Information processing method, information processing device, and information processing program

Abstract

This technology provides the ability to accurately detect abnormalities in production equipment, even when multiple operational data sets exist. [Solution] An information processing method for detecting abnormalities in production equipment equipped with a load includes: acquiring a measurement signal from the load; acquiring operation data indicating the operation of the load based on the measurement signal; calculating multiple abnormality levels of the operation data for each of the multiple reference data based on the operation data and a plurality of reference data for each time point of the operation data; calculating the minimum abnormality level at each time point of the operation data based on the multiple abnormality levels; and displaying the minimum abnormality level on a display device provided by the production equipment.

Description

Technical Field

[0006] , , ,

[0005] , , ,

[0001] The present disclosure relates to a technique for detecting abnormalities in production equipment.

Background Art

[0002] Conventionally, in order to predict and maintain failures in production equipment, a technique for detecting abnormalities in production equipment (hereinafter referred to as an abnormality detection technique) is known. For example, Patent Document 1 discloses a technique for calculating a correlation value between a reference waveform and a current value when an abnormality is determined to have occurred by comparing the two. This technique assumes that the cause of an abnormality in production equipment is estimated by registering a reference waveform.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] In the technique described in Patent Document 1 above, when there are a plurality of reference data serving as a reference for calculating the degree of abnormality of operation data, no consideration has been given to a technique for accurately detecting abnormalities in production equipment. Therefore, the technique described in Patent Document 1 above has room for improvement.

[0005] The present disclosure has been made to solve such problems, and an object thereof is to provide a technique capable of accurately detecting abnormalities in production equipment even when there are a plurality of reference data.

Means for Solving the Problems

[0006] An information processing method in one aspect of the present disclosure is an information processing method for detecting an abnormality in production equipment equipped with a load, comprising: acquiring operation data indicating the operation of the load; pre-calculating multiple abnormality levels of the operation data for each of the multiple reference data based on the operation data and a plurality of reference data; identifying a minimum abnormality level which is the smallest of the plurality of abnormality levels; and displaying the minimum abnormality level on a display device provided in the production equipment. [Effects of the Invention]

[0007] This configuration allows for accurate detection of abnormalities in production equipment, even when multiple reference data sets exist. [Brief explanation of the drawing]

[0008] [Figure 1] This is a block diagram showing an example of a production facility configuration. [Figure 2] This is a block diagram showing an example of a motion controller configuration. [Figure 3] This flowchart shows an example of a process for collecting normal operation data. [Figure 4] This flowchart shows an example of a process for calculating the minimum anomaly score. [Figure 5] This figure shows the first normal operation graph G1, the operational operation graph G2, and the first abnormality graph G3. [Figure 6] This figure shows the second normal operation graph, the operational operation graph, and the second abnormality degree graph. [Figure 7] This is a diagram illustrating the process in step S13 of Figure 4. [Figure 8] This figure shows an example of the first display screen. [Figure 9] This flowchart shows an example of a process that presents similar data to the user. [Figure 10] This figure illustrates the process in step S8 of Figure 2. [Figure 11] This figure shows an example of the second display screen. [Modes for carrying out the invention]

[0009] (Knowledge forming the basis of this disclosure) In the field of anomaly detection technology, a distribution of normal operation data, indicating that production equipment is functioning correctly, is stored in memory beforehand. Operational operation data is then measured during the operation of the production equipment, and if this operational operation data differs from the normal operation data, the operational operation data is considered anomaly. The distribution of normal operation data can be efficiently stored using unsupervised machine learning techniques such as Mahalanobis distance or K-nearest neighbors. Anomalies are then detected by calculating the degree of anomaly (the degree to which the operational operation data differs from the normal operation data) relative to the normal operation data.

[0010] A problem with the above anomaly detection technology arises when multiple sets of normal operation data exist. For example, suppose a first set of normal operation data and a second set of normal operation data are stored in memory. Now, consider a scenario where the operational operation data is generally similar to the first set of normal operation data, but generally different from the second set of normal operation data. In this case, qualitatively, the degree of anomaly in the operational operation data will be calculated as low when using the first set of normal operation data as a reference, but as high when using the second set of normal operation data as a reference. Thus, when multiple sets of normal operation data exist, the degree of anomaly fluctuates depending on the normal operation data set as the reference, making accurate anomaly detection difficult.

[0011] The technology described in Patent Document 1 does not address any method for solving such problems. Therefore, it is difficult to solve the above problems using the technology described in Patent Document 1.

[0012] As a result of extensive research on anomaly detection technology, the inventors have found that by comparing multiple normal operation data and operational operation data, and identifying the minimum anomaly level among multiple anomaly levels, it is possible to appropriately detect anomalies in production equipment.

[0013] The present disclosure has been made based on such findings.

