Control device, inspection system, control method, and program

Abstract

To provide a control device capable of accurately adjusting the attitude of a long flexible robot.SOLUTION: A control device for an inspection device provided with a long flexible robot formed by connecting a plurality of units bendable to desired 1 curvature for each unit and an attitude actuator capable of adjusting the attitudes of the units comprises: a target attitude identification unit that identifies a target attitude composed of the position of a representative point of each of the units advancing along a predetermined route; and an operation amount determination unit that determines an operation amount for setting the unit in the target attitude by using a learning model in which the target attitude of the unit is an input and the operation amount of the attitude actuator is an output.SELECTED DRAWING: Figure 9

Description

[Technical Field]

[0001] The present disclosure relates to a control device, an inspection system, a control method, and a program in the field of inspection of industrial plants, for example. [Background technology]

[0002] Robots used for inspecting narrow spaces are known. Such robots have, for example, an articulated structure and are configured to be long and flexibly bendable so that they can pass through narrow spaces and reach their final destination. Hereinafter, a robot having such a configuration will also be referred to as a "long and flexible robot."

[0003] Patent Document 1 discloses a drive control device that moves a robot along a predetermined route inside an object to be inspected. This robot is configured by connecting multiple tubes. The drive control device adjusts the posture of the tubes along the route by operating wires fixed to each tube with posture actuators, thereby constantly changing the posture of the tubes. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] International Publication No. 2020 / 090959 Summary of the Invention [Problem to be solved by the invention]

[0005] However, with conventional technology, there was a possibility that the target posture of the tube, set by the drive control device to follow the route, would deviate from the actual posture of the tube due to various factors such as stretching due to wire tension, attenuation of tension due to friction between the wire and the wire insertion holes provided in each tube, nonlinearity of wire load and stretch, and variations in wire characteristics.

[0006] The present disclosure has been made in consideration of such problems, and provides a control device, an inspection system, a control method, and a program that can more accurately adjust the posture of a long, flexible robot. [Means for solving the problem]

[0007] According to one aspect of the present disclosure, a control device of an inspection device including a long flexible robot formed by connecting multiple units each capable of bending to a desired curvature, and an attitude actuator capable of adjusting the attitude of the units, includes a target attitude identification unit that identifies a target attitude consisting of the positions of representative points of each of the units progressing along a predetermined path, and an operation amount determination unit that determines an operation amount for bringing the unit into the target attitude using a learning model that takes the target attitude of the unit as input and outputs an operation amount of the attitude actuator.

[0008] According to one aspect of the present disclosure, an inspection system includes the control device according to the above aspect and an inspection device.

[0009] According to one aspect of the present disclosure, a control method for an inspection device including a long flexible robot formed by connecting multiple units each capable of bending to a desired curvature, and an attitude actuator capable of adjusting the attitude of the units, includes the steps of: identifying a target attitude consisting of the positions of representative points of each of the units moving along a predetermined path; and determining an operation amount for bringing the unit into the target attitude using a learning model that takes the target attitude of the unit as input and outputs the operation amount of the attitude actuator.

[0010] According to one aspect of the present disclosure, the program causes an inspection device having a long flexible robot formed by connecting multiple units, each of which can bend to a desired curvature, and an attitude actuator that can adjust the attitude of the units, to execute the following steps: identifying a target attitude consisting of the positions of representative points of each of the units moving along a predetermined path; and determining an operation amount for bringing the unit into the target attitude using a learning model that takes the target attitude of the unit as input and outputs the operation amount of the attitude actuator. [Effects of the Invention]

[0011] According to the control device, inspection system, control method, and program of the present disclosure, the posture of a long flexible robot can be adjusted more accurately. [Brief explanation of the drawings]

[0012] [Figure 1] 1 is a diagram showing the overall configuration of an inspection system according to a first embodiment. [Figure 2] FIG. 1 is a first diagram showing the configuration of an inspection device according to a first embodiment. [Figure 3] FIG. 2 is a second diagram showing the configuration of the inspection device according to the first embodiment. [Figure 4] FIG. 3 is a third diagram showing the configuration of the inspection device according to the first embodiment. [Figure 5] FIG. 4 is a fourth diagram showing the configuration of the inspection device according to the first embodiment. [Figure 6] FIG. 5 is a fifth diagram showing the configuration of the inspection device according to the first embodiment. [Figure 7] FIG. 6 is a sixth diagram showing the configuration of the inspection device according to the first embodiment. [Figure 8] FIG. 2 is a diagram illustrating a hardware configuration of a control device according to the first embodiment. [Figure 9] FIG. 2 is a first diagram showing the functional configuration of a CPU according to the first embodiment. [Figure 10] FIG. 2 is a second diagram showing the functional configuration of the CPU according to the first embodiment. [Figure 11] FIG. 4 is a diagram showing a processing flow of a learning mode of a CPU according to the first embodiment. [Figure 12] FIG. 4 is a diagram showing a processing flow of a control mode of a CPU according to the first embodiment. [Figure 13] FIG. 3 is a first diagram illustrating the effect of the control device according to the first embodiment. [Figure 14] FIG. 6 is a second diagram illustrating the effect of the control device according to the first embodiment. [Figure 15] FIG. 10 is a diagram illustrating a functional configuration of a CPU according to a second embodiment. [Figure 16] FIG. 10 is a first diagram illustrating the function of a control device according to a second embodiment. [Figure 17] FIG. 10 is a diagram showing a processing flow of a learning mode of a CPU according to the second embodiment. [Figure 18] FIG. 10 is a second diagram illustrating the function of the control device according to the second embodiment. [Figure 19] FIG. 10 is a diagram illustrating functions of a control device according to a third embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0013] First Embodiment A control device according to a first embodiment and an inspection system including the control device will be described below with reference to FIGS.

