Abnormality detection device and computer-readable recording medium

Abstract

This abnormality detection device is provided with: a data acquisition unit that acquires data relating to the state of an industrial machine; a feature amount generation unit that generates a feature amount indicating a state feature of the industrial machine on the basis of the state-related data; a feature amount storage unit that stores a feature amount; a feature model storage unit that stores at least one feature model; an inference calculation unit that uses the feature model to calculate an evaluation value for the state of the industrial machine on the basis of the feature amount; an abnormality determination unit that determines, on the basis of the calculation result of the inference calculation unit, that a certain abnormality has occurred in the state of the industrial machine; a state diagnosis unit that diagnoses the state of the industrial machine when the abnormality determination unit determines that an abnormality has occurred in the industrial machine; and a feature model management unit that generates a feature model using the feature amount stored in the feature amount storage unit on the basis of the diagnosis result of the state diagnosis unit, and adds the feature model to the feature model storage unit.

Description

Technical Field

[0001] This disclosure relates to anomaly detection devices and computer-readable recording media. Background Technology

[0002] Failures in industrial machinery such as machine tools and robots are a cause of reduced operating rates on production sites. Even if the machinery's mechanism is functioning correctly, its operational accuracy can deteriorate due to wear and tear on components, leading to defective workpieces and abnormal movements. To prevent reduced operating rates, defective workpieces, and abnormal movements, preventative maintenance of industrial machinery is necessary. This preventative maintenance involves collecting information (control signals, current values, vibration values, etc.) from operating industrial machinery. Furthermore, a method for anomaly detection has been proposed that uses machine learning based on the collected information to model the feature quantities representing the state characteristics of industrial machinery (e.g., Patent Documents 1-3).

[0003] Existing technical documents

[0004] Patent documents

[0005] Patent Document 1: Japanese Patent Application Publication No. 2020-123191

[0006] Patent Document 2: Japanese Patent Application Publication No. 2021-110979

[0007] Patent Document 3: Japanese Patent Application Publication No. 2018-156415 Summary of the Invention

[0008] The problem that the invention aims to solve

[0009] In anomaly detection techniques, for example, the appearance of a feature quantity not included in the feature model (a change in the state of the machine) is detected as an anomaly. Therefore, even when an anomaly is detected, it's unclear whether a problem has actually occurred in the machine or if an unknown normal state is simply being detected as an anomaly. If anomalies continue to be detected even when the industrial machinery is operating normally, the feature model needs to be updated so that the same state is not detected as anomaly in the future.

[0010] Anomalies and changes in the condition of industrial machinery are diverse. These anomalies include bearing damage, increased vibration due to shaft eccentricity, and component detachment. Furthermore, some changes in condition are reversible. Therefore, it is difficult to detect anomalies using only a single feature model best suited for a specific condition.

[0011] Methods for solving problems

[0012] The anomaly detection device disclosed herein notifies of an unknown anomaly when state data containing feature quantities not included in the feature model is encountered, and updates the feature model based on the inspection results of industrial machinery as a result, thereby solving the aforementioned problem.

[0013] Furthermore, one aspect of this disclosure is an anomaly detection device comprising: a data acquisition unit that acquires data related to the state of industrial machinery; a feature quantity generation unit that generates feature quantities representing the state characteristics of the industrial machinery based on the state-related data; a feature quantity storage unit that stores the feature quantities; a feature model storage unit that stores at least one feature model; an inference calculation unit that uses the feature model to calculate an evaluation value of the state of the industrial machinery based on the feature quantities; an anomaly determination unit that determines that the state of the industrial machinery has experienced some kind of anomaly based on the calculation result of the inference calculation unit; a state diagnosis unit that diagnoses the state of the industrial machinery when the anomaly determination unit determines that the industrial machinery has experienced an anomaly; and a feature model management unit that generates a feature model using the feature quantities stored in the feature quantity storage unit based on the diagnosis result of the state diagnosis unit, and adds it to the feature model storage unit. Attached Figure Description

[0014] Figure 1 This is a hardware structure diagram of the anomaly detection device according to the first embodiment.

[0015] Figure 2 This is a block diagram illustrating the function of the anomaly detection device in the first embodiment.

[0016] Figure 3 This is a schematic diagram illustrating an example of the inference computation unit using multiple feature models to perform feature-based inference.

[0017] Figure 4 This is a diagram illustrating an example of inference processing using a feature model in the inference computation unit.