[0014] (1) An information processing method according to one aspect of the present disclosure is an information processing method for detecting an abnormality in production equipment having a load, the method including: obtaining operation data indicating an operation of the load; calculating a plurality of abnormality degrees of the operation data with respect to each of a plurality of reference data based on the operation data and the plurality of reference data; specifying a minimum abnormality degree that is the minimum among the plurality of abnormality degrees; and displaying the minimum abnormality degree on a display device included in the production equipment.

[0015] According to this configuration, a plurality of abnormality degrees of the operation data with respect to each of the plurality of reference data are calculated, and a minimum abnormality degree that is the minimum among the plurality of abnormality degrees is specified. Then, the minimum abnormality degree is presented to the user. In this way, even when there are a plurality of reference data, an abnormality in the production equipment can be accurately detected.

[0016] (2) In the information processing method according to (1) above, calculating the plurality of abnormality degrees may include calculating the plurality of abnormality degrees for each time point of the operation data, and specifying the minimum abnormality degree may include specifying the minimum abnormality degree at each time point of the operation data.

[0017] According to this configuration, the minimum abnormality degree at each time point of the operation data is calculated. For example, when the operation data at a certain time point is similar to any of the plurality of reference data, the minimum abnormality degree at that time point is calculated to be relatively small. On the other hand, when the operation data at a certain time point is not similar to any of the reference data, the minimum abnormality degree at that time point is calculated to be relatively large. And since the minimum abnormality degree at each time point is displayed on the display device, the user can confirm whether the operation data includes a time point at which the minimum abnormality degree is calculated to be large, that is, a time point that is highly likely to indicate an abnormality. In this way, even when there are a plurality of reference data, an abnormality in the production equipment can be accurately detected.

[0018] (3) The information processing method described in (1) or (2) above may further include obtaining designation information that specifies a first time point in the operation data and a second time point different from the first time point; extracting similar data from the plurality of reference data that is similar to the operation data in the interval from the first time point to the second time point; and displaying the similar data on the display device.

[0019] With this configuration, similar data is displayed on the display device, allowing the user to easily see what reference data was primarily used to calculate the minimum anomaly in the interval specified by the user.

[0020] (4) In the information processing method described in (3) above, extracting the similar data may include calculating the minimum anomaly at each time point constituting the interval from the first time point to the second time point, identifying the most cited data that has been adopted the most times as reference data indicating the minimum anomaly in the interval, and determining the most cited data as the similar data.

[0021] This configuration presents the user with the most frequently cited data, which is the reference data that has been used most often to represent the minimum anomaly within the user-specified interval. In other words, it can present the user with reference data that exhibits behavior similar to the behavior of the operational data within the user-specified interval.

[0022] (5) In the information processing method described in (3) above, extracting the similar data may include: calculating a first mean of the minimum anomaly at each time point constituting the interval from the first time point to the second time point; calculating a second mean of the anomaly in the corresponding interval for each of the multiple reference data; extracting the average similar data from the multiple reference data whose second mean is most similar to the first mean; and determining the average similar data as the similar data.

[0023] With this configuration, the user is presented with the mean-similar data in which the second mean is most similar to the first mean. In other words, the user can be presented with reference data that shows behavior similar to the behavior of the behavioral data within the interval specified by the user.

[0024] (6) In the information processing method described in any one of (1) to (5) above, acquiring the operation data may include generating it based on measurement signals measured by operating the production equipment.

[0025] With this configuration, operational data is generated based on measurement signals obtained by operating the production equipment, thus enabling the generation of accurate operational data.

[0026] (7) In the information processing method described in (6) above, the measurement signal may be measured by driving a drive device included in the load.

[0027] With this configuration, accurate operating data can be generated based on measurement signals obtained by driving the drive device.

[0028] (8) The information processing method described in any one of (1) to (7) above may further include determining whether the minimum abnormality is greater than a predetermined threshold, and outputting warning information if the minimum abnormality is greater than the predetermined threshold.

[0029] This configuration simplifies anomaly detection because warning information is output when the minimum anomaly level is greater than a predetermined threshold.

[0030] (9) An information processing device in another aspect of the present disclosure is an information processing device for detecting an abnormality in a production facility equipped with a load, the information processing device includes a processor, the processor performs the following: acquiring operation data indicating the operation of the load; calculating a plurality of abnormality degrees of the operation data for each of the plurality of reference data based on the operation data and a plurality of reference data; identifying a minimum abnormality degree which is the smallest of the plurality of abnormality degrees; and displaying the minimum abnormality degree on a display device provided in the production facility.

[0031] This configuration provides an information processing device that achieves the same effects as the information processing method described above.

[0032] (10) An information processing program in another aspect of the present disclosure is an information processing program that causes a computer to execute an information processing method for detecting an abnormality in production equipment equipped with a load, the program causing the computer to acquire operation data indicating the operation of the load; calculate a plurality of abnormality degrees of the operation data for each of the plurality of reference data based on the operation data and a plurality of reference data; identify a minimum abnormality degree which is the smallest of the plurality of abnormality degrees; and display the minimum abnormality degree on a display device provided in the production equipment.