[0014] (Overview of the overall configuration of the inspection system) FIG. 1 is a diagram showing the overall configuration of an inspection system according to the first embodiment. The inspection system 1 shown in FIG. 1 is used to inspect narrow spaces (for example, the inside of a gas turbine or a steam turbine).

[0015] 1, the inspection system 1 includes a control device 10 and an inspection device 5. The inspection system 1 operates in one of two modes: a learning mode in which a model used to control the inspection device 5 is learned, and a control mode in which the inspection device 5 is controlled using the learned model. Note that the inspection system 1 further includes a position measurement sensor 7 only in the learning mode.

[0016] (Configuration of inspection equipment) First, the inspection device 5 will be described in detail with reference to FIGS. The inspection device 5 is a device that can be inserted into an object to be inspected (such as a gas turbine) to check the inside of the object to be inspected. The inspection device 5 of this embodiment includes an inspection cable 61 (FIG. 3) and a long flexible robot 6.

[0017] First, the overall configuration of the long flexible robot 6 will be described with reference to FIG. The long flexible robot 6 has a tube 62 formed by connecting a plurality of units U in series. Each unit U has a multi-joint structure that can be bent at multiple points, and is structured so that it can be bent in a desired direction within a predetermined range (for example, up to 90°). However, based on the structure described below, one unit U can be bent only with one curvature, and one unit alone cannot be deformed into a shape with two curvatures (for example, an S-shape).

[0018] The units U are connected in series at connecting portions L (portions indicated by black dots in FIG. 2). In this embodiment, one unit U is made up of, for example, three nodes S. The boundaries (connecting portions L) of each unit U and the boundaries of each node S are separated by flanges 632.

[0019] From the tip of the long flexible robot 6, a sensor 612 provided at the tip of an inspection cable 61 (FIG. 3) is inserted.

[0020] Next, the configurations of the inspection cable 61 and the long flexible robot 6 will be described in detail with reference to FIGS. The inspection cable 61 has a highly flexible cable main body 611 and a sensor 612. The cable main body 611 can be bent in any direction intersecting the cable extension direction, which is the direction in which the cable main body 611 extends, by an operator operating an operation unit (not shown). The cable main body 611 is a member separate from the tube 62, and is detachably fixed to the tube 62. The cable main body 611 is provided with an actuator (not shown) for moving the cable so that it can be driven independently, separately from the long flexible robot 6.

[0021] The sensor 612 is fixed to the tip of the cable main body 611. The sensor 612 and the cable main body 611 are built into a tube 62 (described later). The sensor 612 in this embodiment is a camera capable of capturing images of the inside of an inspection object. Captured data such as videos and images captured by the sensor 612 is sent to a camera image monitor or the like via a cable extending from the end (rear end) of the cable main body 611 on which the sensor 612 is not provided. As the inspection cable 61 in this embodiment, for example, a borescope (industrial endoscope) is used for observing and inspecting deep areas that cannot be directly seen with the naked eye.

[0022] The long flexible robot 6 (tube 62) may be any robot as long as it has a bendable structure, and may be, for example, a snake-like robot having a multi-joint structure in which a plurality of highly flexible members are connected.

[0023] Furthermore, the sensor 612 is not limited to being a camera as in this embodiment. For example, the sensor 612 in this embodiment may be a sensor 612 having a dimension measurement function (for example, three-dimensional phase measurement) or a sensor 612 capable of measuring temperature or the presence or absence of scratches.

[0024] The long flexible robot 6 includes a tube 62, a posture actuator 65, and an advance / retreat actuator 67.

[0025] As shown in FIG. 3, the tube 62 has a hollow portion formed therein through which the inspection cable 61 can be inserted. The tube 62 is flexible. The tube 62 has a multi-joint structure that allows it to be bent at multiple locations. Therefore, the tube 62 can be bent in any direction intersecting the tube extension direction, which is the direction in which the tube 62 extends. It is preferable that each joint of the tube 62 has a structure that is easy to bend but difficult to twist and compress. The outer diameter of the tube 62 is set to a size (e.g., 10 mmφ) that allows it to be inserted into a narrow portion of the inspection target. A cable main body 611 is detachably attached to the tube 62. The tube 62 of this embodiment is configured by connecting multiple tube main bodies 63. One tube main body 63 corresponds to one node S shown in FIG. 2.

[0026] The multiple tube bodies 63 are arranged side by side in the direction in which the tube bodies 63 extend and are connected to one another. As shown in Fig. 4, the tube body 63 has a tubular portion 631 with both ends open, and flanges 632 protruding radially outward from the outer peripheral surfaces of both ends of the tubular portion 631. The tubular portion 631 has a cylindrical shape into which the inspection cable 61 can be inserted. The tubular portion 631 has, for example, multiple slits (not shown) formed therein, making it bendable in any direction. The flange 632 has an annular shape and is formed integrally with the tubular portion 631.

[0027] As shown in Fig. 3, the attitude actuator 65 is capable of adjusting the attitude of the tube 62. Here, the attitude of the tube 62 refers to the position and direction of the tip of the tube 62 on an imaginary plane intersecting the tube extension direction. The attitude actuator 65 of this embodiment is fixed to the base end (rear end) of the tube 62. As shown in Fig. 7, the attitude actuator 65 has a plurality of wires 651, a housing unit 652, a pulley 653, a wire driving unit 654, and a wire load detection unit 655.

