-mentioned operations.

[0380] On the other hand, in the series of operations of the above-mentioned predetermined work, the prediction basic data Pd is acquired as follows.

[0381] The first camera 311 captures the status of the digging work performed by the bucket 18, and the image processing unit 312 performs image processing on the captured image to generate data representing the status of the digging work as work status data 212, and the prediction basic data storage unit 117 of the learning data storage unit 115 stores the data.

[0382] Specifically, the first camera 311 primarily captures images of the arm 17, bucket 18, and pit 131 during the thrusting to lifting motion, and primarily captures images of pit 131 during the forward rotation to reverse rotation motion. The image processing unit 312 applies known image processing, such as edge processing, to the captured images to distinguish the areas of the arm 17 and bucket 18 from those of pit 131, thereby generating data on the planar shape and central depth of pit 131. In particular, images captured during the soil discharge motion do not contain the arm 17 and bucket 18 areas, but only the pit 131 area. Therefore, the planar shape and central depth of pit 131 are accurately determined in the operational status data 212 during the soil discharge motion.

[0383] The microphone 313 receives the sound of the engine 26 during the series of operations of the predetermined work. The prediction base data storage unit 117 stores the sound data (operation state data 213).

[0384] In particular, during the shoveling operation, for example, the operator increases the engine output and rotates the bucket 18. This allows for appropriate shoveling of sand and soil on the ground surface of the pit 131. Therefore, during the shoveling operation, the magnitude of the increased engine output is determined by the magnitude of the sound data.

[0385] During the series of actions in the aforementioned predetermined operation, the gyro sensor 314 detects the tilt and vibration (including excitation force, acceleration, angular acceleration, etc.) of the rotating body 15 (driver's seat) or the main body 102 as a reaction from the floor of the pit 131. The learning data storage unit 115 (more precisely, the prediction base data storage unit 117) stores this reaction data 214.

[0386] Especially, in the piercing action, the reaction from the ground when bucket 18 pierces the ground represents the hardness of the ground. Therefore, in the piercing action, the hardness of the ground that bucket 18 pierces is determined based on the size of the reaction data.

[0387] In this manner, in the present embodiment, data that can identify information suitable for determining the next operation is acquired as learning data.

[0388] Machine Learning

[0389] When a predetermined operation performed by hydraulic excavator 20 is completed, control unit 401 causes neural network 500 to perform machine learning using the learning data stored in learning data storage unit 115. At this time, as described above, when time series data Pdi representing learning data at each sampling time ti is input, the previous time series data Pdi-1 to Pdi-n (n is a predetermined positive number) are also input.

[0390] In the prescribed operation of the second action, if Figure 23 As shown in (b), while this is an improvement compared to the initial operation, the corner excavation depth is still insufficient. The operator switches the operation mode to manual mode by operating the operation mode switching control, manually operating the hydraulic excavator 20 to form the pit 131 into the specified cross-sectional shape. The hydraulic excavator 20 is then moved to the next planned pit location and reset to the initial state described above. The operation from this point onward is referred to as the third operation.

[0391] Reference Figure 18 , the operator switches the hydraulic excavator 20 to the semi-automatic mode. The hydraulic excavator 20 then outputs an automatic motion correction command 404 reflecting the learning result, and the basic motion command 402 is corrected by the automatic motion correction command 404. As a result, the motion of the motion unit 103 is corrected to reflect the correction operation in the second motion.

[0392] The operator further corrects the operation of the operation unit 103, which reflects the correction operation in the second operation, so that the shape of the pit 131 becomes the specified cross-sectional shape. As a result, the operation correction instruction 403 corresponding to the further correction operation is output. The basic operation instruction 402 corrected by the automatic operation correction instruction 404 is further corrected by the manual operation correction instruction 403. As a result, in the specified operation in the third operation, as shown in FIG. Figure 23 As shown in (c), a pit 131 having a predetermined cross-sectional shape and having a sufficiently deep corner is formed.

