Free-section drilling machine system

Abstract

We provide an automated, free-section excavator. [Solution] The free-section excavator system comprises a free-section excavator 10 that excavates an excavation target using a cutting drum that moves in response to an operation signal, a group of sensors 30 that acquires information about the excavation target, and a control device 20 that outputs an operation signal based on the information acquired by the group of sensors 30 and the excavation target information. The control device 20 comprises a first trained model 21 that outputs information about the state of the excavation target based on the information acquired by the group of sensors 30, a second trained model 22 that outputs a plan of movement including the path and speed of movement of the cutting drum based on the information about the state of the excavation target output from the first trained model 21 and the excavation target information, and an operation signal creation unit 27 that outputs an operation signal based on the path plan and the operation plan.

Description

Technical Field

[0001] The present invention relates to a full-face tunneling machine system.

Background Art

[0002] A full-face tunneling machine for excavating a face mainly in a mine such as a coal mine or a tunnel is known. Excavation work in a mine is work in a harsh environment such as dust and vibration, and is also work accompanied by danger. For this reason, a tunneling machine that does not require an operator at the work site is required. For example, Patent Document 1 discloses a technique related to the automatic operation of a hydraulic excavator.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] An object of the present invention is to provide a full-face tunneling machine capable of automatic operation.

Means for Solving the Problems

[0005] According to one aspect of the present invention, a free-section excavator system comprises a free-section excavator configured to excavate an object to be excavated using a cutting drum that moves in response to an operation signal; a sensor configured to acquire information about the object to be excavated; and a control device configured to output the operation signal based on the information acquired by the sensor and excavation target information including the target amount of excavation of the object to be excavated. The control device comprises a first trained model obtained by machine learning, configured to output information about the state of the object to be excavated based on the information acquired by the sensor; a second trained model obtained by machine learning, configured to output a plan for the movement path of the cutting drum and a plan for the operation of the cutting drum based on the information about the state of the object to be excavated output from the first trained model and the excavation target information; and an operation signal generation unit configured to output the operation signal based on the plan for the path and the plan for the operation. [Effects of the Invention]

[0006] According to the present invention, it is possible to provide a free-section excavator capable of automatic operation. [Brief explanation of the drawing]

[0007] [Figure 1A] Figure 1A is a schematic front view showing an example of the configuration of a free-section excavator system according to one embodiment. [Figure 1B] Figure 1B is a top view showing a schematic example of the configuration of a free-section excavator system. [Figure 2] Figure 2 is a block diagram illustrating a schematic configuration of the various components related to the operation of a free-section excavator system. [Figure 3] Figure 3 is a schematic diagram showing the rock mass to be drilled. [Figure 4] Figure 4 is a functional block diagram showing an example configuration of the control device for the free-section excavator system related to the automatic excavation operation. [Figure 5] Figure 5 is a flowchart illustrating an example of the operation of an automated drilling system with a free-section drilling machine. [Figure 6] Figure 6 illustrates an example of creating training data for building the first trained model. [Figure 7] Figure 7 illustrates an example of deep reinforcement learning for creating a second pre-trained model. [Figure 8] Figure 8 illustrates an example of the limitations on path planning using the second trained model. [Figure 9] Figure 9 illustrates an example of the limitations on path planning using the second pre-trained model. [Modes for carrying out the invention]

[0008] One embodiment will be described with reference to the drawings. This embodiment relates to a free-section excavator used for excavating a working face in a mine or tunnel. More specifically, this embodiment relates to a free-section excavator system that enables the automatic operation of a free-section excavator.

[0009] [System Configuration] Figure 1A is a schematic front view showing an example configuration of the free-section excavator system 1 of this embodiment. Figure 1B is a schematic top view showing an example configuration of the free-section excavator system 1 of this embodiment. Figure 2 is a block diagram illustrating a schematic configuration of each part related to the operation of the free-section excavator system 1 of this embodiment. The free-section excavator system 1 comprises a free-section excavator 10 and a control device 20 that controls the operation of the free-section excavator 10.

[0010] The free-section excavator 10 includes a mobile body 110, a boom 120 that extends forward from the body 110 and whose tip position can be changed, and a cutting drum 141 that is provided at the tip of the boom 120 and configured to excavate the excavation face.

