Data generation device, classification device, data generation method, classification method, and computer program

Abstract

To provide a data generation device that can generate a data at an abnormal time which may actually occur.SOLUTION: A data generation device comprises: a data acquisition unit for acquiring a first anomaly data that is a time-series anomaly data acquired from a target device; and a data generation unit for generating a second anomaly data that is a time-series anomaly data belonging to a second anomaly class, as a result of classification, which is different from the first anomaly class to which the first anomaly data belongs when the first anomaly data is classified on the basis of the first anomaly data.SELECTED DRAWING: Figure 6

Description

Technical Field

[0001] The present disclosure relates to a data generation device, a class classification device, a data generation method, a class classification method, and a computer program.

Background Art

[0002] In recent years, production systems are increasingly connected to a network in order to improve productivity. For this reason, the number of cases where production systems are attacked by cyberattacks is also increasing, and the need for countermeasures is also recognized.

[0003] Targets of cyberattacks cover a wide range, such as equipment operation programs and controllers. In addition to cyberattacks, abnormalities due to physical interference or misoperations can also occur. Therefore, the types and degrees of abnormalities to be determined when equipment malfunctions cover a wide range. In order to build a system for identifying the cause of an abnormality, data for all abnormality classes to be determined is usually required. However, it is difficult to comprehensively obtain data on various types of abnormal operations in terms of cost and resources.

[0004] Patent Document 1 discloses a method of generating abnormal data by adding periodic fluctuations, an increasing or decreasing trend, spikes, movement deviations, or colored noise to normal data.

Prior Art Documents

Patent Documents

[0005]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0006] However, according to the method described in Patent Document 1, abnormal data is generated by processing normal data, but the processing method is not determined based on reflecting actual abnormal data. Therefore, the abnormal data includes data that is unlikely to actually occur. Even if a learning model for discriminating data anomalies is generated using such abnormal data, there is a possibility that the learning model may not be able to accurately discriminate the data at the time of actual anomalies as abnormal.

[0007] The present disclosure has been made to solve such problems, and an object thereof is to provide a data generation device, a data generation method, and a computer program capable of generating data at the time of anomalies that can actually occur.

[0008] Another object of the present disclosure is to provide a class classification device, a class classification method, and a computer program capable of accurately classifying data at the time of actual anomalies.

Means for Solving the Problems

[0009] A data generation device according to an aspect of the present disclosure includes a data acquisition unit that acquires first abnormal data, which is time-series abnormal data obtained from a target device, and based on the first abnormal data, when the first abnormal data is class classified, a data generation unit that generates second abnormal data, which is time-series abnormal data belonging to a second abnormal class different from a first abnormal class to which the first abnormal data belongs.

[0010] The present invention can be realized not only as a data generation device having such a characteristic processing unit, but also as a data generation method having such a characteristic processing as a step, or as a computer program for causing a computer to execute such a step. Further, it can be realized as a semiconductor integrated circuit that realizes part or all of the data generation device, as a class classification device using the data generation device, or as a system including the data generation device.

Advantages of the Invention

[0011] According to the present disclosure, it is possible to generate data during actual abnormal situations that may occur.

Brief Description of the Drawings

[0012]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Figure 13

Figure 14

Figure 15

Figure 16

Figure 17

Figure 18

Figure 19

Figure 20

Figure 21

Figure 22

Figure 23

Figure 24

Figure 25

Figure 26

[0013] [Outline of Embodiments of the Present Disclosure] First, the outline of the embodiments of the present disclosure will be listed and described. (1) A data generation device according to an embodiment of the present disclosure includes a data acquisition unit that acquires first abnormal data, which is time-series abnormal data obtained from a target device, and based on the first abnormal data, when the first abnormal data is classified into classes, a data generation unit that generates second abnormal data, which is time-series abnormal data belonging to a second abnormal class different from the first abnormal class to which the first abnormal data belongs.

[0014] According to this configuration, the second abnormal data is generated using the actually occurring first abnormal data. Therefore, it is possible to generate the second abnormal data that can actually occur. As a result, it is possible to generate the second abnormal data classified into a class different from the first abnormal data.

[0015] (2) In the above (1), the data generation unit may generate the second abnormal data based on data obtained by temporally reversing a part of the first abnormal data.

[0016] According to this configuration, it is possible to accurately generate the second abnormal data when the target device performs an operation temporally reverse to the operation corresponding to the first abnormal data.

[0017] (3) In the above (1) or (2), the data acquisition unit further acquires third abnormal data, which is time-series abnormal data obtained from the target device and belonging to a third abnormal class different from the first abnormal class, and the data generation unit combines a part of each of the first abnormal data and the third abnormal data to generate fourth abnormal data, which is time-series abnormal data belonging to a fourth abnormal class different from the first abnormal class and the third abnormal class.

[0018] According to this configuration, it is possible to generate the fourth abnormal data by combining the first abnormal data and the third abnormal data. That is, it is possible to generate other abnormal data by combining two abnormal data.

[0019] (4) In any of the above (1) to (3), the data generation unit may generate fifth abnormal data, which is time-series abnormal data belonging to a classification result of a fifth abnormal class different from the first abnormal class and the second abnormal class, by combining parts of the first abnormal data and the second abnormal data with each other.

[0020] According to this configuration, fifth abnormal data can be generated by combining the first abnormal data and the second abnormal data. That is, the data generation unit can generate other fifth abnormal data from the second abnormal data generated by itself.

[0021] (5) In any of the above (1) to (4), the data acquisition unit further acquires normal data obtained from the target device and belonging to a classification result of a normal class, and the data generation unit combines parts of the first abnormal data and the normal data with each other to generate sixth abnormal data, which is time-series abnormal data belonging to a classification result of a sixth abnormal class different from the first abnormal class.

[0022] According to this configuration, sixth abnormal data can be generated by combining the first abnormal data and the normal data.

[0023] (6) In any of the above (1) to (5), the target device includes a robotic arm, and the data obtained from the target device may include at least one of acceleration, displacement, angular velocity, position, voltage, current, speed, angle, magnetic field, temperature, pressure, and water pressure of the robotic arm.

[0024] According to this configuration, second abnormal data that can actually occur can be generated for at least one of acceleration, displacement, angular velocity, position, voltage, current, speed, angle, magnetic field, temperature, pressure, and water pressure of the robotic arm.

[0025] (7) In any of the above (1) to (6), based on the time-series normal data belonging to the normal class as a result of the classification, the time-series abnormal data acquired by the data acquisition unit, and the time-series abnormal data generated by the data generation unit, a model generation unit that generates a classification model for classifying the time-series input data obtained from the target device into any of the abnormal classes and the normal class may be further provided.

[0026] According to this configuration, a classification model capable of classifying normal data and each abnormal data can be generated. Thereby, the data at the time of an actually occurring abnormality can be accurately classified.