[0393] On the other hand, the manual motion correction command 403 corresponding to the further additional correction operation is added to the automatic motion correction command 404 reflecting the learning result of the correction operation in the second motion to generate the motion correction command Pm.

[0394] The learning unit 118 then mechanically learns this motion correction command Pm during subsequent learning. The learning unit 118 then outputs an automatic motion correction command 404 reflecting this motion correction command Pm. Thus, the basic motion command 402 outputted from the basic motion command unit 2119 is corrected by the automatic motion correction command 404, which reflects the correction operations accumulated during the second and third motions. As a result, the motion of the motion unit 103 is corrected to reflect the correction operations during the second and third motions. Consequently, subsequent improvements through the operator's manipulation of the operating unit 101 are no longer necessary.

[0395] Thus, according to the hydraulic excavator 20 of this embodiment, the operator's manual operation correction instructions 403 are accumulated in the learning unit 118, so that the operator's skills (here, digging work) are eventually passed on to the learning unit 118. In addition, the learning unit 118 learns through actual work, so the learning period is short.

[0396] *The second form has a correction operation*

[0397] Next, a case where the operator performs a second type of correction operation will be exemplified.

[0398] Figure 24 (a) to (c) are cross-sectional views schematically showing a process in which the excavation work performed by the hydraulic excavator 20 is improved by a correction operation corresponding to the geology of the planned excavation site. Figure 24 In (a) of FIG. 1 , the dotted line represents the cross-sectional shape of the pit 131 when it is assumed that the operator does not perform a correction operation.

[0399] Reference Figure 24 In (a), if the geology of the site where pit 131 is to be excavated is softer (easier to dig) than the specified geology anticipated by the control program for the specified operation, if the actuator 103 of the hydraulic excavator 20 performs basic operations, a pit 131 having a shape deeper than the specified cross-sectional shape is formed, as shown by the dotted line. However, here, a skilled operator performs correction operations on the hydraulic excavator 20 in semi-automatic mode while digging the pit. As a result, pit 131 having the specified cross-sectional shape is formed. This correction operation is then machine-learned by the learning unit 118. Therefore, if the geology of the site where pit 131 is to be excavated is the same as that of the current operation, the operator does not perform correction operations, and the hydraulic excavator 20, under the control of the control unit 401, digs the pit with the specified cross-sectional shape.

[0400] However, for example, if the geology of the planned site for digging the pit 131 is harder (difficult to dig) than the specified geology expected by the control program for the specified operation, the predicted basic data obtained by the learning unit 118 during the operation (for example, the tilt and vibration of the driver's seat when the bucket 18 penetrates the ground, and the shape of the pit 131 captured by the first camera 311 during digging) are different from the geology of the planned site for digging the pit 131 being softer than the specified geology. Therefore, the learning unit 118 outputs, for example, an automatic action correction instruction 404 with a correction amount of zero. Therefore, if the operator does not perform any correction operation, the action unit 103 performs basic actions according to the basic action instruction, such as Figure 24 As shown in (b), for example, a pit 131 having a cross-sectional shape shallower than a predetermined cross-sectional shape is formed.

[0401] However, in reality, a skilled operator will appropriately deal with the hard ground and dig the pit while performing correction operations on the operation of the operation unit 103. Figure 24 As shown in (c), a pit 131 having a predetermined cross-sectional shape is formed. This correction operation is then machine-learned by the learning unit 118. Therefore, if the geological conditions at the planned site for pit 131 are of the same hardness as those during this operation, the operator does not perform a correction operation, and the hydraulic excavator 20 excavates a pit having the predetermined cross-sectional shape under the control of the control unit 401. Therefore, if the geological conditions at the planned site for pit 131 are of the same hardness as those during this operation, the operator does not perform a correction operation, and the hydraulic excavator 20 excavates a pit having the predetermined cross-sectional shape under the control of the control unit 401.