[0011] The main body 110 includes a travel unit 111, outriggers 112, and a dozer 113. The travel unit 111 is provided so that the free-section excavator 10 can be moved by rotating the left and right crawlers with a travel hydraulic motor. The outriggers 112 are provided as a pair on the left and right at the rear of the main body 110. The outriggers 112 can move up and down, for example by extending and retracting a hydraulic cylinder. The outriggers 112 are configured to press down on the ground during excavation work to stabilize the posture of the main body 110. The dozer 113 is provided at the front lower part of the main body 110. The dozer 113 is configured so that it can push aside and remove the cutting spoil generated when the face is excavated using the cutting drum 141. The travel unit 111, outriggers 112, and dozer 113 operate under the control of the control device 20 and based on operation signals output from the control device 20. Furthermore, each part is equipped with the necessary sensors, and the information acquired by these sensors is input to the control device 20 as sensor information.

[0012] Furthermore, the free-section excavator 10 is equipped with multiple hydraulic drive devices, such as hydraulic cylinders, for the operation of each part. The free-section excavator 10 is also equipped with an electrically operated hydraulic supply unit 114 that supplies hydraulic pressure to these hydraulic drive devices. In addition, each part of the free-section excavator 10 is operated by power supplied from the outside via a power cable. The power cable is stored wound around a cable reel 115 located at the rear of the main body 110. The power cable can be unwound from the cable reel 115 automatically or manually.

[0013] Furthermore, the free-section excavator 10 is equipped with an operating unit 116 that includes levers, switches, pedals, etc., for operation by the operator. Manual operation information regarding the operator's actions acquired by the operating unit 116 is input to the control device 20. The control device 20 may control the operation of each part of the free-section excavator 10 based on this manual operation information. The control device 20 can also use this manual operation information for various analyses.

[0014] The boom 120 is connected to the main body 110 via a swivel section 121. The swivel section 121 is configured to rotate the boom 120 horizontally with respect to the main body 110. The swivel section 121 has a swivel drive section 122 including a hydraulic cylinder for swiveling, a hydraulic circuit, a drive mechanism, etc. configured to swivel the boom 120. Further, the swivel section 121 includes a swivel stroke sensor 123 for detecting the stroke of the hydraulic cylinder for swiveling and a swivel hydraulic pressure gauge 124 for measuring the hydraulic pressure. The swivel drive section 122 operates based on a swivel operation signal output from the control device 20 under the control of the control device 20. Also, the information acquired by the swivel stroke sensor 123 and the swivel hydraulic pressure gauge 124 is input to the control device 20 as sensor information.

[0015] Further, the boom 120 includes a vertical drive section 126 configured to move the boom 120 in a pitching motion to move its tip up and down. The vertical drive section 126 includes a hydraulic cylinder for vertical movement, a hydraulic circuit, a drive mechanism, etc. configured to move the boom 120 vertically. Also, the boom 120 is provided with a vertical stroke sensor 127 for detecting the stroke of the hydraulic cylinder for vertical movement and a vertical hydraulic pressure gauge 128 for measuring the hydraulic pressure. The vertical drive section 126 operates based on a vertical operation signal output from the control device 20 under the control of the control device 20. Also, the information acquired by the vertical stroke sensor 127 and the vertical hydraulic pressure gauge 128 is input to the control device 20 as sensor information.

[0016] In addition, the boom 120 includes a telescopic drive unit 131 configured to be able to extend and retract the boom 120 and move its tip back and forth. The telescopic drive unit 131 includes a hydraulic cylinder for telescoping the boom 120, a hydraulic circuit, a drive mechanism, and the like. Further, the boom 120 is provided with a telescopic stroke sensor 132 for detecting the stroke of the hydraulic cylinder for telescoping and a telescopic hydraulic pressure gauge 133 for measuring the hydraulic pressure. The telescopic drive unit 131 operates under the control of the control device 20 based on the telescopic operation signal output from the control device 20. Also, the information acquired by the telescopic stroke sensor 132 and the telescopic hydraulic pressure gauge 133 is input to the control device 20 as sensor information.

[0017] By combining the operations of the swing drive unit 122, the vertical drive unit 126, and the telescopic drive unit 131, the full face tunneling machine 10 can move the position of the cutting drum 141 provided at the tip of the boom 120 three-dimensionally freely in the front-rear, up-down, and left-right directions within a predetermined range in front of the main body 110. Note that the operations of the swing drive unit 122, the vertical drive unit 126, and the telescopic drive unit 131 may be performed by other means instead of hydraulic cylinders. In that case, appropriate sensors for acquiring the necessary sensor information are provided for each unit.