[0033] This configuration makes it possible to provide an information processing program that achieves the same effects as the information processing method described above.

[0034] This disclosure can also be implemented as a system operated by such an information processing program. Furthermore, it goes without saying that such a computer program can be distributed via computer-readable, non-temporary recording media such as CD-ROMs or via communication networks such as the Internet.

[0035] The embodiments will be described in detail below with reference to the drawings. Note that the embodiments described below are all specific examples of this disclosure. The numerical values, shapes, components, steps, and order of steps shown in the following embodiments are examples only and are not intended to limit this disclosure. Furthermore, components in the following embodiments that are not described in the independent claim representing the highest-level concept will be described as optional components. Also, the contents of each embodiment can be combined.

[0036] (Embodiment 1) Figure 1 is a block diagram showing an example of the configuration of production equipment 1 according to Embodiment 1. Production equipment 1 includes a load 10, a servo amplifier 20, a motion controller 30, an input device 40, and a display 50. The load 10 to the display 50 are connected to each other via LAN cables or the like.

[0037] The load 10 consists of, for example, production equipment used to produce products. The production equipment is, for example, an industrial robot that performs equipment mounting, processing, machining, or transport. The load 10 is installed on a factory production line or the like. The load 10 includes a work arm 11 for gripping and processing parts, a servo motor 12 for driving the work arm 11, and a sensor 13 for detecting the state of the servo motor 12.

[0038] The servo motor 12 is a motor that operates by a specified number of rotations or angle. The servo motor 12 precisely operates the load 10 according to the drive signal output from the servo amplifier 20. For example, the servo motor 12 is incorporated into the joint of the work arm 11 to rotate the work arm 11 by a predetermined angle in the forward or reverse direction.

[0039] Since production equipment 1 is made up of many parts, if a part of production equipment 1 begins to break down, the operation of the servo motor 12 will also deviate from normal operation. Therefore, by measuring the operation of the servo motor 12, it is possible to detect signs of failure in production equipment 1.

[0040] Sensor 13 acquires a measurement signal from the servo motor 12. An example of sensor 13 is a torque sensor that detects the torque of the servo motor 12. Therefore, the measurement signal is the torque signal value. However, this is just an example, and the measurement signal may be an acceleration signal or a velocity signal. In addition, any information that can be used to detect abnormalities in the production equipment 1 may be measured. The measurement signal acquired by sensor 13 is input to the servo amplifier 20.

[0041] The servo amplifier 20 controls the servo motor 12 so that it operates according to the command signal output from the motion controller 30. The servo amplifier 20 generates a drive signal corresponding to the command signal output from the motion controller 30 and inputs it to the servo motor 12. Based on the measurement signal output from the servo motor 12, the servo amplifier 20 provides feedback control to the servo motor 12 so that it performs the operation corresponding to the command signal.

[0042] The motion controller 30 generates a command signal to operate the servo motor 12 and inputs the generated command signal to the servo amplifier 20. The command signal is time-series data that specifies the position of the servo motor 12 from the start to the end of a certain motion pattern. However, this is just an example, and the command signal may be a position command signal, a speed command signal, an acceleration command signal, or a torque command signal that specifies the position, speed, acceleration, or torque of the servo motor 12. Alternatively, the command signal may include at least two of the position command signal, speed command signal, acceleration command signal, and torque command signal.

[0043] The input device 40 is, for example, a touch panel and accepts instructions (operations) from the user.

[0044] The display 50 is composed of, for example, a liquid crystal display or an organic EL display. The display 50 is an example of a display device.

[0045] Figure 2 is a block diagram showing an example of the configuration of a motion controller 30. As shown in Figure 2, the motion controller 30 includes a processor 31 and a memory 32. The processor 31 is composed of, for example, a central processing unit (CPU). The processor 31 includes an acquisition unit 101, a calculation unit 102, and a display control unit 103. The acquisition unit 101 to the display control unit 103 are realized by the processor 31 executing an information processing program stored in the memory 32. However, the acquisition unit 101 to the display control unit 103 may be realized by a dedicated integrated circuit such as an ASIC.

[0046] The acquisition unit 101 acquires the measurement signal detected by the sensor 13. Based on the measurement signal, the acquisition unit 101 generates operational operation data indicating the operation of the load 10. The operational operation data is an example of operation data. The acquisition unit 101 inputs the operational operation data to the operational operation data storage unit 110.

[0047] The calculation unit 102 calculates multiple anomaly scores for each of the multiple normal operation data based on the operational operation data and multiple normal operation data, for each point in time of the operational operation data. Anomaly score is an index that indicates the degree of deviation of the operational operation data from the normal operation data. The calculation unit 102 in this embodiment uses multiple learning models to calculate the minimum anomaly score at each point in time of the operational operation data. The calculation unit 102 calculates the minimum anomaly score at each point in time of the operational operation data based on the multiple anomaly scores obtained using multiple learning models. Details will be described later.