[0402] In this way, although there are multiple forms of the basic action of the action unit 103 that require correction, if a learning unit 118 that performs mechanical learning is used as in the hydraulic excavator 20, each time a phenomenon occurs in which the basic action of the action unit 103 requires correction, the operator's skills can be easily passed on by having the learning unit 118 learn the manual action correction instruction 403 (to be precise, the action correction instruction Pm) corresponding to the form.

[0403] Furthermore, in the hydraulic excavator 20, the basic motions of the motion unit 103 related to a given task that do not require correction are automatically executed by the basic motion command unit 2119. Therefore, the operator only needs to make necessary corrections. This reduces the burden on the operator. Furthermore, even skilled operators experience variations in their work. Therefore, performing only a portion of the work through operator control improves work accuracy compared to performing all work through operator control.

[0404] Summary

[0405] As described above, according to the fourth embodiment, it is possible to provide a construction machine 200 that allows an operator to perform correction operations on the basic motion of the working unit 104 based on the basic motion instruction unit 2119 to learn the work performed by the construction machine 200 and automatically perform the learned work.

[0406] Thus, it is possible to provide the skill-transferring construction machine 200 that can transfer the skills of skilled workers in the construction industry and can automate predetermined work in a short period of time.

[0407] Furthermore, according to the fourth embodiment, by adding a computing unit constituting the control unit 401 and a detector for acquiring prediction base data to an existing hydraulic excavator, the existing hydraulic excavator can be converted into the skill transfer hydraulic excavator 20 of the present invention.

[0408] Furthermore, by referring to the description of the fourth embodiment, the present invention can be easily applied to construction machines other than the hydraulic excavator 20 .

[0409] (Implementation 5)

[0410] The fifth embodiment of the present invention illustrates an embodiment in which the operating state detection unit 113 further includes a second imaging device 331 and an image processing unit 333 in the hydraulic excavator 20 according to the fourth embodiment.

[0411] Figure 25 It is a side view showing the hardware configuration of the skill transfer hydraulic excavator 20 according to the fifth embodiment. Figure 26 Yes Figure 25 A functional block diagram of the structure of a control system of a skill inheritance hydraulic excavator 20.

[0412] Reference Figure 25 as well as Figure 26 In the hydraulic excavator 20 of this embodiment, the operating state detection unit 113 further includes a second imaging device 331 and an image processing unit 333. The remaining configuration is the same as that of the hydraulic excavator 20 of the fourth embodiment.

[0413] The second camera 331 captures the entire hydraulic excavator 20. The image captured by the second camera 331 is processed by the image processing unit 333 to obtain data representing the posture of the hydraulic excavator 20. The image processing unit 333 then outputs the image processing unit 333 as the operating state data 213. The optical axis 332 of the second camera 331 is directed toward the hydraulic excavator 20.

[0414] The second imaging device 331 is constituted by, for example, a common digital camera and is fixed to a fixed object (such as the ground) different from the vehicle of the hydraulic excavator 20 via an appropriate supporting member, or is mounted on a drone.

[0415] The image processing unit 333 processes the image captured by the second imaging device 331, extracts the external shape of the hydraulic excavator 20, and outputs the external shape data as the operation state data 213. The external shape data is posture data that can identify the posture of the operation unit 103 of the hydraulic excavator 20.

[0416] According to embodiment 5, in the series of actions described in embodiment 4, the posture of the hydraulic excavator 20 is determined by the outer shape of the hydraulic excavator 20 in the action status data 213, so the efficiency of the mechanical learning of the learning unit 118 is improved, and the automatic action correction instruction 404 output by the learning unit 118 becomes more appropriate when the action unit 103 is in action.

[0417] (Implementation 6)

[0418] The sixth embodiment of the present invention illustrates an embodiment in which the operating state detection unit 113 further includes a sensor unit 341 in the hydraulic excavator 20 according to the fourth embodiment.

[0419] Figure 27 It is a side view showing the hardware configuration of a skill-transfer hydraulic excavator 20 according to the sixth embodiment. Figure 28 Yes Figure 27 Functional block diagram of the control system structure of the skill inheritance hydraulic excavator 20.