[0018] The full face tunneling machine 10 includes a cutting device 140 including the cutting drum 141 for cutting. The cutting drum 141 has, for example, a configuration in which a plurality of cutting bits are arranged on a substantially hemispherical drum. The cutting drum 141 is configured to cut the face while rotating. The cutting device 140 has a cutting motor 142 that is a drive source of the cutting drum 141 and a speed reducer 143 provided between the cutting motor 142 and the cutting drum 141. The speed reducer 143 may be configured to be able to switch the reduction ratio. Further, the cutting device 140 includes a cutting ammeter 144 for measuring the current flowing through the cutting motor 142. The cutting motor 142 and the speed reducer 143 operate under the control of the control device 20 based on the cutting operation signal output from the control device 20. Also, the information acquired by the cutting ammeter 144 is input to the control device 20 as sensor information.

[0019] Furthermore, the free-section excavator 10 is equipped with various sensors, such as a camera 151, a triaxial inclinometer 152, a machine vibration meter 153, and a microphone 154. The camera 151 may have multiple cameras. The camera 151 is configured to capture images of the area in front of the free-section excavator 10, including at least the area to be excavated. The camera 151 may be configured to capture images in all directions around the free-section excavator 10 using multiple cameras. It is preferable that the camera 151 includes an infrared camera so that images can be obtained even in dusty environments. The image information obtained by the camera 151 is input to the control device 20.

[0020] The triaxial inclinometer 152 is configured to detect the posture of the free-section excavator 10. The posture information obtained by the triaxial inclinometer 152 is input to the control device 20. Based on the information obtained from the triaxial inclinometer 152, the control device 20 can obtain roll, pitch, and yaw inclination information of the main body 110. The machine body vibrometer 153 is configured to detect vibrations of the free-section excavator 10. The vibration information obtained by the machine body vibrometer 153 is input to the control device 20. The microphone 154 acquires sound from around the free-section excavator 10. The sound information obtained by the microphone 154 is input to the control device 20. The information acquired by these various sensors is used by the control device 20 to control the free-section excavator 10.

[0021] Furthermore, the free-section excavator system 1 is equipped with a LiDAR 191. The LiDAR 191 is installed, for example, on the outside of the main body 110 inside the work site, and can acquire information on the position and shape of each part of the entire work site, including the excavation target and the free-section excavator 10. The information acquired by the LiDAR 191 is input to the control device 20. In addition, the free-section excavator system 1 may be equipped with an overhead camera 192 installed inside the work site, separate from the imaging device 151 provided on the main body 110, and capable of photographing the entire work site, including the excavation target and the free-section excavator 10. It is preferable that the overhead camera 192 also includes an infrared camera so that images can be obtained even in environments where dust is generated. The image information captured by the overhead camera 192 is input to the control device 20.

[0022] Furthermore, the free-section drilling machine system 1 is equipped with an input device 193 for inputting information related to the drilling target, such as the drilling range and drilling depth to be drilled by the free-section drilling machine 10, into the control device 20. The input device 193 may also be used to input information about the drilling target, such as the rock type of the rock mass to be drilled. The input device 193 may be located at a distance from the free-section drilling machine 10.

[0023] The free-section excavator system 1 may also be equipped with other sensors. For example, a laser rangefinder or millimeter-wave radar may be provided to measure the distance from a predetermined position on the main body 110 to surrounding objects.

[0024] The control device 20, configured to control the operation of the free-section excavator 10, is a computer equipped with various integrated circuits such as a processor, memory, and storage, as well as various input and output interfaces. The control device 20 operates through the cooperation of these hardware components. The specific configuration of the control device 20 may be anything. Various programs and data are recorded in the control device 20, and the control device 20 operates using these programs, etc. The control device 20 may be mounted on the free-section excavator 10, or part or all of it may be located at a location away from the free-section excavator 10.

[0025] The control device 20 acquires drilling target information input via the input device 193. This drilling target information may include values ​​indicating drilling targets related to the amount of drilling, such as drilling range and drilling depth. The control device 20 also acquires sensor information from various sensors. Based on the acquired drilling target information and sensor information, the control device 20 outputs operation signals to operate each part of the free-section drilling machine 10 in order to achieve the input drilling target. Based on the operation signals, each part of the free-section drilling machine 10 operates and drills the target to be drilled.

[0026] [System operation] The operation of the free-section excavator system 1 according to this embodiment will now be described. For example, when the free-section excavator 10 is positioned in front of the excavation target, such as the face of a tunnel, and the free-section excavator 10 is fixed using the outriggers 112, the automatic excavation operation by the free-section excavator system 1 is started.