[0028] As shown in FIG. 4, a plurality of wires 651 (for example, three in this embodiment) are provided for one tube body 63. The tips of the wires 651 are fixed to the tip surface of the tube body 63. As shown in FIG. 5, the wires 651 are fixed apart from one another so as to be out of phase with each other (for example, 120 degrees) with respect to the tip surface of one tube body 63. Furthermore, the wires 651 are arranged out of phase with each other for each adjacent tube body 63. Therefore, as shown in FIG. 6, the fixing position of the wire 651 in one tube body 63 (tube body 63A) arranged at the tip end side is offset by a predetermined angle (for example, 40 degrees) in another tube body 63 (tube body 63B) adjacent at the base end side. Each tube body 63 is formed with a wire insertion hole 633 for inserting the wire 651. In addition, the tip of the wire 651 may be provided with a spherical portion having a diameter larger than the diameter of the wire insertion hole 633, and this spherical portion may prevent the wire 651 from slipping out of the wire insertion hole 633 (by fixing it to the end face of the tube main body 63).

[0029] 7, the housing 652 is fixed to the base end of the tube 62. The housing 652 houses one end of the wire 651 inside. The housing 652 has a housing through-hole 652A formed therein, through which the cable main body 611 protruding from the base end of the tube 62 can be inserted. The housing through-hole 652A is formed to pass through the housing 652.

[0030] Pulley 653 is rotatably attached inside housing 652. Pulley 653 reverses the direction in which wire 651 extends inside housing 652. A pulley 653 is provided for each wire 651. In other words, one pulley 653 is provided for each wire 651. A plurality of pulleys 653 are provided spaced apart from each other to surround housing through-hole 652A.

[0031] The wire driving unit 654 is fixed inside the housing 652. A wire driving unit 654 is provided for each wire 651. That is, one wire driving unit 654 is provided for each wire 651. The wire driving unit 654 is connected to the base end of the wire 651, which is the end of the wire 651 that is not fixed to the tube main body 63, via a wire load detection unit 655. The wire driving unit 654 is capable of moving the wire 651 forward and backward relative to the pulley 653. As the wire driving unit 654, for example, an electric slider, an electric cylinder, or a ball screw is used.

[0032] The wire load detection unit 655 is disposed between the base end of the wire 651 and the wire driver 654. The wire load detection unit 655 measures the load (wire tensile force) acting on the wire 651 and sends the measurement result to the wire driver 654. If the sent measurement result is equal to or greater than a value determined to be excessive (e.g., a value that may damage the wire 651), the wire driver 654 is driven to loosen the wire 651. If the sent measurement result is equal to or less than a value determined to be insufficient (e.g., a value at which the wire 651 is deemed to be bent), the wire driver 654 is driven to tension the wire 651 to a degree that does not loosen it. The wire load detection unit 655 may be, for example, a load cell that can directly measure the load. Alternatively, the load may be measured indirectly based on the motor current value in the wire driver 654.

[0033] Furthermore, the attitude actuator 65 drives some of the multiple tube bodies 63 that are located near the tip. The number of tube bodies 63 driven by the attitude actuator 65 may be one or more. As shown in FIG. 3 , the tube 62 of this embodiment is divided into an active part 62A that is driven by the attitude actuator 65 and a driven part 62B that is not driven by the attitude actuator 65.

[0034] In the active portion 62A, a wire 651 is fixed to the flange 632 of each tube body 63. The active portion 62A is a region of a predetermined length from the tip of the tube 62. Here, the predetermined length is a length that can reach a desired inspection range.

[0035] The driven part 62B moves following the movement of the active part 62A. In the driven part 62B, the wire 651 is not fixed to the flange 632 of each tube main body 63. The driven part 62B is the region from the base end to the active part 62A in the tube 62. In this embodiment, the driven part 62B is the region sandwiched between the housing part 652 and the active part 62A.

[0036] The advancing / retreating actuator 67 is capable of advancing / retreating the tube 62. Here, advancing / retreating the tube 62 means moving the tube 62 in the tube extension direction. The advancing / retreating actuator 67 of this embodiment is capable of moving the housing part 652 to which the tube 62 is fixed. The advancing / retreating actuator 67 has a guide rail 672 and an advancing / retreating drive part 671.

[0037] The advance / retract drive unit 671 moves on guide rails 672. The housing unit 652 is fixed to the advance / retract drive unit 671. The advance / retract drive unit 671 is, for example, an electric slider. When the advance / retract drive unit 671 moves on the guide rails 672 so as to approach the inspection object, the tube 62 is inserted deep inside the inspection object. Conversely, when the advance / retract drive unit 671 moves on the guide rails 672 so as to move away from the inspection object, the tube 62 is moved from deep inside the inspection object to near the entrance.

[0038] (Configuration of position measurement sensor) The position measurement sensor 7 (FIG. 1) measures the position of a representative point of each unit U of the long flexible robot 6 during learning. The representative points are set, for example, at the tip and central position of each unit U that will become the active part 62A. Furthermore, as shown in FIG. 1, a marker 71 for position measurement is attached to each representative point.

[0039] The position measurement sensor 7 according to this embodiment is a camera. The camera serving as the position measurement sensor 7 measures, from a captured image, the position where the marker 71 is attached, i.e., three-dimensional coordinates indicating the position of the representative point of each unit U. The measured three-dimensional coordinates are output to the control device 10. The position measurement sensor 7 may be any sensor capable of measuring the position of the representative point. For example, in other embodiments, the position measurement sensor 7 is a laser scanner, LiDAR, or the like. Furthermore, if the representative point of each unit can be identified from characteristics such as shape or color, the marker 71 may be omitted.

[0040] (Control device hardware configuration) FIG. 8 is a diagram illustrating a hardware configuration of the control device according to the first embodiment. As shown in FIG. 8, the control device 10 includes a CPU 100, a communication interface 101, a memory 102, an input device 103, an output device 104, and a storage 105.