[0048] The display control unit 103 displays the minimum abnormality level calculated by the calculation unit 102 on the display 50 (Figure 1).

[0049] The memory 32 is composed of storage devices such as RAM and flash memory. As shown in Figure 2, the memory 32 includes an operation data storage unit 110, a normal operation data storage unit 120, and a learning model storage unit 130.

[0050] The operation data storage unit 110 stores the operation data generated by the acquisition unit 101.

[0051] The normal operation data storage unit 120 stores (registers) various types of normal operation data. Normal operation data is used as a standard when calculating the degree of abnormality of operational data. Normal operation data indicates that the production equipment 1 is operating normally. In this embodiment, the normal operation data storage unit 120 stores hundreds of different types of normal operation data. However, this is just an example, and the number of normal operation data stored in the normal operation data storage unit 120 can be changed as appropriate depending on the storage capacity of the memory 32. Therefore, for example, the normal operation data storage unit 120 may store dozens of normal operation data, or it may store thousands of normal operation data. Normal operation data is an example of standard data.

[0052] The learning model memory unit 130 stores multiple learning models.

[0053] Figure 3 is a flowchart showing an example of a process for collecting normal operation data.

[0054] (Step S1) In step S1, normal operation is measured. Specifically, in step S1, the motion controller 30 generates various command signals to operate the servo motor 12. The servo amplifier 20 controls the operation of the servo motor 12 based on the above various command signals. As a result, the servo motor 12 operates in various operation patterns. As a result, various measurement signals are acquired by the acquisition unit 101. Based on these measurement signals, the acquisition unit 101 generates various operational operation data.

[0055] (Step S2) In step S2, normal operation is recorded in memory 32. Specifically, in step S2, among the various operational operation data generated by the acquisition unit 101 in step S1, operational operation data that the user or others determine to indicate that the production equipment 1 is operating normally is registered as normal operation data in the normal operation data storage unit 120.

[0056] As the processes of steps S1 and S2 are repeated, various normal operation data (for example, several hundred types of normal operation data) are accumulated in the normal operation data storage unit 120.

[0057] Figure 4 is a flowchart showing an example of the process for calculating the minimum anomaly score.

[0058] (Step S11) In step S11, operational data is input to the learning model. Specifically, in step S11, production equipment 1 is operated, and the servo motor 12 operates under the control of the servo amplifier 20. At this time, the sensor 13 acquires a measurement signal at a predetermined sampling period. The sensor 13 acquires, for example, a torque signal value as a measurement signal. The acquisition unit 101 acquires this measurement signal, generates operational data based on the measurement signal, and stores the operational data in the operational data storage unit 110. The calculation unit 102 acquires the operational data from the operational data storage unit 110 and inputs the operational data into multiple learning models.

[0059] (Step S12) In step S12, the degree of abnormality of the operational data is calculated for all normal operation data. The method for calculating the degree of abnormality will be explained below using the case where the first normal operation data and the second normal operation data are stored in the normal operation data storage unit 120, and the first learning model and the second learning model are stored in the learning model storage unit 130 as an example.

[0060] The calculation unit 102 inputs the operational data generated by the acquisition unit 101 into the first learning model. The first learning model is a model that calculates the degree of abnormality of the operational data based on the first normal operation data. The first learning model has in advance the distribution of multiple normal operation data. The distribution of multiple normal operation data is learned by using unsupervised learning, such as the k-nearest neighbors method.

[0061] Figure 5 shows the first normal operation graph G1, the operational operation graph G2, and the first abnormality graph G3. The first normal operation graph G1 is a graph showing the first normal operation data, which is one of several normal operation data stored in the normal operation data storage unit 120. In the first normal operation graph G1, the vertical axis v1 represents the torque signal value, and the horizontal axis represents time. The operational operation graph G2 is a graph showing the operational operation data generated by the acquisition unit 101. In the operational operation graph G2, the vertical axis v2 represents the torque signal value, and the horizontal axis represents time.

[0062] In this embodiment, when operational operation data is input from the calculation unit 102, the first learning model compares the operational operation data with the first normal operation data and calculates the degree of abnormality at each point in time of the operational operation data. Hereinafter, the degree of abnormality calculated by the first learning model will be referred to as the first degree of abnormality. For example, the first learning model calculates the first degree of abnormality for each point in time such as t1, t2, and t3 in the operational operation graph G2 in Figure 5. Each point in time corresponds, for example, to the sampling period of the measurement signal from the sensor 13. The first learning model treats, for example, the magnitude of the Mahalanobis distance from the first normal operation data to the operational operation data as the magnitude of the first degree of abnormality.