[0027] Figure 3 is a schematic diagram showing the rock mass that is the target of drilling 91. Generally, the target of drilling 91 has an uneven shape. Also, the hardness of the target of drilling 91 may vary depending on the location. Generally, the operator of a free-section drilling machine considers the shape and hardness of the target of drilling 91, etc., and decides the movement path of the cutting drum, as shown by the dotted line 93 in Figure 3, that is, from which direction the cutting drum will be pressed against each part of the target of drilling 91. The operator also considers the shape and hardness of the target of drilling 91, etc., and decides the movement speed of the cutting drum, etc. Based on these decisions, the operator operates the free-section drilling machine. In the free-section drilling machine system 1 of this embodiment, the control device 20 decides the plan of the path of the cutting drum 141 in such drilling, and the plan of operation including the movement speed of the cutting drum 141, and controls the operation of the free-section drilling machine 10 based on these plans.

[0028] Figure 4 is a functional block diagram of the control device 20 related to automatic excavation operation. As shown in Figure 4, the control device 20 is equipped with a trained model created by machine learning. In this embodiment, the control device 20 includes a first trained model 21, a second trained model 22, and a third trained model 23. Furthermore, a pre-processing unit is provided upstream of each trained model to perform pre-processing of the input data. That is, the control device 20 has the functions of a first pre-processing unit 24, a second pre-processing unit 25, and a third pre-processing unit 26, corresponding to the first trained model 21, the second trained model 22, and the third trained model 23, respectively. The control device 20 also has the function of an operation signal generation unit 27 that generates operation signals for operating each part of the free-section excavator 10 based on the outputs of the first trained model 21, the second trained model 22, and the third trained model 23. Furthermore, the control device 20 includes a function as an excavation target feature calculation unit 28 that calculates the state of the excavation target, such as its hardness, based on sensor information obtained during excavation, and a function as an excavation history recording unit 29 that records the calculated state of the excavation target, such as its hardness, and various history information related to excavation. In Figure 4, the various sensors provided in each part of the free-section excavator system 1 described above are represented as a sensor group 30.

[0029] The automated drilling operation will be explained with reference to the flowchart shown in Figure 5. In step S1, the control device 20 acquires sensor information from the sensor group 30. The sensor information acquired here may include information such as the shape of the drilling target acquired by the LiDAR 191, and images of the drilling target acquired by the imaging device 151 and the overhead camera 192.

[0030] In step S2, the control device 20 derives the state of the drilling target based on this information. That is, the acquired shape and image information of the drilling target is first input to the first preprocessing unit 24. The first preprocessing unit 24 adjusts this information into shape data and image data suitable for input to the first trained model 21. The adjusted shape data and image data are input to the first trained model 21. The first trained model 21 may be created using any machine learning method. The first trained model 21 is not limited to this, but may be a model created using deep learning, or a model created using supervised learning. The first trained model 21 is configured to output data on the state of the drilling target, such as the shape and hardness of each part of the drilling target, in response to the aforementioned shape data and image data. Therefore, the first trained model 21 outputs data on the state of the drilling target in response to the aforementioned input. The state of the drilling target, such as the shape and hardness of the drilling target, output here is also related to the difficulty of cutting by the cutting drum 141.

[0031] In step S3, the control device 20 acquires drilling target information from the input device 193. This drilling target information may include values ​​indicating the drilling target, such as the drilling range and drilling depth. In addition, if information about the drilling target, such as the rock type of the rock mass to be drilled, is input from the input device 193, the control device 20 acquires that information.

[0032] In step S4, the control device 20 derives a plan for the path of movement of the cutting drum 141 during excavation performed by moving the cutting drum 141, and a plan for operation including the movement speed of the cutting drum 141.

[0033] In other words, the drilling target information is first input to the second preprocessing unit 25. The second preprocessing unit 25 adjusts this information into drilling target data suitable for input to the second trained model 22. In addition, the data on the state of the drilling target derived in step S2 is input to the second preprocessing unit 25. The second preprocessing unit 25 adjusts this information into drilling target state data suitable for input to the second trained model 22. When creating the drilling target state data, the second preprocessing unit 25 may use information on the drilling target, such as rock type, input from the input device 193. The adjusted drilling target state data and drilling target data are input to the second trained model 22.