[0041] The CPU 100 performs various functions by operating in accordance with pre-prepared programs. The functions of the CPU 100 will be described in detail later.

[0042] The communication interface 101 is, for example, a connection interface with the long flexible robot 6 and other terminal devices.

[0043] The memory 102 is a so-called main storage device, and provides a storage area necessary for the processing of the CPU 100.

[0044] The input device 103 is a device that accepts operations from an operator, and is, for example, a mouse, a keyboard, a touch sensor, or the like.

[0045] The output device 104 is a device for outputting various information to an operator, and is a display, a speaker, etc.

[0046] The storage 105 is a so-called auxiliary storage device, such as a hard disk drive (HDD) or a solid state drive (SSD).

[0047] (Functional configuration of the control device) FIG. 9 is a first diagram illustrating the functional configuration of the CPU according to the first embodiment. FIG. 10 is a second diagram illustrating the functional configuration of the CPU according to the first embodiment. Next, the functions of the CPU 100 will be described with reference to FIGS. In addition, the CPU 100 according to this embodiment has a learning mode for learning a model (learning model M1 described later) used to control the long flexible robot 6, and a control mode for controlling the long flexible robot 6 using the learned model. First, the functional configuration of the CPU 100 in the control mode will be described with reference to Fig. 9. In the control mode, the CPU 100 has the functions of a desired attitude specifying unit 1001, an operation amount determining unit 1002, and an output unit 1004, as shown in Fig. 9.

[0048] In the control mode, the desired posture specifying unit 1001 specifies a desired posture of each unit U of the long flexible robot 6 moving along a predetermined path. The path is determined, for example, by a known path generating device (not shown) according to the internal structure of the inspection target. The desired posture is expressed, for example, by the three-dimensional coordinates of a representative point of each unit U. The desired posture specifying unit 1001 specifies a desired posture of each unit U so that the position of the representative point of each unit U coincides with the path every time the long flexible robot 6 moves along the path. Here, the desired posture specifying unit 1001 outputs, as a time history, the desired posture of each unit U at each step of the movement of the long flexible robot 6 from the entry position on the path to the target position.

[0049] In the control mode, the operation amount determination unit 1002 determines the operation amount (hereinafter also referred to as "control operation amount") for bringing each unit U into a target posture using a learning model M1 that receives the target posture of each unit U as input and outputs the operation amount of the posture actuator 65. The learning model M1 is a model that learns the relationship between the operation amount of the posture actuator and the joint posture of the long flexible robot 6, and one that has been learned by a learning unit 1003 described below is used. The operation amount is specifically the amount of pulling of each wire 651 by the posture actuator 65 (wire driving unit 654). Here, the operation amount determination unit 1002 outputs the operation amount of each step in which the long flexible robot 6 advances from the entry position of the path to the target position as a time history. The joint posture of the long flexible robot 6 is represented by the position of the representative point of each unit U.

[0050] In the control mode, the output unit 1004 outputs the time history of the manipulated variable for control determined by the manipulated variable determination unit 1002 as a control signal to the actual long flexible robot 6.

[0051] Next, the functional configuration of the CPU 100 in the learning mode will be described with reference to Fig. 10. In the learning mode, the CPU 100 has the functions of a learning unit 1003 and an output unit 1004, as shown in Fig. 10.

[0052] The learning unit 1003 constructs a learning model M1 based on the operation amount of the attitude actuator 65 of the long flexible robot 6 and the position (three-dimensional coordinates) of the representative point of each unit U measured by the position measurement sensor 7. The learning model M1 is a neural network that receives the attitude of the long flexible robot 6 (the position of the representative point of each unit U) as an input and receives the operation amount of the wire 651 as an output. In addition, in the learning mode, the learning unit 1003 determines the operation amount of the attitude actuator 65 (hereinafter also referred to as "learning operation amount") instead of the operation amount determination unit 1002.

[0053] In the learning mode, the output unit 1004 outputs the learning operation amount determined by the learning unit 1003 as a control signal to the actual long flexible robot 6.

[0054] (Learning mode process flow) FIG. 11 is a diagram showing a processing flow of the learning mode of the CPU according to the first embodiment. The process flow of the CPU 100 in the learning mode will be described in detail below with reference to FIG.

[0055] First, the learning unit 1003 determines the learning manipulation amount of the posture actuator 65 (step S1). At this time, the learning unit 1003 randomly determines the value of the learning manipulation amount, for example. Furthermore, the learning manipulation amount determined by the learning unit 1003 is output from the output unit 1004 as a control signal to the actual long flexible robot 6. Then, the long flexible robot 6 changes the posture of each unit U in accordance with the control signal.

[0056] Next, the learning unit 1003 calculates a set of positions of the representative points of each unit U of the long flexible robot 6 for the set of learning operation amounts Wi (i=1 to N, N: number of wires) {x j ,y j ,z j} (j=1 to M, M=the number of representative points) is acquired from the position measurement sensor 7 (step S2).

[0057] The learning unit 1003 generates a set of learning operation variables Wi and a set of positions of representative points of each unit U for the learning operation variables Wi {x j ,y j ,z j}. Then, the learning unit 1003 constructs a learning model M1 based on the acquired learning data (step S3). The learning unit 1003 constructs the learning model M1 using, for example, a deep neural network (DNN).

[0058] Furthermore, the learning unit 1003 stores the constructed learning model M1 in the storage 105 (step S4).

[0059] (Control mode processing flow) FIG. 12 is a diagram showing a processing flow of the control mode of the CPU according to the first embodiment. The flow of processing in the control mode of the CPU 100 will be described in detail below with reference to FIG.