[0420] Reference Figure 28 In the hydraulic excavator 20 of this embodiment, the operating state detection unit 113 further includes a sensor unit 341. The remaining configuration is the same as that of the hydraulic excavator 20 of the fourth embodiment. The sensor unit 341 is composed of sensors S1 to S4.

[0421] Reference Figure 27 The hydraulic excavator 20 is equipped with sensors S1 to S4. Specifically, the rotating body 15 is provided with a sensor S1 that detects the rotation angle of the rotating body 15 around the rotation axis A1. The base end of the boom 16 is provided with a sensor S2 that detects the rotation angle of the boom 16 around the rotation axis A2. The base end of the arm 17 is provided with a sensor S3 that detects the rotation angle of the arm 17 around the rotation axis A3. The distal end of the arm 17 is provided with a sensor S4 that detects the rotation angle of the bucket 18 around the rotation axis A4.

[0422] The sensor unit 341 outputs data on the rotation angles detected by the sensors S1 to S4 as the operation state data 213. The combination of the data on the rotation angles detected by the sensors S1 to S4 is posture data capable of specifying the posture of the operation unit 103 of the hydraulic excavator 20.

[0423] According to embodiment 6, in the series of actions described in embodiment 4, the posture of the hydraulic excavator 20 is determined by the combination of the rotation angles respectively detected by sensors S1 to S4 in the action state data 213, so the efficiency of the mechanical learning of the learning unit 118 is improved, and when the action unit 103 is in action, the automatic action correction instruction 404 output by the learning unit 118 becomes more appropriate.

[0424] (Implementation 7)

[0425] The embodiment of the present invention exemplifies a configuration in which, in any one of the hydraulic excavators 20 of the fourth to sixth embodiments, the operating device generates a pilot pressure, and the control valve is controlled by the pilot pressure.

[0426] Figure 29This is a functional block diagram showing the configuration of an operation command generation unit of a skill-transfer hydraulic excavator, which is an example of a skill-transfer construction machine according to a seventh embodiment of the present invention.

[0427] Reference Figure 18 as well as Figure 29 In the seventh embodiment, the arm operating device 51, the swing operating device 54, the bucket operating device 57, and the boom operating device 71 of the operating unit 101 output a pair of individual manual operation correction commands (52, 53), (55, 56), (58, 59), and (72, 73) of the pilot pressure, respectively. Furthermore, the arm control valve 40, the swing control valve 43, the bucket control valve 44, and the boom control valve 47 are control valves controlled by the pilot pressure.

[0428] On the other hand, the operation command generating unit 2120 includes an adder 2120a, a pressure / electricity converter 2120b, and an electricity / pressure converter 2120c.

[0429] The pressure / electric converter 2120b is composed of, for example, four pairs of piezoelectric elements, which convert four pairs of separate manual action correction instructions (52, 53), (55, 56), (58), (72, 73) of the pilot pressure signals output from the boom operating device 51, the rotation operating device 54, the bucket operating device 57 and the arm operating device 71 of the operating unit 101 into four pairs of separate manual action correction instructions (52, 53), (55, 56), (58, 59), (72, 73) of electrical signals, and output these to the adder 2120a.

[0430] Adder 2120a operates similarly to motion command generator 2120 of Embodiment 4. Specifically, it adds the four pairs of individual basic motion commands and the four pairs of individual automatic motion correction commands corresponding to the four pairs of individual manual motion correction commands (52, 53), (55, 56), (58, 59), and (72, 73) to generate four pairs of individual motion commands. These four pairs of individual motion commands are then output to power / pressure converter 2120c.

[0431] The electric / pressure converter 2120c is composed of 4 pairs of multi-control valves including electromagnetic proportional valves, which convert the 4 pairs of separate action instructions into 4 pairs of separate action instructions (52', 53'), (55', 56'), (58', 59'), (72', 73') of pilot pressure signals, and output them to the boom control valve 40, the rotation control valve 43, the bucket control valve 44 and the arm control valve 47 respectively.