[0034] The second pre-trained model 22 may be created using any machine learning method. The second pre-trained model 22 may be, for example, a model created using deep reinforcement learning. The second pre-trained model 22 is configured to output data for planning the path and motion of the cutting drum 141 that can achieve efficient drilling, in response to the above-mentioned drilling target state data and drilling target data. Therefore, the second pre-trained model 22 outputs data for planning the path and motion of the cutting drum 141 in response to the above-mentioned inputs. The motion of the cutting drum 141 may include the moving speed of the cutting drum 141, the rotational speed of the cutting drum 141, and the pressing force of the cutting drum 141. The path and motion plan of the cutting drum 141 is optimized to most efficiently achieve the drilling target, for example, by determining which direction to press the cutting drum 141 from depending on the unevenness of the drilling target, and how fast to move the cutting drum's position and how much pressing force to generate depending on the hardness of the drilling target.

[0035] Furthermore, the second pre-trained model 22 may be a model created using deep learning, for example, or a model created using supervised learning. The second pre-trained model 22 may be configured to output data for planning the path and operation of the cutting drum 141 that can achieve efficient drilling, given the above-mentioned drilling target state data and drilling target data as inputs. Therefore, the second pre-trained model 22 outputs data for planning the path and operation of the cutting drum 141 in response to the above-mentioned inputs.

[0036] In step S5, the control device 20 derives operation signals to control the operation of each part of the free-section excavator 10. Specifically, the operation signal generation unit 27 creates operation signals to operate the cutting drum 141 along the path and motion plan based on the path and motion plan data derived by the second learned model 22. These operation signals may include, for example, time-dependent operation signals input to the slewing unit 121, vertical drive unit 126, and telescopic drive unit 131 of the boom 120, as well as the cutting motor 142.

[0037] In step S6, the control device 20 outputs the operation signals created by the operation signal generation unit 27 to each part of the free-section excavator 10. As a result, each part of the free-section excavator 10 operates according to the operation signals, and the excavation work is carried out.

[0038] In step S7, the control device 20 acquires various information from sensors provided in various parts of the free-section excavator 10. For example, the output of the stroke sensors of the hydraulic cylinders of the slewing section 121, vertical drive section 126, and telescopic drive section 131 of the boom 120, as well as the output of the hydraulic pressure gauge, can be acquired. In addition, the output of the cutting current meter 144 of the cutting device 140 can be acquired. Based on the outputs of these sensors, it can be determined whether the cutting drum 141 is moving according to the path and operation plan and cutting the target to be excavated.

[0039] In step S8, the control device 20 determines whether or not the excavation according to the route and operation plan has been completed. If it has not been completed, the process proceeds to step S9.

[0040] In step S9, the control device 20 derives an operation adjustment value. That is, although the free-section excavator 10 operates based on the path and operation plan derived in step S4, if there is an error in the state of the excavation target obtained in step S2, or if there is an error in the path and operation plan obtained in step S4, the free-section excavator 10 may not actually perform cutting as planned. Therefore, the operation of the free-section excavator 10 is adjusted based on the sensor information obtained in step S7.

[0041] For example, the current value flowing through the cutting motor 142, obtained by the cutting ammeter 144, is used as sensor information. This current value provides information about the cutting load on the cutting drum 141. In addition, the vibration of the machine body obtained by the machine body vibrometer 153, the stroke values ​​and hydraulic pressure of the hydraulic cylinders of the swivel section 121, the vertical drive section 126, and the telescopic drive section 131 are also used. These values ​​can also be used to obtain information about the cutting load on the cutting drum 141 and information about the movement of the cutting drum 141.

[0042] These values ​​are first input to the third preprocessing unit 26. The third preprocessing unit 26 adjusts this information to data suitable for input to the third trained model 23. The adjusted data is input to the third trained model 23. The third trained model 23 is a model that calculates adjustment values ​​for the operation of the cutting drum 141 based on the data input from the third preprocessing unit 26. Adjustment of the operation of the cutting drum 141 may, for example, reduce the movement speed of the cutting drum 141 if the cutting load is greater than the value assumed when deriving the operation plan in step S4. The third trained model 23 can calculate adjustment values ​​such as the movement speed of the cutting drum 141, the rotational speed of the cutting drum 141, and the pressing force of the cutting drum 141. Any algorithm may be used for the third trained model 23.

[0043] In step S10, the control device 20 calculates and records the characteristics of the drilling target. Specifically, the drilling target characteristic calculation unit 28 of the control device 20 calculates the characteristics of the drilling target, such as the hardness of the part being cut, based on the current value flowing through the cutting motor 142, the vibration of the machine body, the stroke value of each hydraulic cylinder, and the hydraulic pressure. The drilling history recording unit 29 of the control device 20 records the calculated characteristics of the drilling target and the obtained sensor information.