[0027] (8) The class classification device according to another embodiment of the present disclosure includes a data acquisition unit that acquires time-series input data from a target device, and inputs the time-series input data into a classification model generated by the data generation device described in the above (7), thereby classifying the time-series input data into any of the normal class and each abnormal class, and a class classification unit.

[0028] According to this configuration, using a classification model capable of classifying normal data and each abnormal data, the input data can be classified into any class. Thereby, the data at the time of an actually occurring abnormality can be accurately classified.

[0029] (9) The data generation method according to another embodiment of the present disclosure includes a step of acquiring first abnormal data that is time-series abnormal data obtained from a target device, and a step of generating second abnormal data that is time-series abnormal data belonging to a second abnormal class different from the first abnormal class to which the first abnormal data belongs when the first abnormal data is classified based on the first abnormal data.

[0030] This configuration includes the characteristic processing in the above-described data generation device as steps. Therefore, the same operations and effects as those of the above-described data generation device can be achieved.

[0031] (10) The classification method according to another embodiment of the present disclosure includes the steps of obtaining time-series input data from a target device, and classifying the time-series input data into either a normal class or each abnormal class by inputting the time-series input data into a classification model generated by the data generation device described in (7) above.

[0032] This configuration includes the characteristic processing in the above-described classification device as steps. Therefore, the same operations and effects as those of the above-described classification device can be achieved.

[0033] (11) The computer program according to another embodiment of the present disclosure causes a computer to function as a data acquisition unit that acquires first abnormal data, which is time-series abnormal data obtained from a target device, and a data generation unit that generates second abnormal data, which is time-series abnormal data belonging to a second abnormal class different from a first abnormal class to which the first abnormal data belongs when the first abnormal data is classified.

[0034] According to this configuration, the computer can function as the above-described data generation device. Therefore, the same operations and effects as those of the above-described data generation device can be achieved.

[0035] (12) The computer program according to another embodiment of the present disclosure causes a computer to function as a data acquisition unit that acquires time-series input data from a target device, and a class classification unit that classifies the time-series input data into either a normal class or each abnormal class by inputting the time-series input data into a classification model generated by the data generation device described in (7) above.

[0036] According to this configuration, the computer can function as the above-described classification device. Therefore, the same operations and effects as those of the above-described classification device can be achieved.

[0037] [Details of Embodiments of the Present Disclosure] Hereinafter, embodiments of the present disclosure will be described with reference to the drawings. Note that all the embodiments described below are specific examples of the present disclosure. The numerical values, shapes, materials, components, arrangement positions and connection forms of the components, steps, order of steps, etc. shown in the following embodiments are merely examples and do not limit the present disclosure. Among the components in the following embodiments, the components not described in the independent claims are optional components. Also, each drawing is a schematic diagram and is not necessarily drawn precisely.

[0038] Also, the same reference numerals are given to the same components. Since their functions and names are the same, their descriptions will be omitted as appropriate.

[0039] <Embodiment 1> 〔Overall Configuration of Production System〕 FIG. 1 is a diagram showing an example of the overall configuration of a production system according to Embodiment 1 of the present disclosure. The production system 1 includes a robot arm 200 and a computer 100.

[0040] The robot arm 200 is installed, for example, on a product manufacturing line and operates according to commands from a control device (not shown) to perform operations such as attaching parts to semi-finished products, painting, or welding.

[0041] The robot arm 200 includes a base 201, joints 202A, 202B, links 203A, 203B, 203C, an end effector 204, and an acceleration sensor 205. Hereinafter, when the joints 202A, 202B are not distinguished, they are referred to as joint 202. The robot arm 200 is an example of a target device for acquiring abnormal data such as acceleration.

[0042] The base 201 is a component for fixing the robot arm 200 to the floor or the like.

[0043] The first end of link 203A is connected to base 201, and the second end of link 203A is connected to joint 202A. The first end of link 203B is connected to joint 202A, and the second end of link 203B is connected to joint 202B. The first end of link 203C is connected to joint 202B.

[0044] End effector 204, also called a robot hand, is connected to the second end side of link 203C and performs operations such as gripping, processing, screwing, painting, and welding of parts.

[0045] Base 201 and joint 202 are equipped with servo motors inside. It is assumed that base 201 is fixed to a horizontal plane. Link 203A rotates in the horizontal direction by a servo motor provided inside base 201. Also, link 203B rotates in the vertical direction by a servo motor provided inside joint 202A, and link 203C rotates in the vertical direction by a servo motor provided inside joint 202B.

[0046] Acceleration sensor 205 is provided on end effector 204.

[0047] Figure 2 is a diagram showing an example of acceleration sensor 205 installed on end effector 204. The coordinate axes of acceleration sensor 205 are composed of an X-axis parallel to the axis direction of end effector 204, and a Y-axis and a Z-axis orthogonal to the X-axis. Here, the direction of the X-axis of end effector 204 is, for example, the direction from the root part of end effector 204 (the connection part of end effector 204 with link 203C) to the tip of end effector 204. Also, the Y-axis is arranged in a direction orthogonal to the X-axis on the upper surface of end effector 204. Also, the Z-axis is arranged in a direction away from end effector 204 upward. Acceleration sensor 205 detects the acceleration in the X-axis direction, Y-axis direction, and Z-axis direction of end effector 204 and outputs the detection result.

[0048] The acceleration sensor 205 is composed of a piezoelectric acceleration sensor, a servo acceleration sensor, a strain gauge acceleration sensor, a semiconductor acceleration sensor, or the like.

[0049] FIG. 3 is a diagram showing an example of the measurement result by the acceleration sensor 205. The horizontal axis in FIG. 3 indicates the measurement time of the acceleration, and the vertical axis indicates the measured acceleration in the X-axis direction. In FIG. 3, the acceleration is indicated by a circle mark, and time-series data of the acceleration in the X-axis direction is output from the acceleration sensor 205. Note that the acceleration sensor 205 also outputs time-series data of the acceleration in the Y-axis direction and time-series data of the acceleration in the Z-axis direction in the same manner.

[0050] The computer 100 is connected to the acceleration sensor 205 and acquires the acceleration of the end effector 204 from the acceleration sensor 205. The computer 100 determines whether the robot arm 200 is operating normally or abnormally based on the acquired acceleration. The operation determination process of the robot arm 200 by the computer 100 will be described later.

[0051] 〔Operation of the robot arm 200〕 FIG. 4 is a diagram showing an example of the operation of the robot arm 200.

[0052] The control device of the robot arm 200 controls the rotation angle, rotation speed, etc. of the servo motors provided inside the base 201 and the joint 202 so that the robot arm 200 performs a predetermined operation. The predetermined operation is, for example, an operation in which the end effector 204 moves from the stationary position 210 ((0) stationary position) to the upper right of the paper surface along the path ((1) forward path), performs an up-and-down motion including a downward motion and an upward motion starting from a predetermined position ((2) up-and-down motion), moves along the path ((3) return path) from the predetermined position to the stationary position 210, and returns to the stationary position 210 ((0) stationary position). Note that hereinafter, the operation is also referred to as motion, and the motion is also referred to as operation.