[0432] In addition, 4 pairs of individual manual action correction instructions (52, 53), 55, 56), (58, 59), (72, 73) constitute part of the manual action correction instruction 403, and 4 pairs of individual action instructions (52', 53'), (55', 56'), (58', 59'), (72', 73') constitute part of the action instruction 201.

[0433] According to the seventh embodiment, the present invention can be applied to the hydraulic excavator 20 in which the operating device generates a manual operation correction command of the pilot pressure and the control valve is controlled by the operation command of the pilot pressure.

[0434] In addition, according to embodiment 7, by adding a computing unit constituting the control unit 401, a detector for obtaining prediction basic data, a piezoelectric element constituting the pressure / electricity converter 2120b, and a multi-control valve constituting the electricity / pressure converter 2120c to the existing hydraulic excavator, the existing hydraulic excavator can be converted into the skill inheritance hydraulic excavator 20 of the present invention.

[0435] (Other embodiments)

[0436] The hydraulic excavator 10 according to the second embodiment may further include the sensor unit 341 according to the third embodiment in the operating state detection unit 113 .

[0437] In Embodiments 1 to 3, the operating state detection unit 113 may include an operating sensor for detecting the operation of each control valve 40, 43, 44, 47, instead of the microphone 313, or in addition to the microphone 313. Examples of the operating sensor include a pressure sensor provided in an oil passage connected to each control valve 40, 43, 44, 47, and a position sensor for detecting the position of the valve body of each control valve 40, 43, 44, 47.

[0438] In Embodiments 1 to 3, the operating state detection unit 113 may include a rotational speed sensor that detects the rotational speed of the engine 26 and outputs data on the detected rotational speed (drive source state data), instead of or in addition to the microphone 313. Examples of the rotational speed sensor include a rotary encoder, a tachometer, and a tachometer.

[0439] In addition, in the first to third embodiments, the learning method of the learning unit 118 may be teacher-less learning.

[0440] In addition, in Embodiments 1 to 3, the driving source of the hydraulic excavator 10 may be a driving source other than the engine 26. As such a driving source, an electric motor is exemplified.

[0441] The hydraulic excavator 20 according to the fifth embodiment may further include the sensor unit 341 according to the sixth embodiment in the operating state detection unit 113. In addition, the hydraulic excavator 20 may also be corrected as in the seventh embodiment.

[0442] In any of the fourth to seventh embodiments, the operation state detection unit 113 may include an operation sensor for detecting the operation of each control valve 40 , 43 , 44 , 47 instead of or in addition to the microphone 313 .

[0443] In any of Embodiments 4 to 7, the operating state detection unit 113 may include a rotational speed sensor that detects the rotational speed of the engine 26 and outputs data on the detected rotational speed (drive source state data), instead of or in addition to the microphone 313. Examples of the rotational speed sensor include a rotary encoder, a tachometer, and a tachometer.

[0444] In any of the fourth to seventh embodiments, the learning method of the learning unit 118 may be teacher-less learning.

[0445] In any of the fourth to seventh embodiments, the driving source of the hydraulic excavator 20 may be a driving source other than the engine 26. An electric motor is exemplified as such a driving source.

[0446] In any of the fourth to sixth embodiments, the operation command generating unit 2120 may be configured to apply hydraulic pilot pressure signals to each other.

[0447] Numerous modifications and other embodiments will be apparent to those skilled in the art based on the above description, and therefore, the above description should be interpreted as illustrative only.

[0448] Industrial application possibilities

[0449] The construction machine with a learning function according to the present invention is useful as a construction machine capable of learning work performed by a human operator and automatically performing the learned work.

[0450] Furthermore, the skill-transferring construction machine of the present invention is useful as a skill-transferring construction machine capable of transmitting the skills of skilled workers in the construction industry and automating predetermined operations in a short period of time.