[0060] First, the desired posture specifying unit 1001 specifies the time history of the desired posture of each unit U at each step when the long flexible robot 6 advances from the entry position to the target position on a predetermined path (step S10).

[0061] Next, the manipulated variable determination unit 1002 inputs the target posture identified by the target posture identification unit 1001 into the learning model M1 constructed by the learning unit 1003 and stored in the storage 105. Then, the learning model M1 outputs the manipulated variable of the posture actuator 65 for the target posture. Based on the manipulated variable output from the learning model M1, the manipulated variable determination unit 1002 determines the time history of the manipulated variable for control corresponding to the time history of the target posture (step S11).

[0062] Next, the output unit 1004 outputs the time history of the manipulated variable for control determined by the manipulated variable determination unit 1002 as a control signal to the actual long flexible robot 6 (step S12). Then, the attitude actuator 65 operates each wire 651 in accordance with the time history of the manipulated variable for control in each step. This allows the long flexible robot 6 to change its attitude so as to follow the path as it progresses along the path.

[0063] (effect) With reference to FIGS. 13 and 14, the effects of controlling the long flexible robot 6 using the learning model M1 will be described. Here, a plurality of target postures were given to each of the control device 10 according to this embodiment and a control device of the prior art (for example, the drive control device described in Patent Document 1), and the posture of the long flexible robot 6 was controlled. FIG. 13 shows the error of the actual posture of the long flexible robot 6 with respect to each target posture. The error (RMSE) of the actual posture with respect to the target posture is calculated by the following formula (1). In formula (1), {x j,y j ,z j} are the three-dimensional coordinates (actual posture) of the representative point of the long flexible robot 6 measured by the position measurement sensor 7, and {x j,ref ,y j,ref ,z j,ref} represents the target posture, and M represents the number of representative points.

[0064]

number

[0065] Conventional control devices do not take into consideration disturbances such as the weight of each unit U and the inspection cable 61, the stretch and friction of the wire 651, and the characteristics of each wire 651. In contrast, the control device 10 according to this embodiment can construct a learning model M1 that incorporates these disturbances by learning the relationship between the actual posture of the long flexible robot 6 and the operation amount. As a result, as shown in FIG. 13, the control device 10 according to this embodiment can significantly reduce errors for all target postures compared to conventional control devices.

[0066] Moreover, Fig. 14 shows the positional error of each representative point of the long flexible robot 6 for a certain target posture. As shown in Fig. 14, in the conventional control device, the disturbance described above is not taken into consideration, and as a result, the position of each representative point significantly deviates from the target posture. In contrast, the control device 10 according to this embodiment was able to significantly reduce the error from the target posture for all representative points.

[0067] As described above, according to the control device 10 according to the first embodiment and the inspection system 1 including the same, the posture of the long flexible robot 6 can be adjusted more accurately.

[0068] <Second embodiment> Next, a control device according to a second embodiment of the present disclosure and an inspection system including the control device will be described with reference to FIGS. Components common to the above-described embodiment are denoted by the same reference numerals, and detailed description thereof will be omitted.

[0069] In the first embodiment, in the learning mode, the manipulated variable determiner 1002 randomly determined the value of the manipulated variable of the posture actuator 65. That is, in the control device 10 according to the first embodiment, no restrictions were placed on the posture of the long flexible robot 6 to be learned. However, in reality, the path inside the inspection object is finite, so it is considered more efficient to narrow down learning to postures that can be taken when moving along the path. For this reason, the control device 10 according to the present embodiment selects in advance postures that can be taken by the long flexible robot 6 in the learning mode, and efficiently constructs the learning model M1 by narrowing down learning to the selected postures.

[0070] FIG. 15 is a diagram illustrating a functional configuration of a control device according to the second embodiment. 15, the control device 10 according to this embodiment further includes an operation device 106 for an operator to manually adjust the posture of each unit U of the long flexible robot 6. The operation device 106 is, for example, an input device such as a joystick or a touch panel.

[0071] FIG. 16 is a first diagram illustrating the function of the control device according to the second embodiment. FIG. 17 is a diagram showing a processing flow of the learning mode of the CPU according to the second embodiment. First, the learning unit 1003 of the control device 10 sets a target posture to be learned (step S20). In this embodiment, the operator determines a plurality of target postures that the long flexible robot 6 can take on the path to be inspected, and instructs the learning unit 1003 via the input device 103.

[0072] Next, the learning unit 1003 calculates the wire path length based on the estimated wire elongation value obtained by inputting the current operation amount (wire pull amount) of the posture actuator 65 and the wire tension into the elongation model M3. The learning unit 1003 also estimates the current joint posture from the wire path length using forward kinematics, and derives the Jacobian matrix J (step S21).

[0073] As shown in Figure 18, the active part 62A of the long flexible robot 6 consists of three units U1, U2, and U3, and three wires 651 are fixed to each unit. The Jacobian matrix J is a sensitivity matrix that converts the velocity of the tip of each unit U1, U2, and U3 into the operation velocity of the attitude actuator 65. The first term on the right side of the following equation (2) is the Jacobian matrix J, and the elements of this matrix are basically determined by the attitude angle of each unit U. q·i (i = 1 to 9) on the left side of equation (2) is the pulling speed of each wire 651. x· and y· in the second term on the right side of equation (2) are the attitude speed of the tip of each unit U, and q·m is the average wire pulling speed of each unit U. Note that "q·", "x·", and "y·" correspond to the notation where a dot symbol "·" is added above "q", "x", and "y", respectively.