[0063] The first anomaly graph G3 in Figure 5 shows the first anomaly at each point in time of the operational data. In the first anomaly graph G3, the vertical axis v3 represents the magnitude of the first anomaly, and the horizontal axis represents time. As shown in Figure 5, the torque signal value of the operational graph G2 and the torque signal value of the first normal operation graph G1 at time t1 are close. Therefore, in the first anomaly graph G3, the first anomaly at time t1 is calculated to be small. On the other hand, the torque signal value of the operational graph G2 at time t2 is higher than the torque signal value of the first normal operation graph G1 at time t2. Therefore, in the first anomaly graph G3, the first anomaly at time t2 is calculated to be large. Also, the torque signal value of the operational graph G2 at time t3 is higher than the torque signal value of the first normal operation graph G1 at time t3. Therefore, in the first anomaly graph G3, the first anomaly at time t3 is calculated to be large. The first learning model performs a similar process at each point in time within the interval from time ts to time te of the operational data. Time ts corresponds to the start time of measurement of the measurement signal (torque in this example) used to generate the operational data, and time te corresponds to the end time of measurement of the measurement signal.

[0064] In parallel with the calculation of the first abnormality score by the first learning model, the calculation unit 102 inputs operational data to the second learning model and causes the second learning model to calculate the abnormality score of the operational data. Figure 6 shows the second normal operation graph G4, the operational operation graph G2, and the second abnormality score graph G5. The second normal operation graph G4 is a graph of the second normal operation data, which is one of the multiple normal operation data stored in the normal operation data storage unit 120. In the second normal operation graph G4, the vertical axis v4 represents torque, and the horizontal axis represents time. The operational operation graph G2 shown in Figure 6 is the same as the operational operation graph G2 shown in Figure 5, so no explanation is given.

[0065] In this embodiment, when operational operation data is input from the calculation unit 102, the second learning model compares the operational operation data with the second normal operation data and calculates the degree of anomaly at each point in time of the operational operation data. Hereinafter, the degree of anomaly calculated by the second learning model will be referred to as the second degree of anomaly. Similar to the first degree of anomaly, methods such as the Mahalanobis distance may be used to calculate the second degree of anomaly.

[0066] The second anomaly graph G5 in Figure 6 shows the second anomaly at each point in time of the operational data. In the second anomaly graph G5, the vertical axis v5 represents the magnitude of the second anomaly, and the horizontal axis represents time. As shown in Figure 6, the torque signal value of the operational graph G2 at time t1 is increased compared to the torque signal value of the second normal operation graph G4. Therefore, in the second anomaly graph G5, the second anomaly at time t1 is calculated to be large. Also, the torque signal value of the operational graph G2 at time t2 is increased compared to the torque signal value of the first normal operation graph G1 at time t2. Therefore, in the second anomaly graph G5, the second anomaly at time t2 is calculated to be large. On the other hand, the torque signal value of the operational graph G2 and the torque signal value of the second normal operation graph G4 at time t3 are close. Therefore, in the second anomaly graph G5, the second anomaly at time t3 is calculated to be small. The second learning model performs a similar process at each point in time included in the interval from time ts to time te of the operational data.

[0067] (Step S13) In step S13, the calculation unit 102 determines the minimum anomaly level at each point in time of the operational data based on the first and second anomaly levels. Figure 7 is a diagram illustrating the process in step S13. Figure 7 shows the first anomaly level graph G3, the second anomaly level graph G5, and the minimum anomaly level graph M1 overlaid on top of each other. In Figure 7, the vertical axis v6 represents the anomaly level, and the horizontal axis represents time. For the sake of explanation, the first anomaly level at time t1, the first anomaly level at time t2, and the first anomaly level at time t3 will be referred to as the first a anomaly level, the first b anomaly level, and the first c anomaly level, respectively. Also, the second anomaly level at time t1, the second anomaly level at time t2, and the second anomaly level at time t3 will be referred to as the second a anomaly level, the second b anomaly level, and the second c anomaly level, respectively.

[0068] As shown in Figure 7, the first anomaly (1a anomaly) at time t1 is smaller than the second anomaly (2a anomaly) at time t1. Therefore, in this case, the calculation unit 102 determines the 1a anomaly as the minimum anomaly at time t1. Also, the second anomaly (2b anomaly) at time t2 is smaller than the first anomaly (1b anomaly) at time t2. In this case, the calculation unit 102 determines the 2b anomaly as the minimum anomaly at time t2. Also, the second anomaly (2c anomaly) at time t3 is smaller than the first anomaly (1c anomaly) at time t3. In this case, the calculation unit 102 determines the 2c anomaly as the minimum anomaly at time t3. By repeating the above process, the calculation unit 102 determines the minimum anomaly at each time point.

[0069] (Step S14) The display control unit 103 presents the user with the minimum anomaly score. Figure 8 shows an example of the first display screen 51 displayed on the display 50. As shown in Figure 8, the first display screen 51 includes a minimum anomaly score graph M1. The minimum anomaly score graph M1 is a graph that shows the change in the minimum anomaly score of operational data over time. The vertical axis of the minimum anomaly score graph M1 represents the minimum anomaly score, and the horizontal axis represents the time axis. In the example shown in Figure 8, the display control unit 103 displays the minimum anomaly score graph M1 and the operational graph G2 superimposed on each other.