[0053] More specifically, the control device sets the stationary position of the end effector 204, the operating speed (moving speed) of the forward path of the end effector 204 (hereinafter referred to as the "forward path speed"), the turning-back position in the vertical movement of the end effector 204, and the operating speed (moving speed) of the return path of the end effector 204 (hereinafter referred to as the "return path speed") as control parameters of the robot arm 200, and controls the rotation angle, rotation speed, etc. of the servo motors provided inside the base 201 and the joints 202 so as to perform the above-described predetermined operation according to the control parameter values.

[0054] 〔Configuration of Computer 100〕 FIG. 5 is a block diagram showing an example of the hardware configuration of the computer 100.

[0055] The computer 100 includes an I / F (interface) unit 110, a storage device 120, a processor 130, and a bus 140.

[0056] The I / F unit 110 connects the external device and the computer 100. Here, the I / F unit 110 is connected to the acceleration sensor 205 of the robot arm 200 shown in FIG. 1 and receives acceleration data from the acceleration sensor 205. The I / F unit 110 and the acceleration sensor 205 are connected by wire or wirelessly.

[0057] The storage device 120 is composed of a volatile memory element such as SRAM (Static Random Access Memory) or DRAM (Dynamic Random Access Memory), a non-volatile memory element such as flash memory or EEPROM (Electrically Erasable Programmable Read-Only Memory), or a magnetic storage device such as a hard disk.

[0058] The memory device 120 stores a data generation program 121 and a class classification program 122, which are computer programs executed by the processor 130. The memory device 120 also stores data used or generated during the execution of the computer program. The data includes a classification model 123 for classifying acceleration data. Details of the classification model 123 will be described later.

[0059] The processor 130 is composed of various processors suitable for controlling the computer 100, such as a CPU (Central Processing Unit), an MPU (Micro Processing Unit), and a GPU (Graphics Processing Unit). The processor 130 reads and executes the data generation program 121 and the class classification program 122 stored in the memory device 120. Note that the processor 130 may be a DSP (Digital Signal Processor), an FPGA (Field Programmable Gate Array), or an ASIC (Application Specific Integrated Circuit) configured to execute the processing procedures indicated by the data generation program 121 and the class classification program 122.

[0060] 〔Functional Configuration of Data Generation Device〕 FIG. 6 is a block diagram showing an example of the functional configuration of the data generation device.

[0061] The data generation device 10 is functionally realized by executing the data generation program 121 on the processor 130. The data generation device 10 includes a data acquisition unit 11, a data generation unit 12, and a model generation unit 13.

[0062] The data acquisition unit 11 acquires the time-series acceleration data measured by the acceleration sensor 205. The data acquisition unit 11 may acquire the acceleration data in all three axial directions, or may acquire the acceleration data in some axial directions. Hereinafter, when referring to acceleration data, it also means the time-series acceleration data.

[0063] Note that the data acquisition unit 11 may send the label of the class for which acquisition is desired to the control device of the robot arm 200 and acquire the acceleration data belonging to the class. The labels include, for example, "Normal" which is the label of the normal class, and "Outward path abnormality", "Vertical movement abnormality", "Return path abnormality", "Both outward and return paths are abnormal", "Outward path and vertical movement are abnormal", "Vertical movement and return path are abnormal", and "All of outward path, vertical movement, and return path are abnormal" which are the labels of the abnormal classes.

[0064] The classes with the labels "Outward path abnormality", "Vertical movement abnormality", and "Return path abnormality" are the classes including acceleration data with abnormal outward path movement, the class including acceleration data with abnormal vertical movement, and the class including acceleration data with abnormal return path movement, respectively.

[0065] The class with the label "Both outward and return paths are abnormal" is the class including acceleration data with abnormal outward path movement and return path movement. The class with the label "Outward path and vertical movement are abnormal" is the class including acceleration data with abnormal outward path movement and vertical movement. The class with the label "Vertical movement and return path are abnormal" is the class including acceleration data with abnormal vertical movement and return path movement. The class with the label "All of outward path, vertical movement, and return path are abnormal" is the class including acceleration data with all of outward path movement, vertical movement, and return path movement being abnormal.

[0066] For example, the data acquisition unit 11 sends the label "normal" to the control device of the robotic arm 200. The control device that receives the label "normal" sets predetermined normal values for the forward speed, the turning-back position in the vertical movement, and the return speed of the robotic arm 200. The control device controls the robotic arm 200 so that the robotic arm 200 executes the above-described predetermined operation according to the set respective speeds. The data acquisition unit 11 acquires, from the acceleration sensor 205, acceleration data classified into the class of the label "normal" (hereinafter referred to as "acceleration data of the label 'normal'". The same applies to other labels) while the robotic arm 200 is executing the predetermined operation.

[0067] Also, for example, the data acquisition unit 11 sends the label "abnormal forward path" to the control device of the robotic arm 200. The control device that receives the label "abnormal forward path" sets a predetermined abnormal value for the forward speed of the robotic arm 200 and sets predetermined normal values for the turning-back position in the vertical movement and the return speed. The control device controls the robotic arm 200 so that the robotic arm 200 executes the above-described predetermined operation according to the set respective speeds. The data acquisition unit 11 acquires, from the acceleration sensor 205, acceleration data of the label "abnormal forward path" while the robotic arm 200 is executing the predetermined operation. The data acquisition unit 11 can similarly acquire acceleration data of other labels.

[0068] The data generation unit 12 generates abnormal data based on the data acquired by the data acquisition unit 11. Here, the abnormal data is acceleration data that belongs to the result of class classification into any one of the above-described seven abnormal classes. Note that the normal data is acceleration data that belongs to the result of class classification into the normal class.

[0069] The data generation unit 12 generates second abnormal data that belongs to the result of class classification into a second abnormal class different from the first abnormal class based on the first abnormal data that belongs to the first abnormal class by class classification.

[0070] The following describes two methods for generating the second abnormal data. <<First Generation Method>> Based on data obtained by temporally reversing a part of the abnormal data belonging to a certain abnormal class, the data generation unit 12 generates abnormal data belonging to another abnormal class.

[0071] Figure 7 is a flowchart showing an example of a method for generating abnormal data. The data acquisition unit 11 acquires acceleration data with the label "return path abnormality" from the acceleration sensor 205 (step S11). Figure 8 is a diagram showing an example of a graph of acceleration data with the label "return path abnormality". The horizontal axis of Figure 8 indicates the measurement time of acceleration, and the vertical axis indicates the measured acceleration in the Z-axis direction.