[0074]

number

[0075] The forward kinematics model M2 is constructed based on the geometrical specifications of each unit U (e.g., segment length, joint length, wire pitch circle diameter, wire phase, etc.) and the wire path length obtained from the stretch model M3, and is a mechanical model that calculates the tip position of each unit U from the amount of pull of the wire 651. The stretch model M3 is a model that estimates the wire stretch from the wire tension based on the tension-stretch characteristic (nonlinear) of the wire 651. When the wire stretch changes, the wire path length changes, affecting the bending angle of the unit U. Generally, a wire with large stretch tends to have a smaller bending angle than a wire with small stretch. In this embodiment, the stretch model M3 is used to calculate the estimated amount of stretch of the wire 651 based on the measurement results of the wire load detection unit 655. Note that the forward kinematics model M2 and the stretch model M3 may be those used in control devices of the prior art.

[0076] Next, the learning unit 1003 uses the Jacobian matrix J to determine a set of learning manipulation amounts Wi (i=1 to N, N=9 in the example of FIG. 16) for bringing each unit U of the long flexible robot 6 into a target posture (step S22). The learning manipulation amounts determined by the learning unit 1003 are output from the output unit 1004 as control signals to the actual long flexible robot 6. Then, the long flexible robot 6 changes the posture of each unit U in accordance with the control signals.

[0077] When the posture of each unit U of the long flexible robot 6 changes, the learning unit 1003 acquires the position of the representative point of each unit U of the long flexible robot 6 from the position measurement sensor 7 (step S23).

[0078] Next, the difference between the actual posture and the target posture of each unit U is calculated from the actual position of each unit U measured by the position measurement sensor 7. Furthermore, using the Jacobian matrix J, a posture velocity target value is calculated so as to reduce the error between the actual posture and the target posture, and the operator manually inches via the operation device 106 to adjust the long flexible robot 6 so that it assumes a posture (shape) close to the target posture (step S24). An posture close to the target posture is, for example, a posture in which the position of the representative point of each unit U is located within a predetermined distance from the target position of each representative point in the target posture.

[0079] Next, the learning unit 1003 acquires learning data consisting of the position of the representative point of each unit U after adjustment measured by the position measurement sensor 7 and the operation amount of the posture actuator 65 (the pulling amount of each wire 651) (step S25). Then, the learning unit 1003 constructs a learning model M1 based on the acquired learning data (step S26). Note that the learning unit 1003 may be configured to sequentially collect learning data while the posture of the long flexible robot 6 is being manually adjusted.

[0080] Furthermore, the learning unit 1003 stores the learned learning model M1 in the storage 105 (step S27).

[0081] As described above, the control device 10 according to the second embodiment manually adjusts the posture of the long flexible robot 6 so as to approach a target posture that the long flexible robot 6 can take, and acquires the posture of the long flexible robot 6 after this adjustment and the corresponding operation amount as learning data. In this way, the learning model M1 can be learned more efficiently with a smaller amount of learning data than when the operation amount is randomized.

[0082] Furthermore, the control device 10 according to the second embodiment determines the learning operation amount for the target posture using the Jacobian matrix J derived from the forward kinematics model M2 and the extension model M3. In this way, the control device 10 can efficiently construct a learning model.

[0083] <Third embodiment> Next, a control device and an inspection system including the control device according to a third embodiment of the present disclosure will be described with reference to FIG. Components common to the above-described embodiment are denoted by the same reference numerals, and detailed description thereof will be omitted.

[0084] FIG. 19 is a diagram illustrating the function of the control device according to the third embodiment. In the second embodiment, the learning unit 1003 determines the learning operation amount by manual operation using the Jacobian matrix J. In contrast, the learning unit 1003 according to the present embodiment automatically determines the learning operation amount as shown in FIG.

[0085] Furthermore, the learning unit 1003 determines the learning manipulation amount using the learning model M1. The position measurement sensor 7 measures the actual posture of each unit U based on this learning manipulation amount in real time. The learning unit 1003 sequentially corrects the learning manipulation amount using the Jacobian matrix J based on the error between the actual posture of the unit U at each time and the target posture. The learning unit 1003 also performs feedback control based on the corrected learning manipulation amount and the posture of the unit U, automatically updating the learning model M1 in real time. In this way, the learning model M1 can be tuned more efficiently.

[0086] As described above, several embodiments of the present invention have been described, but all of these embodiments are presented as examples and are not intended to limit the scope of the invention. These embodiments can be implemented in various other forms, and various omissions, substitutions, and modifications can be made without departing from the spirit of the invention. These embodiments and their modifications are included in the scope of the invention and its equivalents as defined in the claims, as well as in the scope and spirit of the invention.

[0087] <Additional Notes> The control device, inspection system, control method, and program described in the above-described embodiments can be understood, for example, as follows.

[0088] (1) According to a first aspect of this embodiment, a control device (10) of an inspection device (5) including a long flexible robot (6) formed by connecting a plurality of units each capable of bending to a desired curvature, and an attitude actuator (65) capable of adjusting the attitude of the units, includes a target attitude specification unit (1001) that specifies a target attitude consisting of the positions of representative points of each unit proceeding along a predetermined path, and an operation amount determination unit (1002) that determines an operation amount for bringing the unit into the target attitude using a learning model that receives the target attitude of the unit as an input and outputs an operation amount of the attitude actuator (65).

[0089] In this way, the control device can adjust the robot's posture more accurately.

[0090] (2) According to a second aspect of this embodiment, the control device (10) according to the first aspect further includes a learning unit (1003) that constructs a learning model based on the position of the representative point acquired from a sensor that measures the three-dimensional position of the representative point and the operation amount of the attitude actuator (65).