[0044] Subsequently, the process returns to step S5. In deriving the operation signal in step S5, the operation adjustment value derived in step S9 is taken into consideration. That is, the operation plan from step S4 is adjusted by the adjustment value to create the operation signal. As a result, the free-section excavator 10 can perform excavation operations according to the characteristics of the excavation target that became apparent when the excavation target was actually excavated. In this way, excavation continues while feeding back the actual excavation operation.

[0045] In step S8, if it is determined that the excavation is complete, the above process ends.

[0046] In the above explanation, it was stated that in step S2, the state of the drilling target is derived based on information such as the shape of the drilling target acquired by the LiDAR 191 and images of the drilling target acquired by the imaging device 151 and the overhead camera 192. However, the following is also possible. That is, after drilling to the target drilling depth acquired in step S3 is performed, consider advancing the free-section drilling machine 10 to perform further drilling. At this time, the new drilling target is the part adjacent to the drilling target that was drilled immediately before. The hardness of each part of the drilling target that was drilled immediately before is calculated in step S10 based on sensor information at the time of drilling and recorded in the drilling history recording unit 29. Therefore, in step S2, the information recorded in the drilling history recording unit 29 is read out as adjacent drilling information, and the first preprocessing unit 24 and the first trained model 21 may use this adjacent drilling information in addition to the shape and images of the drilling target to derive the state of the drilling target. By doing so, the state of the drilling target can be derived with even greater accuracy.

[0047] [About the first pre-trained model] As described above, the first trained model 21 is configured to output data on the state of the excavation target, such as the shape and hardness of each part of the excavation target, based on data on the shape and image of the excavation target, or data on the shape and image of the excavation target and data on adjacent excavation information. In order to create such a model using supervised learning with deep learning, appropriate training data is required. There are various ways to prepare such training data, but for example, training data may be created based on data obtained when excavation is actually performed using the free-section excavator 10. This excavation may be performed according to the operation of the operator.

[0048] Figure 6 is a diagram illustrating the creation of training data. Let's consider excavating an actual rock mass or other excavation target 311. First, the shape and image 312 of the excavation target 311 are acquired using equipment similar to the LiDAR 191, imaging device 151, and overhead camera 192. This information on the shape and image 312 of the excavation target 311 corresponds to the information acquired in the process of step S1 described above.

[0049] Next, the excavation target 311 is excavated using an excavator similar to the free-section excavator 10. At this time, sensor information 313 is acquired that corresponds to the current value flowing through the cutting motor 142, the vibration of the machine body, the stroke value of each hydraulic cylinder, and the hydraulic pressure, etc., as each part of the excavation target 311 is being cut. This sensor information 313 corresponds to the information acquired in the process of step S7 described above.

[0050] Based on the obtained sensor information 313, information on the characteristics of the drilling target 311, such as its hardness, is calculated. This calculation of information on the characteristics of the drilling target 311, such as its hardness, corresponds to the process performed in step S10 described above. This information on the characteristics of the drilling target 311, such as its hardness, corresponds to the state of the drilling target obtained in the process of step S2 described above.

[0051] As described above, the information on the shape and image 312 of the target to be excavated 311 is associated with the information on the characteristics of the target to be excavated 311, such as its hardness, obtained by actually excavating the target to be excavated 311, thereby creating training data 316. By machine learning using this training data 316, a first trained model 21 can be created that, in response to input data on the shape and image of the target to be excavated, outputs data on the state of the target to be excavated, such as the shape and hardness of each part of the target.

[0052] Furthermore, the sensor information 313 obtained during the excavation described above may also include information on the operator's operation, i.e., information on the operation signal. In this case, the information on the characteristics 314 of the excavation target obtained may be the relationship between the operation signal and the sensor information obtained in response to that operation signal, or the relationship between the operation signal and the cutting status obtained from the sensor information. With training data 316 that includes such information on the characteristics 314 of the excavation target, a first trained model 21 can be created that outputs data on the state of the excavation target, such as what kind of cutting status is obtained for what kind of operation signal, in response to input data on the shape and image of the excavation target. The second trained model 22 may also be configured to output the path and operation plan of the cutting drum 141 when such data on the state of the excavation target, such as what kind of cutting status is obtained for what kind of operation signal, is input.