[0018] Figure 5 This is a diagram illustrating another example of inference processing using a feature model in the inference computation unit.

[0019] Figure 6 This is a diagram illustrating another example of inference processing using a feature model in the inference computation unit.

[0020] Figure 7 This is a diagram illustrating another example of inference processing using a feature model in the inference computation unit.

[0021] Figure 8 This is a screenshot showing the judgment result of the anomaly judgment unit.

[0022] Figure 9This is a screenshot illustrating the threshold setting screen.

[0023] Figure 10 This is a schematic diagram illustrating an example of a feature model storage unit for each environment's storage feature model.

[0024] Figure 11 This is a block diagram illustrating the function of the anomaly detection device in the second embodiment. Detailed Implementation

[0025] The following is related to the appendix. Figure 1 The embodiments of this disclosure will be described below. Furthermore, in the following description, structures having the same or similar functions will be labeled with the same reference numerals. Also, repeated descriptions of these structures will sometimes be omitted.

[0026] In this application, "based on XX" means "at least based on XX," but also includes cases based on other elements besides XX. Furthermore, "based on XX" is not limited to directly using XX, but also includes cases based on operations or processing performed on XX. "XX" can be any element (e.g., any information).

[0027] [First Implementation Method]

[0028] Figure 1 This is a schematic hardware structure diagram showing the main parts of the anomaly detection device according to the first embodiment of this disclosure. The anomaly detection device 1 of this disclosure can be installed, for example, on a control device for controlling industrial machinery. Furthermore, the anomaly detection device 1 of this disclosure can be installed on a personal computer installed alongside the control device for controlling industrial machinery, a personal computer connected to the control device via a wired / wireless network, a unit computer, a fog computer 6, a cloud server 7, or other computers. In this embodiment, an example of installing the anomaly detection device 1 on a control device for controlling industrial machinery is shown.

[0029] The CPU 11 of the anomaly detection device 1 in this embodiment is a processor that controls the anomaly detection device 1 as a whole. The CPU 11 reads the system / program stored in the ROM 12 via the bus 22 and controls the anomaly detection device 1 as a whole according to the system / program. Temporary calculation data, display data, and various data input from external sources are temporarily stored in the RAM 13.

[0030] The non-volatile memory 14 is composed of, for example, a memory backed up by a battery (not shown), an SSD (Solid State Drive), etc., and maintains its storage state even when the power supply to the fault detection device 1 is disconnected. Programs and data read from the external device 72 via the interface 15, programs and data input via the input device 71, and programs and data obtained from the industrial machinery 2 are stored in the non-volatile memory 14. The data stored in the non-volatile memory 14 can also be expanded in the RAM 13 during execution / use. Furthermore, various systems / programs, such as known parsing programs, are pre-written into the ROM 12.

[0031] Interface 15 is used to connect the CPU 11 of the anomaly detection device 1 to an external device 72 such as a USB memory, Compact Flash (registered trademark), or SD card. It can read, for example, control programs and various data used for controlling the industrial machine 2 from the external device 72. Furthermore, control programs and various data edited within the anomaly detection device 1 can be stored on the external device 72. The PLC (Programmable Logic Controller) 16 outputs signals and performs control via I / O unit 17 to the industrial machine 2 and its peripheral devices (e.g., tool changers, actuators of robots, sensors installed on the industrial machine 2) through a timing program built into the anomaly detection device 1. Additionally, the PLC 16 receives signals from various switches and peripheral devices on the control panel of the main body of the industrial machine 2, performs necessary signal processing, and then transmits them to the CPU 11.

[0032] In the display device 70, data read into the memory and data obtained as a result of executing programs are output and displayed via the interface 18. In addition, the input device 71, which consists of a keyboard, indicator devices, etc., transmits instructions and data based on the operator's operation to the CPU 11 via the interface 19.

[0033] Interface 20 is used to connect the CPU 11 of the anomaly detection device 1 to a wired or wireless network 5. The network 5 can communicate using technologies such as RS-485 serial communication, Ethernet communication, optical communication, wireless LAN, Wi-Fi, and Bluetooth. Computers such as the fog computer 6 and cloud server 7 are connected to the network 5 and exchange data with the anomaly detection device 1.