[0072] The data generation unit 12 divides the acquired acceleration data with the label "return path abnormality" into a reciprocating motion part and another part (stationary part) (step S12). Note that a method for detecting a boundary (for example, a boundary between the reciprocating motion part and another part) in the acceleration data will be described later. Referring to Figure 4, the reciprocating motion is, for example, an operation in which the end effector 204 moves from the stationary position 210 ((0) stationary position) to (1) the forward path, performs (2) an up-and-down motion starting from a predetermined position, moves from the predetermined position to (3) the return path, and returns to the stationary position 210 ((0) stationary position).

[0073] The data generation unit 12 temporally reverses the reciprocating motion part by horizontally flipping the reciprocating motion part (the part to the left of the dashed line 601 in Figure 8) divided in step S12 on the graph (step S13).

[0074] The data generation unit 12 generates acceleration data by combining the reciprocating motion part after inversion in step S13 and the other part divided in step S12 (step S14).

[0075] FIG. 9 is a diagram showing an example of a graph of acceleration data obtained as a result of combination. The vertical axis and the horizontal axis in FIG. 9 are the same as those in FIG. 8. That is, the reciprocating motion part in FIG. 9 (the part to the left of the broken line 601 in FIG. 9) is a time-reversed version of the reciprocating motion part in FIG. 8 (the part to the left of the broken line 601 in FIG. 8). Also, the stationary part in FIG. 9 (the part to the right of the broken line 601 in FIG. 9) is the same as the stationary part in FIG. 8 (the part to the right of the broken line 601 in FIG. 8).

[0076] The acceleration data generated in step S14 is acceleration data with the label "outward path anomaly". The reciprocating motion is a temporally symmetric motion. Therefore, the data generation unit 12 can generate the reciprocating motion part of the acceleration data when there is an abnormality in the outward path speed by inverting the reciprocating motion part of the acceleration data when there is an abnormality in the return path speed.

[0077] Here, an example of the process of generating the acceleration data with the label "outward path anomaly" from the acceleration data with the label "return path anomaly" has been described. However, as long as the part of the acceleration data that is a temporally symmetric motion can be extracted, the acceleration data of other labels can be generated in the same way. For example, the data generation unit 12 may extract the reciprocating motion part from the acceleration data with the label "outward path anomaly", invert the reciprocating motion part in time, and combine it with other parts to generate the acceleration data with the label "return path anomaly". Similarly, the data generation unit 12 can also generate the acceleration data with the label "up and down motion and return path anomaly" by performing the same inversion process as above on the acceleration data with the label "outward path and up and down motion anomaly". Also, the data generation unit 12 can generate the acceleration data with the label "outward path and up and down motion anomaly" by performing the same inversion process as above on the acceleration data with the label "up and down motion and return path anomaly".

[0078] ≪Second generation method≫ The data generation unit 12 generates abnormal data belonging to a classification result in another class different from the plurality of classes by combining a part of the abnormal data belonging to each of the plurality of classes with each other.

[0079] Figure 10 is a flowchart showing an example of a method for generating abnormal data. The data acquisition unit 11 acquires acceleration data with the label "abnormal vertical movement" from the acceleration sensor 205 (step S21).

[0080] Figure 11 is a diagram showing an example of a graph of acceleration data with the label "abnormal vertical movement". The horizontal axis and the vertical axis in Figure 11 are the same as those in Figure 8. Note that the graphs in Figures 12 and 13 described later are also the same as those in Figure 8. In Figure 11, the acceleration data of the vertical movement part is shown by a solid-line rectangular frame 701A.

[0081] The data acquisition unit 11 acquires acceleration data with the label "abnormal return path" from the acceleration sensor 205 (step S22).

[0082] Figure 12 is a diagram showing an example of a graph of acceleration data with the label "abnormal return path". In Figure 12, the acceleration data of the return path part is shown by a solid-line rectangular frame 702A.

[0083] The data generation unit 12 divides the acceleration data with the label "abnormal vertical movement" acquired in step S21 (Figure 11) at the time of the boundary between (2) vertical movement and (3) return path shown in Figure 4 (step S23).

[0084] The data generation unit 12 divides the acceleration data with the label "abnormal return path" acquired in step S22 (Figure 12) at the time of the boundary between (2) vertical movement and (3) return path shown in Figure 4 (step S24).

[0085] The data generation unit 12 extracts the acceleration data from (1) the forward path to (2) the vertical movement shown in FIG. 4, which is divided in step S23, from the acceleration data with the label "abnormal vertical movement" (step S25). For example, the data generation unit 12 extracts the acceleration data that includes the rectangular frame 701A in FIG. 11 and is to the left of the rectangular frame 701A. That is, the data generation unit 12 extracts the acceleration data included in the dashed rectangular frame 701B.

[0086] The data generation unit 12 extracts the acceleration data after (3) the return path shown in FIG. 4, which is divided in step S24, from the acceleration data with the label "abnormal return path" (step S26). For example, the data generation unit 12 extracts the acceleration data that includes the rectangular frame 702A in FIG. 12 and is to the right of the rectangular frame 702A. That is, the data generation unit 12 extracts the acceleration data included in the dashed rectangular frame 702B.

[0087] The data generation unit 12 generates acceleration data by combining the acceleration data extracted in step S25 and the acceleration data extracted in step S26 (step S27).

[0088] FIG. 13 is a diagram showing an example of a graph of the acceleration data obtained as a result of the combination. That is, the graph is a graph obtained by combining the acceleration data included in the rectangular frame 701B and the acceleration data included in the rectangular frame 702B. The acceleration data generated in step S27 is the acceleration data with the label "abnormal vertical movement and return path". In this way, the data generation unit 12 extracts from two acceleration data with different labels the parts including abnormal operations so that they do not overlap with each other temporally. The data generation unit 12 can generate acceleration data including two abnormal operations by combining the partial data of the acceleration data including the two extracted abnormal operations.

[0089] Note that the acceleration data of the two labels acquired by the data acquisition unit 11 is not limited to the above. For example, the data acquisition unit 11 acquires the acceleration data of each of the labels "outward path anomaly" and "return path anomaly. The data generation unit 12 extracts partial data including the acceleration data of the outward path movement from the label "outward path anomaly", and extracts partial data including the acceleration data of the return path movement from the label "return path anomaly". The data generation unit 12 generates the acceleration data of the label "both the outward path and the return path are abnormal" by combining the two partial data.

[0090] Also, the number of acceleration data acquired by the data acquisition unit 11 is not limited to two, and may be three or more. For example, by combining the partial data of the acceleration data of each of the labels "outward path anomaly", "return path anomaly", and "vertical movement anomaly", the acceleration data of the label "all of the outward path, vertical movement, and return path are abnormal" may be generated.