[0451] Description of Reference Numerals

[0452] 1...Hydraulic drive system; 8...Flow control device; 10...Hydraulic excavator with learning function (hydraulic excavator); 11...Boom cylinder; 12...Arm cylinder; 13...Bucket cylinder; 14...Swing motor; 15...Swing unit; 16...Boom; 17...Arm; 18...Bucket; 19...Traveling unit; 20...Hydraulic excavator; 21...First main pump; 22...First flow control device; 23...Second main pump; 24...Second flow control device; 25...Auxiliary pump; 40...Arm control valve; 43...Swing control valve; 44...Bucket control valve; 47...Boom control valve; 50...Accelerator; 5 1...Arm operating device; 54...Rotation operating device; 57...Bucket operating device; 71...Boom operating device; 81-86, 91, 92...Pressure sensor; 100...Construction machinery with learning function (construction machinery); 101...Operating unit; 102...Main unit; 103...Actuating unit; 104...Working unit; 105...Hydraulic drive system; 106...Hydraulic circuit; 107...Pump unit; 110...Action command detection unit; 112...Working status detection unit; 113...Action state detection unit; 114...Reaction detection unit; 115...Learning data storage unit; 116...Command data storage unit; 117... 7...Prediction Basic Data Storage Unit; 118...Learning Unit; 119...Operation Unit Driver; 120...Prediction Action Command Converter; 131...Pit; 200...Skill Inheritance Construction Machinery; 201...Action Command; 202...Driving Force; 211, 211'...Command Data; 212...Working Status Data; 213...Action Status Data; 214...Reaction Data; 311...First Camera; 312...Image Processing Unit; 313...Microphone; 314...Gyro Sensor; 321...Optical Axis; 331...Second Camera; 332...Optical Axis; 333...Image Processing Unit; 341...Transmission Sensor unit; 1400...Neural network; 1401...Learning data and teacher data generation unit; 1402...Data input unit; 1403...Learning evaluation unit; 500...Neural network; 501...Learning data and teacher data generation unit; 502...Data input unit; 503...Learning evaluation unit; 1000...Construction machine with learning function; 1101...Command data generation unit; 1102...Command storage unit; 1103...Prediction command; 2110...Motion correction command generation unit; 2116...Motion correction command storage unit; 2119...Basic motion command unit; 2120...Motion command generation unit; A1-A4...The first to fourth rotation axes; M1 to M5...servo motors; Pd...prediction basic data; Pd'...learning prediction basic data; Pf...prediction action command; Pf'...learning command data; S1 to S4...sensors.

Claims

1. A construction machine with a learning function, wherein: have: an action unit having an operating unit and causing the operating unit to operate in a manner of performing an operation; an operating unit that outputs an instruction corresponding to an operator's operation; a work status detecting unit that detects a status of the work performed by the work unit and outputs the detected work status as work status data; an action state detecting unit that detects an action state of the action unit and outputs the detected action state as action state data; a reaction detection unit that detects a reaction received by the action unit from a work object due to the work of the work unit and outputs the detected reaction as reaction data; a learning data storage unit that stores the instructions as instruction data in a time series, and stores prediction basic data including the work status data, the operation state data, and the reaction data in a time series; a learning unit that performs machine learning on the instruction data stored in the learning data storage unit using the prediction base data stored in the learning data storage unit, and, after completing the machine learning, receives input of the prediction base data when the action unit operates, and outputs a predicted instruction of the instruction; as well as a hydraulic drive system that drives the motion unit based on the command, the predicted command, or the command and the predicted command, The operation unit is configured to output, as the command, an action command corresponding to the operator's operation. The learning data storage unit is configured to store the motion instructions as instruction data in a time series, and to store prediction basic data including the work status data, the motion state data, and the reaction data in a time series. The learning unit is configured to perform machine learning on the instruction data stored in the learning data storage unit using the prediction basic data stored in the learning data storage unit during learning, and to receive input of the prediction basic data and output a prediction action instruction as the prediction instruction when automatic control after the machine learning is completed. The hydraulic drive system is configured to drive the motion unit according to the motion command or the predicted motion command. A main body portion is provided with the action portion, The reaction detection unit detects the reaction including at least one of the inclination, acceleration, and angular acceleration of the operation unit or the main body, and outputs the detected reaction as the reaction data.