[0034] The axis control circuit 30, used to control the control axes of the industrial machine 2, receives position commands for the control axes from the CPU 11 and outputs the commands for those control axes to the servo amplifier 40. The servo amplifier 40 receives these commands to drive the servo motors 50 of the control axes, causing each component of the industrial machine 2 to move along its respective control axis. Each servo motor 50 has a built-in position detector, and the position feedback signal from the position detector is fed back to the axis control circuit 30. The axis control circuit 30 performs feedback control of the servo motors 50 based on the position feedback signal. Furthermore, in... Figure 1 In the hardware structure diagram, one axis control circuit 30, one servo amplifier 40, and one servo motor 50 are shown, but in reality, the number of control axes of the industrial machine 2 that is to be controlled is much larger. For example, in the case of controlling a typical machine tool with three linear axes, three sets of axis control circuits 30, servo amplifiers 40, and servo motors 50 are prepared to move the spindle on which the tool is mounted and the workpiece relative to each other in the directions of the three linear axes (X-axis, Y-axis, and Z-axis).

[0035] The spindle control circuit 60 receives a spindle rotation command and outputs a spindle speed signal to the spindle amplifier 61. The spindle amplifier 61 receives this spindle speed signal and causes the spindle motor 62 of the industrial machine 2 to rotate at the commanded speed, driving the spindle. The spindle motor 62 is coupled with a position encoder 63. The position encoder 63 outputs feedback pulses synchronously with the spindle rotation, and these feedback pulses are read by the CPU 11. Furthermore, in... Figure 1 In the hardware structure diagram, one spindle control circuit 60, one spindle amplifier 61, and one spindle motor 62 are shown, but in reality, the number of control axes of the industrial machine 2 that is the object of control is prepared. In addition, industrial machines 2 that do not have a spindle sometimes do not have these structures.

[0036] Interface 21 is used to connect CPU 11 and machine learning device 100. Machine learning device 100 includes: processor 101, which uniformly controls the entire machine learning device 100; ROM 102, which stores the system / program, etc.; RAM 103, which provides temporary storage for various processes involved in machine learning; and non-volatile memory 104, which stores models, etc. Machine learning device 100 can observe information obtainable by anomaly detection device 1 via interface 21. Furthermore, anomaly detection device 1 obtains the processing results output from machine learning device 100 via interface 21, stores or displays the obtained results, or sends them to other devices via network 5, etc.

[0037] Figure 2 This diagram is a schematic block diagram showing the functions of the anomaly detection device 1 according to the first embodiment of this disclosure. Figure 1The CPU 11 of the anomaly detection device 1 and the processor 101 of the machine learning device 100 executed the system / program to control the operation of each part of the anomaly detection device 1 and the machine learning device 100, thereby realizing the various functions of the anomaly detection device 1 of this embodiment.

[0038] The anomaly detection device 1 of this embodiment includes a data acquisition unit 110, a feature quantity generation unit 120, an inference calculation unit 140, an anomaly determination unit 150, a status diagnosis unit 160, a feature model management unit 170, and an output unit 190. Furthermore, the RAM 103 or non-volatile memory 104 of the machine learning device 100 includes a feature model storage unit 210 for storing at least one feature model and a feature quantity storage unit 220 for storing feature quantities.

[0039] The data acquisition unit 110 acquires data related to a predetermined state obtained from the industrial machine 2. Examples of data related to the state of the industrial machine 2 include feed rate, feed axis load, spindle speed, spindle load, position feedback value, and vibration value. The data acquisition unit 110 may also acquire environmental information indicating the predetermined state, such as signals (cutting signals, etc.), program number, and tool number. The data acquisition unit 110 outputs the acquired data to the feature quantity generation unit 120.

[0040] The feature quantity generation unit 120 generates feature quantities representing the characteristics of the state of the industrial machinery 2 based on data related to the state of the industrial machinery 2 acquired by the data acquisition unit 110. The feature quantities generated by the feature quantity generation unit 120 are useful information for judging abnormalities in the state of the industrial machinery 2. The feature quantities generated by the feature quantity generation unit 120 can be obtained by performing predetermined statistical calculations (average, variance, etc.), frequency characteristic calculations, or standardization based on predetermined benchmarks on the acquired data related to the state. The feature quantities generated by the feature quantity generation unit 120 include multiple values ​​generated based on data related to at least one state. Examples of feature quantities that can be considered as feature quantities include, for example, the feature quantity obtained by sampling the load of the spindle acquired by the data acquisition unit 110 at a predetermined sampling period over a predetermined past period. Additionally, the peak value of the vibration value of the servo motor 50 installed on the industrial machinery 2 over a predetermined past period is also an example of a feature quantity. The feature quantity generation unit 120 outputs the generated feature quantities to the inference calculation unit 140. In addition, the feature quantity generation unit 120 stores the generated feature quantity in the feature quantity storage unit 220 in association with the time when the data related to the state of the generation source is obtained.