[0091] In this way, the control device can construct a relationship model (learning model) between the wire pulling amount and the joint posture.

[0092] (3) According to a third aspect of this embodiment, in the control device (10) according to the second aspect, the learning unit (1003) calculates a Jacobian matrix representing a sensitivity matrix that converts the velocity of the tip of each unit into the operation velocity of the attitude actuator (65) from a forward kinematics model that has the operation amount of the attitude actuator (65) as an input and the position of the representative point as an output, determines a learning operation amount for bringing the unit closer to the target attitude using the Jacobian matrix, and constructs a learning model based on the target attitude, the learning operation amount, and the position of the representative point of the unit after operation with the learning operation amount.

[0093] In this way, the control device can construct a learning model more efficiently with a smaller amount of learning data than when learning by randomizing the learning operation variables.

[0094] (4) According to a fourth aspect of this embodiment, in the control device (10) according to the second aspect, the learning unit (1003) determines a learning operation amount for bringing the unit closer to the target posture using a learning model, and updates the learning model based on the deviation between the target posture and the actual posture of the long flexible robot (6) consisting of the target posture, the learning operation amount, and the position of the representative point of the unit after operation with the learning operation amount acquired from the position measurement sensor (7) that measures the position of the representative point of the unit in real time, and the posture velocity target value.

[0095] The control device uses the learning model in this way to determine the learning operation variables, and automatically performs feedback control in real time based on the error between the actual posture based on this learning operation variable and the target posture, thereby updating the learning model in real time, thereby enabling the learning model to be learned more efficiently.

[0096] (5) According to a fifth aspect of the present embodiment, an inspection system (1) includes the control device (10) according to any one of the first to fourth aspects and an inspection device (5).

[0097] (6) According to a sixth aspect of this embodiment, a control method for an inspection device (5) including a long flexible robot (6) in which a plurality of units, each of which can bend to a desired curvature, are connected, and an attitude actuator (659) capable of adjusting the attitude of the units, includes the steps of: specifying a target attitude consisting of the positions of representative points of each of the units proceeding along a predetermined path; and determining an operation amount for bringing the units into the target attitude using a learning model that receives the target attitude of the units as an input and outputs an operation amount of the attitude actuator (659).

[0098] (7) According to a seventh aspect of this embodiment, the program causes an inspection device (5) equipped with a long flexible robot (6) formed by connecting a plurality of units each capable of bending to a desired curvature, and an attitude actuator (65) capable of adjusting the attitude of the units, to execute the steps of: identifying a target attitude consisting of the positions of representative points of each unit proceeding along a predetermined path; and determining an operation amount for bringing the unit into the target attitude using a learning model in which the target attitude of the unit is input and the operation amount of the attitude actuator (65) is output. [Explanation of symbols]

[0099] 1. Inspection system 10 Control device 100 CPU 1001 Target posture identification section 1002 Manipulated amount determination section 1003 Learning Department 1004 Output section 101 Communication Interface 102 memory 103 Input Device 104 Output Device 105 Storage 106 Operating equipment 5. Inspection equipment 6 Long flexible robot 61 Inspection cable 611 Cable body 612 Sensors 62 tubes 62A Active part 62B Driven part 63, 63A, 63B tube body 631 Cylindrical part 632 flange 633 Wire insertion hole 65 Posture Actuator 651 Wire 652 Housing 652A Housing Through Hole 653 Pulley 654 Wire drive unit 655 Wire load detection unit 67 Advance / retreat actuator 671 Advance / retreat drive unit 672 Guide Rail 7 Position measurement sensor 71 Marker J Jacobian matrix L connection part M1 Learning Model M2 forward kinematics model M3 elongation model

Claims

1. A control device for an inspection device comprising a long flexible robot in which a plurality of units, each of which can be bent to a desired curvature, are connected, and an attitude actuator capable of adjusting the attitude of the units, a target attitude specifying unit that specifies a target attitude consisting of the positions of representative points of each of the units moving along a predetermined path; an operation amount determination unit that determines an operation amount for bringing the unit into the target attitude using a learning model that receives the target attitude of the unit as an input and outputs an operation amount of the attitude actuator; a learning unit that constructs the learning model based on the positions of the representative points acquired from a position measurement sensor that measures the three-dimensional positions of the representative points and the operation amounts of the attitude actuators; Equipped with The learning unit a Jacobian matrix representing a sensitivity matrix for converting the velocity of the tip of each unit into the operation velocity of the attitude actuator is calculated from a forward kinematics model having the operation amount of the attitude actuator as an input and the position of the representative point as an output; determining an operation amount of the attitude actuator for moving the unit closer to the target attitude using the Jacobian matrix; a speed of the tip of each unit is calculated from the Jacobian matrix so that an error between the attitude of the unit after operation using the operation amount of the attitude actuator and the target attitude is reduced, and when inching is performed manually using an operating device, the learning model is trained based on the position of the representative point of the unit after inching and the operation amount of the attitude actuator; Control device.

2. A control device for an inspection device comprising a long flexible robot in which a plurality of units, each of which can be bent to a desired curvature, are connected, and an attitude actuator capable of adjusting the attitude of the units, a target attitude specifying unit that specifies a target attitude consisting of the positions of representative points of each of the units moving along a predetermined path; an operation amount determination unit that determines an operation amount for bringing the unit into the target attitude using a learning model that receives the target attitude of the unit as an input and outputs an operation amount of the attitude actuator; a learning unit that constructs the learning model based on the positions of the representative points acquired from a position measurement sensor that measures the three-dimensional positions of the representative points and the operation amounts of the attitude actuators; Equipped with The learning unit a Jacobian matrix representing a sensitivity matrix for converting the velocity of the tip of each unit into the operation velocity of the attitude actuator is calculated from a forward kinematics model having the operation amount of the attitude actuator as an input and the position of the representative point as an output; determining an operation amount of the attitude actuator for bringing the unit closer to the target attitude using the learning model; the Jacobian matrix is ​​used to sequentially correct the manipulated variable of the attitude actuator based on the deviation between the position of the representative point of the unit acquired in real time from the position measurement sensor and the target attitude, and the learning model is automatically updated in real time based on the manipulated variable of the attitude actuator after the correction and the position of the representative point of the unit. Control device.