[0053] Furthermore, consider further excavation of the excavation target 321 after the series of excavations described above have been performed. In this case, the information on the characteristics of the excavation target 311, such as its hardness, is the same as the information on the characteristics of the excavation target obtained in the process of step S10 described above, and is the same as the information on the adjacent excavation information that is recorded in the excavation history recording unit 29 and then read out. The information on the characteristics of the adjacent excavation target 325, such as its hardness, can be used in the next training data 326.

[0054] Furthermore, similar to the process for creating the training data 316 described above, the shape and image 322 of the excavation target 321 are acquired using the same equipment as the LiDAR 191, imaging device 151, overhead camera 192, etc. Subsequently, the excavation target 321 is excavated using an excavator similar to the free-section excavator 10. At this time, sensor information 323 is acquired, corresponding to the current value flowing through the cutting motor 142, the vibration of the machine body, the stroke value of each hydraulic cylinder, and the hydraulic pressure, etc., while each part of the excavation target 321 is being cut. Based on the obtained sensor information 323, information on the characteristics 324 of the excavation target, such as the hardness of the excavation target 321, is calculated.

[0055] As described above, the information on the shape and image 322 of the drilling target 321, the information on the characteristics of the drilling target 325 such as the hardness of the adjacent drilling target 311, and the information on the characteristics of the drilling target 324 such as the hardness of the drilling target 321 obtained by actually drilling the drilling target 321 are associated to create training data 326. By machine learning using this training data 326, a first trained model 21 can be created that takes data on the shape and image of the drilling target and data on adjacent drilling information as inputs and outputs data on the state of the drilling target, such as the shape and hardness of each part of the drilling target.

[0056] [Regarding the second pre-trained model] (Reinforcement learning) As described above, the second pre-trained model 22 may be a model created using, for example, deep reinforcement learning. The second pre-trained model 22 is configured to output data for the path and operation plan of the cutting drum 141 that enables efficient drilling, in response to the drilling target state data and drilling target data described above. Deep reinforcement learning for creating such a second pre-trained model 22 will be explained with reference to Figure 7.

[0057] For example, reinforcement learning can be performed using a machine learning model 411 with a deep neural network as the agent and a simulator 412 as the environment. The simulator 412 has a drilling target model that mimics rock. This drilling target model can mimic the shape and hardness of various drilling targets. When various conditions equivalent to drilling by the cutting drum 141 are applied to this drilling target model, the simulator 412 outputs drilling results according to predetermined rules. The simulator 412 can be constructed using various sensor information obtained when drilling an actual drilling target such as rock, similar to the example described with reference to Figure 6.

[0058] In reinforcement learning using model 411 and simulator 412, model 411 receives the drilling target model from simulator 412 as a state and applies various cutting drum paths and movements corresponding to the drilling target as actions to this drilling target model. Simulator 412 simulates the cutting drum path and movement at that time and calculates the corresponding drilling result. The time required for drilling, the power consumption, the amount of work, etc. in this drilling simulation can be used as the reward value for reinforcement learning. A shorter drilling time results in a higher score. A lower power consumption results in a higher score. A smaller difference between the planned path and the path the cutting drum actually moved in the simulation results in a higher score. Model 411 learns to find the cutting drum path and movement that maximizes the reward.

[0059] Through this reinforcement learning, a second trained model 22 can be obtained that outputs data for planning the path and operation of the cutting drum 141, enabling efficient drilling, given the input of drilling target state data and drilling target data.

[0060] Reinforcement learning is not limited to being performed using a simulator. For example, it may be performed by actually excavating the target object using the free-section excavator 10. Also, the retraining of the second trained model 22 may be continued while the free-section excavator system 1 is operating.

[0061] (Deep learning) Furthermore, the second pre-trained model 22 may be created not by reinforcement learning, but by, for example, supervised learning. For example, the second pre-trained model 22 may be created by deep learning. For example, a large amount of training data of drilling paths and drilling operations based on operator operations according to the state of the drilling target may be prepared, and the second pre-trained model 22 may be created by supervised learning using this data.

[0062] (Restrictions on route planning) The path plan for excavation by the cutting drum 141, output by the second trained model 22 created using reinforcement learning or deep learning, may be difficult for a human to understand its intended purpose, and its content may be difficult for a human to verify or understand. Therefore, restrictions may be placed on the path plan.