[0091] Referring to FIG. 6 again, the model generation unit 13 generates a classification model 123 that classifies the input data (acceleration data) acquired from the acceleration sensor 205 into either the normal class or each abnormal class based on the normal data and abnormal data acquired by the data acquisition unit 11 and the abnormal data acquired by the data generation unit 12. That is, the model generation unit 13 generates the classification model 123 using the normal data and each abnormal data as teacher data.

[0092] The classification model 123 is, for example, a classifier learned by TSF (Time Series Forest). TSF is a supervised learning method that performs classification by the random forest method. Note that the classification model 123 is not limited to the above classifier. The classification model 123 may be, for example, a CNN (Convolutional Neural Network) or an RNN (Recurrent Neural Network) machine-learned by a method such as deep learning, or an SVM (Support Vector Machine). Further, the classification model 123 may be a machine learning model of the k-nearest neighbor method.

[0093] 〔Operation of data generation device 10〕 FIG. 14 is a flowchart showing an example of the operation of the data generation device 10. The data generation device 10 acquires the acceleration data in the Z-axis direction of each of the labels "normal", "abnormal up and down movement", and "abnormal return path" from the acceleration sensor 205 (step S31). The data generation device 10 acquires, for example, the acceleration data during the time from when the robot arm 200 starts operating from the stationary position 210, returns to the stationary position 210, and starts the next operation.

[0094] An example of the graph of the acceleration data of the label "abnormal up and down movement" is the same as the graph shown in FIG. 11. An example of the graph of the acceleration data of the label "abnormal return path" is the same as the graph shown in FIG. 12.

[0095] The data generation device 10 generates the acceleration data of the label "abnormal forward path" by inverting the partial data of the acceleration data of the label "abnormal return path" (step S32). An example of the details of the process of step S32 is as described with reference to FIG. 7.

[0096] FIG. 15 is a diagram showing an example of a graph of acceleration data of the label "outbound anomaly". The horizontal axis and the vertical axis of the graph in FIG. 15 are the same as those in FIG. 8. The same applies to the graph in FIG. 16 described later. By the process of step S32, for example, the acceleration data of the label "return anomaly" shown in FIG. 12 is used to generate the acceleration data of the label "outbound anomaly" shown in FIG. 15. In FIG. 15, the acceleration data of the outbound part is indicated by a solid-line rectangular frame 705A.

[0097] The data generation device 10 generates the acceleration data of each of the labels "outbound and return are abnormal", "outbound and vertical movement are abnormal", "vertical movement and return are abnormal", and "outbound, vertical movement, and return are all abnormal" by cutting and pasting the acceleration data acquired in step S31 or the acceleration data generated in step S32 (step S33). An example of the details of the process of step S33 is as described with reference to FIG. 10.

[0098] FIG. 16 is a diagram showing an example of a graph of acceleration data of the label "outbound, vertical movement, and return are all abnormal". By the process of step S33, for example, the acceleration data of the outbound movement indicated by the rectangular frame 705A in FIG. 15, the acceleration data of the vertical movement indicated by the rectangular frame 701A in FIG. 11, and the acceleration data of the return movement indicated by the rectangular frame 702A in FIG. 12 are included, and the acceleration data on the right side of the rectangular frame 702A (the acceleration data included in the dashed-line rectangular frame 702B) are combined to generate the acceleration data shown in FIG. 16.

[0099] The data generation device 10 generates a classification model 123 that classifies the input data acquired from the acceleration sensor 205 into one of eight classes using the acceleration data of eight labels acquired or generated from step S31 to step S33 as teacher data (step S34).

[0100] [Regarding the method for detecting boundaries in acceleration data] Next, a method for detecting the boundary between two adjacent operations in the acceleration data, which is executed by the data generation unit 12 of the data generation device 10, will be described.

[0101] In the first generation method shown in FIG. 7, the data generation unit 12 detected the boundary between the reciprocating motion part and the other part (stationary part), and divided the two. Here, the boundary can be detected by any one of Detection Example 1 to Detection Example 3 below. (Detection Example 1) Set the time immediately before a certain time (for example, 1 second) at the end of the cycle as the boundary (Detection Example 2) Set the time when a certain time has elapsed since the start of the cycle as the boundary (Detection Example 3) Set the time when the value of the acceleration data reaches the value at rest as the boundary

[0102] The period of the acceleration data is specified as follows, for example. The acceleration in the Z-axis direction when the end effector 204 is at rest is ideally 1.0G. Therefore, the data generation unit 12 sets the start and end of the cycle to the time immediately before the second predetermined time (for example, 0.01 second) when the acceleration data first exceeds the threshold value 1.1 after the state where the acceleration data is equal to or less than the threshold value 1.1 continues for the first predetermined time (for example, 0.5 second).

[0103] In Detection Example 1 and Detection Example 2, it is assumed that the time of the reciprocating motion part or the stationary part is known in advance. In Detection Example 3, the data generation unit 12 may, for example, set the time when the acceleration data first becomes equal to or less than the threshold value 1.1 as the boundary when the state where the acceleration data is equal to or less than the threshold value 1.1 continues for the first predetermined time (for example, 0.5 second). Alternatively, the data generation unit 12 may set the time immediately after the third predetermined time (for example, 0.01 second) when the acceleration data first exceeds the threshold value 1.1 as the boundary when looking at the acceleration data while going back in time from the end of the cycle.

[0104] In the second generation method shown in FIG. 10, the data generation unit 12 divided the acceleration data at the boundary between adjacent operations. Here, the boundary can be detected by any one of Detection Example 4 to Detection Example 6 below. (Detection Example 4) Using the time when the value of the acceleration data exceeds or falls below the threshold as the boundary (Detection Example 5) Using the time when the value of the acceleration data reaches a maximum or minimum as the boundary (Detection Example 6) Using the time when a certain period of time has elapsed since the start of the cycle as the boundary

[0105] FIG. 17 is a diagram showing an example of a graph of the acceleration in the Y-axis direction of the end effector 204 for one cycle. The horizontal axis of FIG. 17 indicates the measurement time of the acceleration, and the vertical axis indicates the measured acceleration in the Y-axis direction.

[0106] For example, according to Detection Example 4, the data generation unit 12 detects, as the boundary between the forward movement and the vertical movement, the time when the acceleration in the Y-axis direction falls below a predetermined threshold. Assume that the data generation unit 12 starts comparing the acceleration with the threshold in order of the measurement time from the end of the forward movement. It is assumed that the approximate time of the forward movement is known in advance. Therefore, it is also assumed that the end of the forward movement is known in advance. In this case, assume that the acceleration at the position indicated by the arrow 501 first falls below the threshold. The data generation unit 12 detects the time corresponding to the acceleration that first falls below the threshold as the boundary between the forward movement and the vertical movement.