2. The construction machine with a learning function according to claim 1, wherein: The operation state detection unit includes a drive source state detection unit that detects a state of a drive source including at least one of an output of a drive source that drives a pump that pressurizes the working oil of the hydraulic drive system and an operation sound, and outputs the detected state of the drive source as drive source state data. The operation status data includes the driving source status data.

3. The construction machine with a learning function according to claim 1 or 2, wherein: The operation status data includes the instruction data.

4. The construction machine with a learning function according to claim 1 or 2, wherein: The motion state detection unit further includes a posture detection unit that detects the posture of the motion unit and outputs the detected posture as posture data. The action state data includes the posture data.

5. A construction machine with a learning function, wherein: have: an action unit having an operating unit and causing the operating unit to operate in a manner of performing an operation; an operating unit that outputs an instruction corresponding to an operator's operation; a work status detecting unit that detects a status of the work performed by the work unit and outputs the detected work status as work status data; an action state detecting unit that detects an action state of the action unit and outputs the detected action state as action state data; a reaction detection unit that detects a reaction received by the action unit from a work object due to the work of the work unit and outputs the detected reaction as reaction data; a learning data storage unit that stores the instructions as instruction data in a time series, and stores prediction basic data including the work status data, the operation state data, and the reaction data in a time series; a learning unit that performs machine learning on the instruction data stored in the learning data storage unit using the prediction base data stored in the learning data storage unit, and, after completing the machine learning, receives input of the prediction base data when the action unit operates, and outputs a predicted instruction of the instruction; and a hydraulic drive system that drives the motion unit based on the command, the predicted command, or the command and the predicted command, The construction machine with a learning function is a skill-transmitting construction machine equipped with a control unit. The skill-transmitting construction machine is a construction machine capable of transmitting the skills of skilled operators in the construction industry. The operation unit is configured to output, as the command, a manual operation correction command corresponding to the operator's operation. The hydraulic drive system is configured to drive the motion unit according to the basic motion instruction, the automatic motion correction instruction and the manual motion correction instruction, and The control unit includes: a basic motion instruction unit configured to output the basic motion instruction for causing the operation unit to perform a basic motion through the operation unit; an action correction instruction generating unit for generating an action correction instruction by adding the manual action correction instruction to the automatic action correction instruction; The learning data storage unit includes a motion correction instruction storage unit that stores the motion correction instructions as the instruction data in a time series, and a prediction basic data storage unit that stores the prediction basic data in a time series; as well as The learning department, The learning unit is configured to use the prediction basic data stored in the prediction basic data storage unit to mechanically learn the action correction instruction stored in the action correction instruction storage unit. After completing the mechanical learning, the learning unit receives the input of the prediction basic data when the action unit is in action, and outputs the automatic action correction instruction as the prediction instruction.

6. The construction machine with a learning function according to claim 5, wherein: A main body portion is provided with the action portion, The reaction detection unit detects the reaction including at least one of the inclination, acceleration, and angular acceleration of the operation unit or the main body, and outputs the detected reaction as the reaction data.

7. The construction machine with a learning function according to claim 5 or 6, wherein: The operation state detection unit includes a drive source state detection unit that detects a state of a drive source including at least one of an output of a drive source that drives a pump that pressurizes the working oil of the hydraulic drive system and an operation sound, and outputs the detected state of the drive source as drive source state data. The operation status data includes the driving source status data.

8. The construction machine with a learning function according to claim 5 or 6, wherein: The motion state detection unit further includes a posture detection unit that detects the posture of the motion unit and outputs the detected posture as posture data. The action state data includes the posture data.

9. The construction machine with a learning function according to claim 5 or 6, wherein: The manual action correction instruction is an electrical instruction signal, The action unit includes: a hydraulic actuator that drives the working unit; and a control valve that hydraulically controls the action of the hydraulic actuator according to the basic action instruction, the automatic action correction instruction, and the manual action correction instruction. The control valve is a solenoid valve.

Images