[0070] In steps S11 to S14, for the sake of simplicity, an example was described in which the calculation unit 102 calculates the minimum abnormality of the operational operation data at each point in time using the first normal operation data and the second normal operation data, but this is just one example. The calculation unit 102 may determine the minimum abnormality based on more normal operation data. For example, if the normal operation data storage unit 120 stores hundreds of types of normal operation data, the calculation unit 102 can use these hundreds of types of normal operation data to calculate the minimum abnormality of the operational operation data at each point in time.

[0071] Figure 9 is a flowchart illustrating an example of a process for presenting similar data to the user.

[0072] (Step S21) Production equipment 1 accepts a range specification operation from the user via input device 40 on a user interface (e.g., first display screen 51) where operational data and minimum abnormality levels are displayed. Specifically, production equipment 1 acquires specification information that specifies a first time point in the operational data and a second time point different from the first time point.

[0073] (Step S22) The display control unit 103 searches for the normal operation data that has been adopted the most times as normal operation data showing the minimum abnormality level within the range (interval) from the first time point to the second time point specified by the user (hereinafter referred to as the most frequently cited data). In other words, the display control unit 103 extracts similar data that is similar to the operation data within the range from the first time point to the second time point from among multiple normal operation data.

[0074] Figure 10 is a diagram illustrating the process in step S22. In Figure 10, the vertical axis represents the degree of abnormality and the torque signal value, and the horizontal axis represents time. In the example shown in Figure 10, the user inputs an operation to specify time t10 as the first time point and time t14 as the second time point. That is, in Figure 10, the range (interval) 100 from time t10 to time t14 is specified by the user. In this case, when the display control unit 103 calculates the minimum degree of abnormality at each time point within the range 100 from time t10 to time t14, it searches for the normal operation data that has been adopted the most times as the minimum degree of abnormality. For example, suppose that at time t10, t11, t12, and t13, the first normal operation data was adopted as the normal operation data showing the minimum degree of abnormality. Also, suppose that at time t14, the second normal operation data was adopted as the normal operation data showing the minimum degree of abnormality. In this case, the display control unit 103 determines the first normal operation data to be the most frequently cited data. The most frequently cited data is an example of similar data.

[0075] (Step S23) The display control unit 103 displays the most frequently cited data on the display 50. Figure 11 is a diagram showing an example of the second display screen 52 displayed on the display 50 in step S23. In Figure 11, the vertical axis represents the degree of abnormality and the torque signal value, and the horizontal axis represents time. As shown in Figure 11, the second display screen 52 includes the minimum abnormality graph M1 and the first normal operation graph G1 which shows the first normal operation data, which is the most frequently cited data. In the example shown in Figure 11, the minimum abnormality graph M1 and the first normal operation graph G1 are displayed separately, but the display control unit 103 may display them superimposed.

[0076] As described above, in this embodiment, multiple anomaly scores for operational operation data are calculated for each of the multiple normal operation data, and the minimum anomaly score, which is the smallest of the multiple anomaly scores, is identified. The minimum anomaly score is then presented to the user. In this way, even when multiple normal operation data exist, anomalies in the production equipment 1 can be accurately detected.

[0077] Furthermore, in this embodiment, the minimum abnormality level at each point in time of multiple operation data is calculated, so even if there are countless normal operation data, abnormalities in the production equipment can be accurately detected. Specifically, in this embodiment, operational operation data is generated using measurement signals such as torque or rotational speed that can be measured from the servo motor 12. Since the production equipment 1 in which the servo motor 12 is used includes drive parts such as the work arm 11, there are virtually infinitely many operation patterns for the servo motor 12. For example, during acceleration, constant speed, deceleration, etc., the measurement signals measured from the servo motor 12 change in various ways, so there are virtually infinitely many patterns of operational operation data.

[0078] In contrast, in this embodiment, multiple machine learning models execute in parallel the process of calculating the degree of abnormality based on a finite number of normal operation data points, and the minimum value among the degrees of abnormality calculated by each learning model is output to the display 50.

[0079] For example, as explained using Figures 5 to 7, the minimum anomaly at time t1 is calculated to be low based on the first anomaly, and the minimum anomaly at time t3 is calculated to be low based on the second anomaly. On the other hand, at time t2, both the first and second anomaly levels increase, so as a result, the minimum anomaly at time t2 is calculated to be high. The user can recognize that an anomaly has occurred in production equipment 1 by checking the second display screen 52 (Figure 8) which shows the minimum anomaly.

[0080] In this way, by calculating the minimum degree of anomaly at each point in time in the operational data, it is possible to accurately detect anomalies in production equipment 1, even when there are countless data points that can be generated as operational data.

[0081] Furthermore, in the production equipment 1 according to this embodiment, when a user specifies a range, the most frequently cited data is displayed, which was used most often when determining the minimum abnormality level within that range. This allows the user to easily confirm what kind of normal operation data was mainly used to calculate the minimum abnormality level.