[0107] Also, according to Detection Example 4, the data generation unit 12 detects, as the boundary between the vertical movement and the return movement, the time when the acceleration in the Y-axis direction exceeds a predetermined threshold. Assume that the data generation unit 12 starts comparing the acceleration with the threshold in order of the measurement time from the start of the vertical movement. Assume that the start time of the vertical movement is detected as the time corresponding to the acceleration at the position indicated by the arrow 501. In this case, assume that the acceleration at the position indicated by the arrow 502 first exceeds the threshold. The data generation unit 12 detects the time corresponding to the acceleration that first exceeds the threshold as the boundary between the vertical movement and the return movement.

[0108] FIG. 18 is a diagram showing an example of a graph of the acceleration in the Z-axis direction of the end effector 204 for one cycle. The horizontal axis of FIG. 18 indicates the measurement time of the acceleration, and the vertical axis indicates the measured acceleration in the Z-axis direction.

[0109] For example, according to Detection Example 5, the data generation unit 12 detects the time when the acceleration in the Z-axis direction reaches its maximum as the boundary between the forward movement and the vertical movement. Assume that the data generation unit 12 has detected the maximum acceleration at the position indicated by arrow 503 near the boundary between the forward movement and the vertical movement. It is assumed that the time near the boundary between the forward movement and the vertical movement is known in advance. The data generation unit 12 detects the time corresponding to the maximum acceleration at the position indicated by arrow 503 as the boundary between the forward movement and the vertical movement.

[0110] Also, according to Detection Example 5, the data generation unit 12 detects the time when the acceleration in the Z-axis direction reaches its maximum as the boundary between the vertical movement and the return movement. Assume that the data generation unit 12 has detected the maximum acceleration at the position indicated by arrow 504 near the boundary between the vertical movement and the return movement. It is assumed that the time near the boundary between the vertical movement and the return movement is known in advance. The data generation unit 12 detects the time corresponding to the maximum acceleration at the position indicated by arrow 504 as the boundary between the vertical movement and the return movement.

[0111] Also, the data generation unit 12 determines that the end effector 204 is stationary at the stationary position 210 from 1 second before the end time of the cycle to the end time for 1 second.

[0112] 〔Functional Configuration of Classifying Device〕 FIG. 19 is a block diagram showing an example of the functional configuration of the classifying device. The classifying device 20 is functionally realized by executing the classification program 122 on the processor 130. The classifying device 20 includes a data acquisition unit 21, a classification unit 22, and an output unit 23.

[0113] The data acquisition unit 21 acquires input data to be input to the classification model 123. Specifically, the data acquisition unit 21 acquires the acceleration data in the Z-axis direction output from the acceleration sensor 205 at a predetermined time interval. The data acquisition unit 21 acquires the acceleration data for one cycle. Here, one cycle may be, for example, the time from when the robot arm 200 starts operating from the stationary position 210 until it returns to the stationary position 210 and starts the next operation, or it may be a predetermined time width.

[0114] The class classification unit 22 classifies the input data acquired by the data acquisition unit 21 into one of a normal class and a plurality of abnormal classes by inputting the input data into the classification model 123.

[0115] The output unit 23 outputs the classification result in the class classification unit 22. For example, the output unit 23 may cause the classification result to be displayed by outputting it to a display device such as a display. Also, the output unit 23 may transmit the classification result to the control device of the robot arm 200 or an external computer. Thereby, the control device can take measures to correct the operation of the robot arm 200 or stop the robot arm 200 based on the classification result.

[0116] 〔Operation of the class classification device 20〕 FIG. 20 is a flowchart showing an example of the operation of the class classification device 20. The class classification device 20 acquires the acceleration data for one cycle in the Z-axis direction of the end effector 204 from the acceleration sensor 205 (step S41).

[0117] The class classification device 20 inputs the acquired acceleration data for one cycle into the classification model 123 (step S42).

[0118] The class classification device 20 acquires the accuracy of the input acceleration data belonging to each of the eight classes from the classification model 123, and determines that the input acceleration data belongs to the class with the highest accuracy (step S43).

[0119] The classification device 20 outputs the determination result in step S43 to a display device to be displayed, or transmits it to the control device of the robot arm 200 or an external computer (step S44).

[0120] As described above, according to the first embodiment of the present disclosure, abnormal data belonging to a certain abnormal class is used to generate abnormal data belonging to another abnormal class. Therefore, it is possible to generate abnormal data that can actually occur.

[0121] Specifically, by temporally reversing a part of the abnormal data belonging to a certain abnormal class, it is possible to generate abnormal data belonging to another abnormal class. Therefore, it is possible to accurately generate abnormal data when the robot arm 200 performs an operation that is temporally reverse to the operation corresponding to the abnormal data belonging to a certain abnormal class.

[0122] In addition, by combining a part of two or more abnormal data having different classes from each other, it is possible to generate abnormal data belonging to another abnormal class.

[0123] In addition, the two or more abnormal data to be combined are not limited to those acquired from the acceleration sensor 205, and may be abnormal data generated by the data generation device 10.

[0124] In addition, the data generation device 10 can generate a classification model 123 that can classify abnormal data acquired or generated by the data generation device 10. Thereby, it is possible to accurately classify the data at the time of an actually occurring abnormality.

[0125] In addition, the classification device 20 can classify input data into any class using a classification model that can classify normal data and each abnormal data. Thereby, it is possible to accurately classify the data at the time of an actually occurring abnormality.

[0126] <Modification Example 1 of Embodiment 1> FIG. 21 is a diagram showing an example of the overall configuration of a production system according to Modification 1 of Embodiment 1 of the present disclosure. The production system 1A includes a robot arm 200A and a computer 100.

[0127] The robot arm 200A is a device similar to the robot arm 200. However, it is provided with an acceleration sensor 205A instead of the acceleration sensor 205. The acceleration sensor 205A is provided on the link 203C.

[0128] FIG. 22 is a diagram showing an example of the acceleration sensor 205A provided on the link 203C. The coordinate axes of the acceleration sensor 205A are composed of an X-axis parallel to the axis direction of the link 203C, a Y-axis orthogonal to the X-axis, and a Z-axis. Here, the axis direction of the link 203C is, for example, the direction of the central axis of the link 203C, which is the direction from the joint 202B toward the end effector 204. Also, the Z-axis is arranged in the direction away from the link 203C. The acceleration sensor 205A detects the accelerations in the X-axis direction, Y-axis direction, and Z-axis direction of the link 203C and outputs the detection results.

[0129] The acceleration sensor 205A is composed of a piezoelectric acceleration sensor, a servo acceleration sensor, a strain gauge type acceleration sensor, a semiconductor type acceleration sensor, or the like.

[0130] The operation of the robot arm 200A is the same as the operation of the robot arm 200 shown in FIG. 4. The configuration of the computer 100 is the same as that in Embodiment 1. The acceleration sensor 205A is connected to the I / F unit 110 of the computer 100.