[0063] For example, as explained with reference to Figures 8 and 9, the path planning may be restricted as follows: The excavation target 91 is divided into multiple blocks 95 of adjusted size, as schematically shown by the dashed grid in Figure 8. The excavation paths permitted within each of these blocks 95 are restricted to a finite number of pre-prepared movement patterns, as schematically shown in the lower part of Figure 9. The movement patterns are not limited to movement in the up, down, left, and right planes, but can also be three-dimensional movement patterns including the front and back directions. Restrictions may be placed on machine learning so that the excavation path is planned by combinations of these finite patterns assigned to each block 95. Furthermore, the movement patterns may take into account the position of the block 95. For example, if a block 95 is located at the edge, patterns that move outward from it can be excluded to prevent over-excavation.

[0064] In this way, the path plan output from the second pre-trained model 22 can be made easier for humans to understand. As a result, unexpected excavation operations are prevented in the free-section excavator system 1, and the operability and maintainability of the free-section excavator system 1 are improved. Furthermore, even when retraining is performed, learning efficiency can be improved by learning in units of patterns. In addition, the human-readable path plan output from the second pre-trained model 22 can be used to evaluate the operator's operation or to train the operator's skills.

[0065] [About the Free-Cross Excavation Machine System] According to this embodiment, by realizing automatic operation in a free-section excavator used, for example, in tunnel excavation work, human operation and the operator's experience become unnecessary. The rock mass to be excavated by a free-section excavator has various conditions such as rock type, hardness, and cross-sectional shape, and operation of the free-section excavator is required according to these conditions. Such operation requires experience and knowledge, and there are challenges in the personalization of knowledge and the transfer of skills. According to this embodiment, excavation work using a free-section excavator can be mechanized, and the number of personnel can be reduced and the efficiency can be improved. [Explanation of Symbols]

[0066] 1: Free-section drilling machine system 10: Free-section excavator, 110: Main body, 111: Traveling section, 112: Outrigger, 113: Dozer, 114: Hydraulic supply section, 115: Cable reel, 116: Operating section 120: Boom, 121: Swivel section, 122: Swivel drive section, 123: Swivel stroke sensor, 124: Swivel hydraulic pressure gauge, 126: Up / down drive section, 127: Up / down stroke sensor, 128: Up / down hydraulic pressure gauge, 131: Telescopic drive section, 132: Telescopic stroke sensor, 133: Telescopic hydraulic pressure gauge 140: Cutting device, 141: Cutting drum, 142: Cutting motor, 143: Reducer, 144: Cutting ammeter 151: Imaging device, 152: Triaxial inclinometer, 153: Aircraft vibration meter, 154: Microphone 191: LiDAR, 192: Overhead camera, 193: Input device 20: Control device, 21: First trained model, 22: Second trained model, 23: Third trained model, 24: First pre-processing unit, 25: Second pre-processing unit, 26: Third pre-processing unit, 27: Operation signal generation unit, 28: Excavation target feature calculation unit, 29: Excavation history recording unit 30: Sensor group

Claims

1. A free-section drilling machine configured to excavate the target to be excavated using a cutting drum that moves in response to an operating signal, A sensor configured to acquire information about the target to be excavated, A control device configured to output the operation signal based on the information acquired by the sensor and drilling target information including the drilling volume of the target drilling target. Equipped with, The control device is A first trained model obtained by machine learning is configured to output information regarding the state of the target to be excavated based on the information acquired by the aforementioned sensor, A second pre-trained model obtained by machine learning is configured to output a plan for the movement path of the cutting drum and a plan for the operation of the cutting drum, based on the information regarding the state of the drilling target and the drilling target information output from the first pre-trained model. An operation signal generation unit configured to output the operation signal based on the planned route and the planned operation. Equipped with, Free-section excavation machine system.

2. The second trained model is a model obtained through reinforcement learning, where the reward is at least one of the following: the time required for drilling, the power consumption, and the difference between the planned path and the actual path of the cutting drum. The free-section excavator system according to claim 1.

3. The sensor includes a LiDAR or imaging device. The first trained model is configured to output information about the state of the drilling target, including at least information about the shape of the drilling target and the hardness of each part, using at least the shape or image of the drilling target acquired by the LiDAR or imaging device. The free-section excavator system according to claim 1.

4. The sensor further includes an ammeter for the cutting motor, The first trained model is configured to output information about the state of the drilling target, including information about the hardness of each part of the drilling target, based on the current value obtained by the ammeter of the cutting motor. The free-section excavator system according to claim 3.

5. The aforementioned second trained model is configured to divide the excavation target into multiple blocks of adjusted size, select one movement pattern from a pre-prepared set of movement patterns for the cutting drum within each block, and output a plan of the movement path of the cutting drum represented by the combination of the block and the movement pattern. A free-section excavator system according to any one of claims 1 to 4.

Images