[0082] Furthermore, the second display screen 52 according to this embodiment displays the operation graph G2 and the most frequently cited data (for example, the first normal operation graph G1). This allows the user to compare the operation data and the most frequently cited data. This allows the user to confirm whether the behavior of the minimum abnormality in range 100 indicates an abnormality in the production equipment 1 or an unknown normal operation. An unknown normal operation refers to an operation that is not registered in the normal operation data storage unit 120 and does not indicate a failure of the production equipment 1.

[0083] For example, suppose the user has prior knowledge that an increase in the torque of servo motor 12 compared to the most frequently cited data indicates a possible failure of production equipment 1, while a decrease in the torque of servo motor 12 compared to the most frequently cited data indicates no failure. In such a case, presenting the user with normal operation data, such as the most frequently cited data, which formed the basis for calculating the minimum anomaly score for a portion of the operational operation data, helps the user interpret the calculation results of the minimum anomaly score. In the example shown in Figure 11, the torque signal value in range 100 is decreased compared to the torque signal value in the most frequently cited data (first normal operation graph G1). Therefore, in this case, the user can interpret that the behavior of the torque signal value in the operational operation data in range 100 does not indicate a failure of production equipment 1. In other words, although the minimum anomaly score increases from time t10 to time t14, the user can interpret this increase in the minimum anomaly score not as an indication of a failure of production equipment 1, but as an indication of unknown normal operation.

[0084] The following variations of this disclosure may be adopted.

[0085] (1) In Embodiment 1, an example was described in which the display control unit 103 extracts the most frequently cited data in step S22, but this is just one example. In step S22, the display control unit 103 may calculate the average value of the minimum abnormality at each point in time within the range 100 specified by the user (hereinafter referred to as the first average value). Then, the display control unit 103 may extract from a plurality of normal operation data the normal operation data (hereinafter referred to as the average similar data) in which the average value of the abnormality at each point in the corresponding range 105 corresponding to range 100 (second average value) is most similar to the first average value. The average similar data is an example of similar data.

[0086] For example, suppose the first average value of the minimum anomaly from time t10 to time t14 in the range 100 shown in Figure 10 is 10. Here, suppose the second average value of the first anomaly from time t10 to time t14 in the corresponding range 105 of the first normal operation data is 11, and the second average value of the second anomaly from time t10 to time t14 in the corresponding range 105 of the second normal operation data is 15. In this case, the display control unit 103 determines the second normal operation data as the average similar data. Then, in step S23, the display control unit 103 displays the second normal operation graph G4 on the display 50.

[0087] If the most frequently cited data contains outliers, it is highly likely that the most frequently cited data is unsuitable as reference data (referred to as reference data) for determining whether the operational behavior data indicates an anomaly or an unknown normal behavior. In contrast, according to this modified version, the user can be presented with normal behavior data that shows behavior that is on average similar (overall similar) to the behavior of the operational behavior data within the range specified by the user. In this way, it is possible to suppress the display of normal behavior data containing outliers as reference data on the display 50 within the range corresponding to the range specified by the user.

[0088] (2) The memory 32 may further include a threshold storage unit. The threshold storage unit may store predetermined thresholds set in advance by the user or the like. The calculation unit 102 may determine whether any of the minimum abnormality levels at each point in time of the operational data is greater than a predetermined threshold. If there is a point in time when the minimum abnormality level is greater than the predetermined threshold, the calculation unit 102 may output warning information. The warning information may be a message such as "An abnormality may have occurred." The warning information may be output to the display 50, for example. In addition, the warning information may be output to an information processing terminal such as a smartphone owned by the user.

[0089] According to this modified version, warning information is output when the minimum anomaly level is greater than a predetermined threshold, thus simplifying anomaly detection.

[0090] (3) In Embodiment 1, an example was described in which the anomaly score is calculated using the Mahalanobis distance, but the method for calculating the anomaly score can be changed as appropriate. For example, the learning model may calculate the anomaly score using the difference. As an example, if the torque at time t1 of the first normal operation data is 10 Nm and the torque at time t1 of the operational operation data is 5 Nm, the first learning model may calculate the difference between the two, 5 Nm, as the first anomaly score at time t1.

[0091] (4) In Embodiment 1, an example was described in which the sensor 13 detects an abnormality in the production equipment 1 based on a measurement signal acquired from the servo motor 12. However, the object of measurement of the measurement signal is not limited to the servo motor 12. For example, if the production equipment 1 includes various drive devices such as a stepping motor and an induction motor, an abnormality in the production equipment 1 may be detected based on measurement signals that can be measured from these drive devices. In addition, an abnormality in the production equipment 1 may be detected based on measurement signals that can be measured from various components of the production equipment 1 other than drive devices.