[0041] The inference calculation unit 140 calculates an evaluation value for the state of the industrial machinery 2 based on the feature quantities input from the feature quantity generation unit 120, using at least one feature model stored in the feature model storage unit 210. Figure 3 This is a schematic diagram illustrating an example of the inference calculation unit 140 using multiple feature models to perform feature-based inference. For example... Figure 3 As illustrated, the inference calculation unit 140 is configured to use a machine learning machine that employs each of the plurality of feature models 1-1 to 1-n stored in the feature model storage unit 210, and inputs the feature quantities 1 to m generated by the feature quantity generation unit 120 as input data to each of these machine learning machines. Then, it calculates the inference results 1 to n as its output. Each inference result 1 to m becomes an evaluation value 1 to m obtained by inferring the state of the industrial machinery 2. The inference calculation unit 140 outputs at least one evaluation value as an inference result to the anomaly determination unit 150.

[0042] At least one feature model capable of inferring the state of industrial machinery 2 is pre-stored in the feature model storage unit 210. This feature model can also be generated based on data related to the state obtained from industrial machinery 2. Alternatively, for example, feature models used in other industrial machinery of the same model as industrial machinery 2 can be used. Each feature model can be, for example, a feature model generated by machine learning of feature quantities generated based on data related to the state obtained during normal operation of industrial machinery 2 (hereinafter referred to as a normality feature model). The normality feature model is used to infer whether, when at least one feature quantity is input, it falls within the range of feature quantities obtained from industrial machinery 2 operating normally. By using the normality feature model, an anomaly can be determined when the feature quantity deviates from the range of feature quantities under normal conditions. Each feature model can be, for example, a feature model generated by machine learning of feature quantities generated based on data related to the state obtained during abnormal operation of industrial machinery 2 (hereinafter referred to as an anomaly feature model). The anomaly feature model is used to infer whether, when at least one feature quantity is input, it falls within the range of feature quantities obtained from industrial machinery 2 operating in a predetermined abnormal state. By using the anomaly feature model, an anomaly can be determined when the feature quantity falls within the range of feature quantities under abnormal conditions. Each feature model can be generated through machine learning based on data associated with different states. Furthermore, each feature model can be a machine learning model using different methods. For example, it can be a device for inference processing using multi-layer neural networks, or a known machine learning device such as a VAE (Variational Auto Encoder), Bayesian network, support vector machine, Gaussian mixture model, or cluster model. Feature models can also be models for inference based on methods such as supervised learning, unsupervised learning, and reinforcement learning.

[0043] The anomaly determination unit 150 determines that the state of the industrial machine 2 has become abnormal based on at least one evaluation value input from the inference calculation unit 140. For example, the anomaly determination unit 150 compares the evaluation values ​​inferred from the normality feature model with predetermined thresholds. Next, if the comparison result with the thresholds meets predetermined conditions, the evaluation value determines that the state of the industrial machine 2 is abnormal. Then, if all evaluation values ​​indicate that the industrial machine 2 is in an abnormal state, the anomaly determination is made. On the other hand, if a certain feature model indicates a normal state, the industrial machine 2 is determined not to be in an abnormal state.

[0044] Figure 4 This is a diagram illustrating an example of inference processing using a feature model by the inference calculation unit 140. Figure 4 In the example, normality feature models 1 to 4 are stored in the feature model storage unit 210. Then, inferences are made using normality feature models 1 to 4 based on feature quantities generated from data related to the state obtained from the industrial machine 2. Figure 4 In the example, an evaluation value of 4.5 was calculated as the inference result using Normality Feature Model 1. Similarly, an evaluation value of 7.5 was calculated as the inference result using Normality Feature Model 2, an evaluation value of 6.4 was calculated as the inference result using Normality Feature Model 3, and an evaluation value of 3.3 was calculated as the inference result using Normality Feature Model 4. Here, the evaluation value calculated using Normality Feature Model 1 is set to be greater than a threshold of 6 to indicate that the state of Industrial Machinery 2 is abnormal. Similarly, when the evaluation value calculated using Normality Feature Model 2 is greater than a threshold of 7, when the evaluation value calculated using Normality Feature Model 3 is greater than a threshold of 5, and when the evaluation value calculated using Normality Feature Model 4 is greater than a threshold of 6, the state of Industrial Machinery 2 is respectively determined to be abnormal. Figure 4 In the example, the evaluation values ​​calculated using normality feature models 2 and 3 exceed the threshold, therefore the state of industrial machinery 2 is determined to be abnormal. However, the evaluation values ​​calculated using normality feature models 1 and 4 are below the threshold, therefore the state of industrial machinery 2 is determined to be normal. In such cases, the anomaly determination unit 150 determines that industrial machinery 2 has not experienced any anomalies.