[0131] FIG. 23 is a flowchart showing an example of the operation of the data generation device 10. The data generation device 10 acquires the acceleration data in the Z-axis direction of each of the label "normal" and the label "both the forward path, vertical movement, and return path are abnormal" from the acceleration sensor 205A (step S51). The data generation device 10 acquires, for example, the acceleration data during the time from when the robot arm 200 starts operating from the stationary position 210, returns to the stationary position 210, and starts the next operation.

[0132] The data generation device 10 generates the acceleration data of each of the label "forward path abnormal", the label "vertical movement abnormal", the label "return path abnormal", the label "both the forward path and return path are abnormal", the label "both the forward path and vertical movement are abnormal", and the label "both the vertical movement and return path are abnormal" by cutting and pasting the acceleration data acquired in step S51 (step S52). For example, the data generation device 10 extracts the acceleration data of the forward path movement from the acceleration data of the label "both the forward path, vertical movement, and return path are all abnormal". The data generation device 10 extracts the acceleration data other than the forward path movement from the acceleration data of the label "normal". The data generation device 10 generates the acceleration data of the label "forward path abnormal" by combining the two extracted acceleration data. The data generation device 10 can similarly generate the acceleration data of other classes by combining the acceleration data of a part of the movement extracted from the acceleration data of the label "both the forward path, vertical movement, and return path are all abnormal" and the acceleration data of other than the above-mentioned part of the movement extracted from the acceleration data of the label "normal".

[0133] The data generation device 10 uses the acceleration data of the two labels acquired in step S51 and the acceleration data of the six labels generated in step S52 as teacher data to generate a classification model 123 that classifies the input data (acceleration data) acquired from the acceleration sensor 205A into one of eight classes (step S53). An example of the operation of the class classification device 20 is the same as that shown in FIG. 20.

[0134] According to Modification 1, abnormal data of other classes can be generated by combining normal data and abnormal data of a certain class.

[0135] <Modification Example 2 of Embodiment 1> FIG. 24 is a diagram showing an example of the overall configuration of a production system according to Modification Example 2 of Embodiment 1 of the present disclosure. The production system 1B includes a robot arm 200B and a computer 100.

[0136] The robot arm 200B is the same device as the robot arm 200. However, it is provided with displacement sensors 206A and 206B instead of the acceleration sensor 205.

[0137] The displacement sensor 206A is provided on the link 203C, detects the displacement (amount of movement) from the reference position, and outputs the detection result. The displacement sensor 206B is provided on the end effector 204, detects the displacement (amount of movement) from the reference position, and outputs the detection result.

[0138] The operation of the robot arm 200B is the same as the operation of the robot arm 200 shown in FIG. 4. Here, the reference positions of the displacement sensor 206A and the displacement sensor 206B can be the positions of the displacement sensor 206A and the displacement sensor 206B when the end effector 204 is present at the stationary position 210, respectively. The configuration of the computer 100 is the same as that in Embodiment 1. The displacement sensors 206A and 206B are connected to the I / F unit 110 of the computer 100.

[0139] FIG. 25 is a flowchart showing an example of the operation of the data generation device 10. The data generation device 10 acquires displacement data classified into the class with the label "normal" and displacement data classified into the class with the label "outward path anomaly" from the displacement sensors 206A and 206B (step S61). That is, the data acquisition unit 11 of the data generation device 10 transmits the label "normal" to the control device of the robot arm 200B. The control device that has received the label "normal" sets predetermined normal values for the outward path speed, the turning-back position in the vertical movement, and the return path speed of the robot arm 200B. The control device controls the robot arm 200B so that the robot arm 200B executes the above-described predetermined operation according to the set speeds. The data acquisition unit 11 acquires displacement data classified into the class with the label "normal" (hereinafter referred to as "displacement data with the label 'normal'"; the same applies to other labels) from the displacement sensors 206A and 206B while the robot arm 200B is executing the predetermined operation. The displacement data for each label is assumed to include the displacement data acquired from the displacement sensor 206A and the displacement data acquired from the displacement sensor 206B.

[0140] Also, the data acquisition unit 11 transmits the label "outward path anomaly" to the control device of the robot arm 200B. The control device that has received the label "outward path anomaly" sets a predetermined abnormal value for the outward path speed of the robot arm 200B and sets predetermined normal values for the turning-back position in the vertical movement and the return path speed. The control device controls the robot arm 200B so that the robot arm 200B executes the above-described predetermined operation according to the set speeds. The data acquisition unit 11 acquires displacement data with the label "outward path anomaly" from the displacement sensors 206A and 206B while the robot arm 200B is executing the predetermined operation.

[0141] The data generation device 10 generates displacement data with the label "return path anomaly" by inverting the partial data of the displacement data with the label "outward path anomaly" acquired from the displacement sensors 206A and 206B (step S62). An example of the details of the process in step S62 is the same as that described with reference to FIG. 7 except that the data to be inverted has changed to the displacement data with the label "outward path anomaly".

[0142] The data generation device 10 uses the displacement data with the label "normal" and the displacement data with the label "forward path anomaly" acquired in step S61, and the displacement data with the label "return path anomaly" generated in step S62 as teacher data to generate a classification model 123 that classifies the input data (displacement data) acquired from the displacement sensors 206A and 206B into any of three classes (step S63).

[0143] Figure 26 is a flowchart showing an example of the operation of the class classification device 20. The class classification device 20 acquires displacement data for one cycle of the link 203C and the end effector 204 from the displacement sensors 206A and 206B (step S71).

[0144] The class classification device 20 inputs the displacement data for one cycle acquired in step S71 into the classification model 123 (step S72).

[0145] The class classification device 20 acquires the accuracy with which the input displacement data belongs to each of the three classes from the classification model 123, and determines that the input displacement data belongs to the class with the highest accuracy (step S73).

[0146] The class classification device 20 outputs the determination result in step S73 to the display device for display, or transmits it to the control device of the robot arm 200 or an external computer (step S74).

[0147] According to Modification 2, the operation of the robot arm 200C can be classified into classes using the displacement sensors 206A and 206B.

[0148] <Modification 3 of Embodiment 1> In Embodiment 1, the acceleration data in the Z-axis direction output by the acceleration sensor 205 is used as normal data and abnormal data, but the classification model 123 can also be generated in the same manner for the acceleration data in the X-axis direction and the Y-axis direction output by the acceleration sensor 205.

[0149] Alternatively, instead of the acceleration sensor 205, other sensors such as a gyro sensor, a position sensor, a voltage sensor, a current sensor, a speed sensor, an angle sensor, a magnetic sensor, a temperature sensor, a pressure sensor, or a water pressure sensor may be used. In this case, the sensor values output by the other sensors (for example, the angular velocity output by the gyro sensor, the position output by the position sensor, the voltage output by the voltage sensor, the current output by the current sensor, the speed output by the speed sensor, the angle output by the angle sensor, the magnetic field output by the magnetic sensor, the temperature output by the temperature sensor, the pressure output by the pressure sensor, the water pressure output by the water pressure sensor) may be treated in the same way as the acceleration data, a classification model 123 may be generated, and class classification may be performed from the input sensor values.