[0092] (5) In Embodiment 1, an example was described in which the processor 31 and memory 32 are implemented in the motion controller 30, but the processor 31 and memory 32 may be implemented in, for example, the servo amplifier 20. In addition, the processor 31 and memory 32 may be implemented in, for example, a personal computer or server (for example, a cloud server) that is connected to the production equipment 1 in a way that enables data transmission and reception.

[0093] (6) In Embodiment 1, an example was described in which the degree of abnormality at each time point included in the interval from time point ts to time point te is calculated, but this is only one example. The calculation unit 102 may, for example, calculate the degree of abnormality in the interval from time point ts to time point t3. In other words, this disclosure is applicable not only when calculating the degree of abnormality at all time points from the start to the end of the operational data, but also when calculating the degree of abnormality at each time point included in a part of the operational data.

[0094] (7) In Embodiment 1, an example was described in which normal operation data is stored in the normal operation data storage unit 120 in steps S1 and S2. However, the normal operation data may be stored in the normal operation data storage unit 120 at an appropriate timing after step S11.

[0095] (8) In Embodiment 1, an example was described in which the acquisition unit 101 generates operational data indicating the operation of the load 10 based on the measurement signal. However, the load 10 may generate the operational data. In this case, the acquisition unit 101 only needs to acquire the operational data generated by the load 10. [Industrial applicability]

[0096] This disclosure is useful in the field of technology for detecting abnormalities in production equipment. [Explanation of symbols]

[0097] 1: Production equipment 10: Load 11: Work arm 12: Servo motor 13: Sensor 20: Servo amplifier 30: Motion controller (an example of an information processing device and computer) 31: Processor 32: Memory 50: Display (an example of a display device) 51: 1st display screen 52:Second display screen G1: First normal operation graph G2: Operation graph G3: First Anomaly Score Graph G4: Second normal operation graph G5: Second Anomaly Score Graph M1: Minimum Anomaly Graph

Claims

1. An information processing method for detecting abnormalities in production equipment equipped with a load, To acquire operational data indicating the operation of the aforementioned load, Based on the aforementioned operational data and a plurality of reference data, a plurality of anomaly degrees of the operational data are calculated for each of the plurality of reference data. Identifying the minimum anomaly, which is the smallest anomaly among the aforementioned multiple anomaly degrees, This includes displaying the minimum abnormality on a display device provided by the production equipment, Information processing methods.

2. Calculating the multiple anomaly degrees includes calculating the multiple anomaly degrees for each point in time of the operation data, Identifying the minimum anomaly includes identifying the minimum anomaly at each point in time of the operation data. The information processing method according to claim 1.

3. moreover, To obtain specification information that specifies a first time point in the aforementioned operation data and a second time point different from the first time point, Extracting similar data from the plurality of reference data that is similar to the operation data in the interval from the first time point to the second time point, This includes displaying the aforementioned similar data on the display device, The information processing method according to claim 2.

4. Extracting the aforementioned similar data is To calculate the minimum anomaly at each time point that constitutes the interval from the first time point to the second time point, To identify the most frequently cited data that was adopted the most times as the reference data indicating the minimum anomaly in the aforementioned interval, This includes determining the most frequently cited data as the similar data, The information processing method according to claim 3.

5. Extracting the aforementioned similar data is To calculate the first mean of the minimum anomaly at each time point that constitutes the interval from the first time point to the second time point, The second average value of the abnormality in the corresponding interval for the aforementioned interval is calculated for each of the multiple reference data, Extracting the average similar data from the plurality of reference data in which the second average value is most similar to the first average value, This includes determining the average similar data as the similar data, The information processing method according to claim 3.

6. Acquiring the aforementioned operational data includes generating it based on measurement signals measured by operating the production equipment. The information processing method according to claim 1 or 2.

7. The aforementioned measurement signal is measured by driving the drive device included in the load. The information processing method according to claim 6.

8. moreover, Determining whether the minimum anomaly is greater than a predetermined threshold, This includes outputting warning information when the minimum abnormality level is greater than the predetermined threshold, The information processing method according to claim 1 or 2.

9. An information processing device for detecting abnormalities in production equipment equipped with a load, The aforementioned information processing device includes a processor, The aforementioned processor, To acquire operational data indicating the operation of the aforementioned load, Based on the aforementioned operational data and a plurality of reference data, a plurality of anomaly degrees of the operational data are calculated for each of the plurality of reference data. Identifying the minimum anomaly, which is the smallest anomaly among the aforementioned multiple anomaly degrees, The minimum abnormality level is displayed on a display device provided in the production equipment. Information processing device.

10. An information processing program that causes a computer to execute an information processing method for detecting abnormalities in production equipment equipped with a load, To the aforementioned computer, To acquire operational data indicating the operation of the aforementioned load, Based on the aforementioned operational data and a plurality of reference data, a plurality of anomaly degrees of the operational data are calculated for each of the plurality of reference data. Identifying the minimum anomaly, which is the smallest anomaly among the aforementioned multiple anomaly degrees, The minimum abnormality level is displayed on a display device provided in the production equipment, and the following is performed: Information processing program.

Images