[0045] Figure 5 This is a diagram illustrating another example of inference processing using the feature model in the inference calculation unit 140. Figure 5In the example, normality feature models 1 to 4 are stored in the feature model storage unit 210. Then, inferences are made using normality feature models 1 to 4 based on feature quantities generated from data related to the state obtained from the industrial machine 2. Figure 5 In the example, an evaluation value of 7.0 is calculated based on the inference result of using normality feature model 1. Similarly, an evaluation value of 8.2 is calculated based on the inference result of using normality feature model 2, an evaluation value of 6.8 is calculated based on the inference result of using normality feature model 3, and an evaluation value of 6.2 is calculated based on the inference result of using normality feature model 4. In this case, the evaluation value calculated using any of normality feature models 1 to 4 exceeds the threshold. Therefore, the anomaly determination unit 150 determines that an anomaly has occurred in the industrial machine 2.

[0046] The anomaly determination unit 150 can, when determining that certain anomalies have occurred in the industrial machine 2 using evaluation values ​​inferred from the normality feature model, further determine whether the anomaly is a known anomaly using evaluation values ​​inferred from the anomaly feature model. In this case, the anomaly determination unit 150 compares each evaluation value inferred from the anomaly feature model with a predetermined threshold. Then, if the comparison result with the threshold satisfies a predetermined condition, it is determined that a known anomaly exists in the state of the industrial machine 2. Furthermore, if, among all the evaluation values ​​inferred from the anomaly feature model, the comparison result with the threshold does not satisfy a predetermined condition, it is determined that an unknown anomaly has occurred in the industrial machine 2. Moreover, if the anomaly feature model related to the state of the industrial machine 2 is not stored in the feature model storage unit 210, it can determine that an unknown anomaly has occurred in the industrial machine 2 even if certain anomalies have occurred using evaluation values ​​inferred from the normality feature model.

[0047] Figure 6 This is a diagram illustrating another example of the inference processing performed by the inference calculation unit 140 using the feature model. Figure 6 In the example, the feature model storage unit 210 stores normality feature models 1-4 and abnormality feature models 1 and 2. Furthermore, the anomaly determination unit 150, based on feature quantities generated from data related to the state obtained from the industrial machine 2, and based on the results of inferences using normality feature models 1-4, determines that some kind of anomaly has occurred in the industrial machine 2. Next, the anomaly determination unit 150 refers to the results of inferences based on abnormality feature models 1 and 2. Figure 6 In the example, an evaluation value of 8.2 was calculated as the inference result using Anomaly Feature Model 1. Conversely, an evaluation value of 9.6 was calculated as the inference result using Anomaly Feature Model 2. Figure 6In the example, the evaluation values ​​of both anomaly feature models 1 and 2 did not meet the set threshold conditions. Therefore, the anomaly determination unit 150 determined that an unknown anomaly had occurred in the industrial machinery 2.

[0048] Figure 7 This is a diagram illustrating another example of the inference processing performed by the inference calculation unit 140 using the feature model. Figure 7 In the example, with Figure 6 Similarly, based on the results of inferences drawn using normality feature models 1-4, the anomaly determination unit 150 determines that some kind of anomaly has occurred in the industrial machine 2. On the other hand, the evaluation value using anomaly feature model 1 meets the condition of being smaller than the threshold 7, while the evaluation value using anomaly feature model 2 does not meet the condition of the threshold. Therefore, the anomaly determination unit 150 determines that a known anomaly involving anomaly feature model 1 has occurred in the industrial machine 2.

[0049] The anomaly determination unit 150 outputs the determination result related to the state of the industrial machine 2 to the output unit 190. Figure 8 This is a screenshot illustrating the judgment results of the anomaly judgment unit. Figure 8 In the example, the arrangement shows the shift of evaluation values ​​based on multiple feature models. Furthermore, if the anomaly determination unit 150 determines that an unknown anomaly has occurred in the industrial machine 2, it notifies the status diagnosis unit 160 that the industrial machine 2 is in a state abnormality.