[0150] That is, the data obtained from the robot arm 200 may include at least one of the acceleration, displacement, angular velocity, position, voltage, current, speed, angle, magnetic field, temperature, pressure, and water pressure of the robot arm 200.

[0151] According to Modification 3, abnormal data that can actually occur can be generated for at least one of the acceleration, displacement, angular velocity, position, voltage, current, speed, angle, magnetic field, temperature, pressure, and water pressure of the robot arm 200.

[0152] [Appendix] Each process (each function) of the above-described embodiment is realized by a processing circuit including one or more processors. The processing circuit may be composed of an integrated circuit in which one or more memories, various analog circuits, and various digital circuits are combined in addition to the one or more processors. The one or more memories store a program (instruction) for causing the one or more processors to execute each process. The one or more processors may execute each process according to the program read from the one or more memories, or may execute each process according to a logic circuit designed in advance to execute each process. The processor may be various processors suitable for controlling a computer, such as a CPU, GPU, DSP, FPGA, ASIC, etc. Note that the plurality of physically separated processors may cooperate with each other to execute each process. For example, the processors mounted on each of the plurality of physically separated computers may cooperate with each other to execute each process via a network such as a LAN (Local Area Network), WAN (Wide Area Network), or the Internet. The program may be installed in the memory via the network from an external server device or the like, or may be distributed in a state stored in a recording medium such as a CD-ROM (Compact Disc Read Only Memory), DVD-ROM (Digital Versatile Disk Read Only Memory), or semiconductor memory, and may be installed in the memory from the recording medium.

[0153] Also, at least a part of the above embodiment and the above modification example may be arbitrarily combined.

[0154] The embodiments disclosed this time should be considered to be illustrative in all respects and not restrictive. The scope of the present invention is shown not by the above meaning, but by the scope of claims, and is intended to include all modifications within the meaning and scope equivalent to the scope of claims.

Description of Reference Numerals

[0155] 1 Production system 1A Production system 1B Production system 1C Production system 1D Production system 10 Data generation device 11 Data acquisition unit 12 Data generation unit 13 Model generation unit 20 Classifier 21 Data acquisition unit 22 Classification unit 23 Output unit 100 Computer 110 I / F unit 120 Storage device 121 Data generation program 122 Classification program 123 Classification model 130 Processor 140 Bus 200 Robot arm (target device) 200A Robot arm (target device) 200B Robot arm (target device) 200C Robot arm (target device) 201 Base 202 Joint 202A Joint 202B Joint 203A Link 203B Link 203C Link 204 End effector 205 Acceleration sensor 205A Acceleration sensor 206A Displacement sensor 206B Displacement sensor 210 Rest position 300A Belt conveyor 300B Belt conveyor 301A Driving pulley 301B Driven pulley 301C Driven pulley 301D Driven pulley 301E Driving pulley 301F Driven pulley 301G Driven pulley 301H Driven pulley 302A Motor 302B Motor 303A Belt 303B Belt 304A Current sensor 304B Current sensor 310 Object to be transported 400 Crane system 401 Transverse rail 402 Hoist crane 403 Wire 404 Tip member 404A Hook 405 Acceleration sensor 410 Object to be conveyed 501 Arrow 502 Arrow 503 Arrow 504 Arrow 601 Dashed line 701A Rectangular frame 701B Rectangular frame 702A Rectangular frame 702B Rectangular frame 705A Rectangular frame

Claims

1. A data acquisition unit that acquires first abnormal data, which is time-series abnormal data obtained from a target device; A data generation device comprising: a data generation unit that generates second abnormal data, which is time-series abnormal data belonging to a second abnormal class different from a first abnormal class to which the first abnormal data belongs when the first abnormal data is classified into classes based on the first abnormal data.

2. The data generation device according to claim 1, wherein the data generation unit generates the second abnormal data based on data obtained by temporally reversing a part of the first abnormal data.

3. The data acquisition unit further acquires third abnormal data, which is time-series abnormal data obtained from the target device and belonging to a third abnormal class different from the first abnormal class as a result of classification; The data generation device according to claim 1 or 2, wherein the data generation unit generates fourth abnormal data, which is time-series abnormal data belonging to a fourth abnormal class different from the first abnormal class and the third abnormal class, by combining parts of the first abnormal data and the third abnormal data.

4. The data generation device according to claim 1 or 2, wherein the data generation unit generates fifth abnormal data, which is time-series abnormal data belonging to a fifth abnormal class different from the first abnormal class and the second abnormal class, by combining parts of the first abnormal data and the second abnormal data.

5. The data acquisition unit further acquires normal data, which is normal data obtained from the target device and belonging to a normal class as a result of classification; The data generation device according to claim 1 or 2, wherein the data generation unit generates sixth abnormal data, which is time-series abnormal data belonging to a sixth abnormal class different from the first abnormal class, by combining parts of the first abnormal data and the normal data.

6. The target device includes a robotic arm; The data obtained from the target device includes at least one of acceleration, displacement, angular velocity, position, voltage, current, speed, angle, magnetic field, temperature, pressure, and water pressure of the robotic arm. The data generation device according to claim 1 or 2.

7. A model generation unit that generates a classification model for classifying time-series input data obtained from the target device into any of the abnormal classes and the normal class based on the time-series normal data belonging to the result of class classification in the normal class, the time-series abnormal data acquired by the data acquisition unit, and the time-series abnormal data generated by the data generation unit. The data generation device according to claim 1 or claim 2.

8. A data acquisition unit that acquires time-series input data from a target device, A class classification device comprising: a class classification unit that classifies the time-series input data into either a normal class or an abnormal class by inputting the time-series input data into a classification model generated by the data generation device according to claim 7.

9. A step of acquiring first abnormal data that is time-series abnormal data obtained from a target device, A step of generating second abnormal data that is time-series abnormal data belonging to a second abnormal class different from the first abnormal class to which the first abnormal data belongs when the first abnormal data is classified based on the first abnormal data. A data generation method including:

10. A step of acquiring time-series input data from a target device, A step of classifying the time-series input data into either a normal class or an abnormal class by inputting the time-series input data into a classification model generated by the data generation device according to claim 7. A class classification method including:

11. A computer, A data acquisition unit that acquires first abnormal data that is time-series abnormal data obtained from a target device, A computer program for causing the computer to function as a data generation unit that generates second abnormal data that is time-series abnormal data belonging to a second abnormal class different from the first abnormal class to which the first abnormal data belongs when the first abnormal data is classified based on the first abnormal data.

12. A computer, A data acquisition unit that acquires time-series input data from a target device, A computer program for causing the computer to function as a class classification unit that classifies the time-series input data into either a normal class or an abnormal class by inputting the time-series input data into a classification model generated by the data generation device according to claim 7.

Images