3. The control device according to claim 1 or 2; the inspection device; An inspection system comprising:

4. A control method for an inspection device including a long flexible robot in which a plurality of units, each of which can be bent to a desired curvature, are connected, and an attitude actuator capable of adjusting the attitude of the units, comprising: specifying a target attitude consisting of the positions of representative points of each of the units moving along a predetermined path; determining an amount of operation for bringing the unit into the target attitude using a learning model that has the target attitude of the unit as an input and the amount of operation of the attitude actuator as an output; constructing the learning model based on the positions of the representative points acquired from a position measurement sensor that measures the three-dimensional positions of the representative points and the operation amounts of the attitude actuators; A control method comprising: The step of constructing the learning model includes: a Jacobian matrix representing a sensitivity matrix for converting the velocity of the tip of each unit into the operation velocity of the attitude actuator is calculated from a forward kinematics model having the operation amount of the attitude actuator as an input and the position of the representative point as an output; determining an operation amount of the attitude actuator for moving the unit closer to the target attitude using the Jacobian matrix; a speed of the tip of each unit is calculated from the Jacobian matrix so that an error between the attitude of the unit after operation using the operation amount of the attitude actuator and the target attitude is reduced, and when inching is performed manually using an operating device, the learning model is trained based on the position of the representative point of the unit after inching and the operation amount of the attitude actuator; Control method.

5. An inspection device including a long flexible robot in which a plurality of units, each of which can be bent to a desired curvature, are connected, and an attitude actuator capable of adjusting the attitude of the units, specifying a target attitude consisting of the positions of representative points of each of the units moving along a predetermined path; determining an amount of operation for bringing the unit into the target attitude using a learning model that has the target attitude of the unit as an input and the amount of operation of the attitude actuator as an output; constructing the learning model based on the positions of the representative points acquired from a position measurement sensor that measures the three-dimensional positions of the representative points and the operation amounts of the attitude actuators; A program for executing The step of constructing the learning model includes: a Jacobian matrix representing a sensitivity matrix for converting the velocity of the tip of each unit into the operation velocity of the attitude actuator is calculated from a forward kinematics model having the operation amount of the attitude actuator as an input and the position of the representative point as an output; determining an operation amount of the attitude actuator for moving the unit closer to the target attitude using the Jacobian matrix; a speed of the tip of each unit is calculated from the Jacobian matrix so that an error between the attitude of the unit after operation using the operation amount of the attitude actuator and the target attitude is reduced, and when inching is performed manually using an operating device, the learning model is trained based on the position of the representative point of the unit after inching and the operation amount of the attitude actuator; program.

6. A control method for an inspection device comprising a long flexible robot in which a plurality of units, each of which can be bent to a desired curvature, are connected, and an attitude actuator capable of adjusting the attitude of the units, comprising: specifying a target attitude consisting of the positions of representative points of each of the units moving along a predetermined path; determining an amount of operation for bringing the unit into the target attitude using a learning model that has the target attitude of the unit as an input and the amount of operation of the attitude actuator as an output; constructing the learning model based on the positions of the representative points acquired from a position measurement sensor that measures the three-dimensional positions of the representative points and the operation amounts of the attitude actuators; A control method comprising: The step of constructing the learning model includes: a Jacobian matrix representing a sensitivity matrix for converting the velocity of the tip of each unit into the operation velocity of the attitude actuator is calculated from a forward kinematics model having the operation amount of the attitude actuator as an input and the position of the representative point as an output; determining an operation amount of the attitude actuator for bringing the unit closer to the target attitude using the learning model; the Jacobian matrix is ​​used to sequentially correct the manipulated variable of the attitude actuator based on the deviation between the position of the representative point of the unit acquired in real time from the position measurement sensor and the target attitude, and the learning model is automatically updated in real time based on the manipulated variable of the attitude actuator after the correction and the position of the representative point of the unit. Control method.

7. An inspection device comprising a long flexible robot in which a plurality of units, each of which can be bent to a desired curvature, are connected, and an attitude actuator capable of adjusting the attitude of the units, specifying a target attitude consisting of the positions of representative points of each of the units moving along a predetermined path; determining an amount of operation for bringing the unit into the target attitude using a learning model that has the target attitude of the unit as an input and the amount of operation of the attitude actuator as an output; constructing the learning model based on the positions of the representative points acquired from a position measurement sensor that measures the three-dimensional positions of the representative points and the operation amounts of the attitude actuators; A program for executing The step of constructing the learning model includes: a Jacobian matrix representing a sensitivity matrix for converting the velocity of the tip of each unit into the operation velocity of the attitude actuator is calculated from a forward kinematics model having the operation amount of the attitude actuator as an input and the position of the representative point as an output; determining an operation amount of the attitude actuator for bringing the unit closer to the target attitude using the learning model; the Jacobian matrix is ​​used to sequentially correct the manipulated variable of the attitude actuator based on the deviation between the position of the representative point of the unit acquired in real time from the position measurement sensor and the target attitude, and the learning model is automatically updated in real time based on the manipulated variable of the attitude actuator after the correction and the position of the representative point of the unit. program.

Images