[0050] When the status diagnosis unit 160 receives a notification from the anomaly determination unit 150 that an anomaly has occurred in the status of the industrial machine 2, it diagnoses the status of the industrial machine 2. The status diagnosis unit 160 can diagnose the status of the industrial machine 2 immediately upon receiving a notification from the anomaly determination unit 150, but it is preferable to analyze the status of the industrial machine 2 only when the same notification has continued for a predetermined period or when the same notification has been given a predetermined number of times. This allows a certain amount of feature quantities generated under the same status to be stored in the feature quantity storage unit 220. For example, the status diagnosis unit 160 can cause the industrial machine 2 to perform a predetermined diagnostic action, and automatically diagnose the status of the industrial machine 2 based on the result. In this case, a control program related to the diagnostic action is prepared in advance. Then, the operator is prompted to execute the control program related to the diagnostic action. Then, based on the result, any unknown anomalies occurring in the industrial machine 2 are automatically diagnosed. Alternatively, the status diagnosis unit 160 can simply prompt the operator to inspect the industrial machine 2, and diagnose the status of the industrial machine 2 by inputting information related to the results of that inspection.

[0051] In the diagnostic results of the industrial machine 2's status, there may be cases where no abnormality has occurred and cases where an abnormality has occurred. Furthermore, in the case where an abnormality has occurred, information related to the content of the abnormality may be included. If the status diagnosis unit 160 diagnoses that the industrial machine 2 has no abnormality, this means that although the anomaly determination unit 150 determines that the industrial machine 2 has experienced an unknown anomaly, the industrial machine 2 is not in an abnormal state. In such a case, the status diagnosis unit 160 instructs the feature model management unit 170 to generate a new normality feature model. On the other hand, if the status diagnosis unit 160 diagnoses that the industrial machine 2 has experienced an abnormality, it means that a new anomaly has been identified. In such a case, the status diagnosis unit 160 may also instruct the feature model management unit 170 to generate a new anomaly feature model related to the occurred anomaly. The status diagnosis unit 160 outputs the diagnostic results of the industrial machine 2's status to the output unit 190.

[0052] The feature model management unit 170 generates feature models based on instructions from the status diagnosis unit 160. When instructed by the status diagnosis unit 160 to generate a new normal feature model, the feature model management unit 170 reads multiple feature quantities calculated during the period when the anomaly was detected from the feature quantity storage unit 220. Then, it generates a normal feature model by performing machine learning on the read feature quantities. Conversely, when instructed by the status diagnosis unit 160 to generate a new abnormal feature model, the feature model management unit 170 reads multiple feature quantities calculated during the period when the anomaly was detected from the feature quantity storage unit 220. Then, it generates an abnormal feature model by performing machine learning on the read feature quantities. At this time, the feature model management unit 170 can also allow the operator to specify which range of feature quantities to use to generate the model. The threshold used to determine the evaluation value based on each feature model can be appropriately determined based on the type of feature model generated and the distribution of the feature quantities used for learning. For example, for the feature quantities used in generating the feature model, a value equivalent to a predetermined percentile (e.g., the 75th percentile) can be calculated, and the value obtained by multiplying it by a predetermined multiple (e.g., 1.5) can be used as the threshold. In this case, multiple percentile values ​​can be preset to calculate strict and lenient thresholds. The calculated threshold value can also be prompted to the operator for confirmation and correction. Additionally, as... Figure 9 As illustrated, the shift in anomaly calculated based on the generated feature model can also be displayed on the display device 70, prompting the operator to set a threshold. The feature model management unit 170 stores the generated feature model in the feature model storage unit 210. The feature model stored in the feature model storage unit 210 is used for subsequent determination of the state of the industrial machinery 2.

[0053] The output unit 190 outputs the judgment result of the anomaly judgment unit 150 and the diagnosis result of the status diagnosis unit. The output unit 190 can also display the judgment result and diagnosis result on the display device 70. In addition, the output unit 190 can also send the output to a host computer such as the fog computer 6 or the cloud server 7 via the network 5. In addition, the output can also be recorded to the log recording area set on the RAM 13 or the non-volatile memory 14 of the anomaly detection device 1.

[0054] In this embodiment, the anomaly detection device 1, equipped with the above-described structure, detects an unknown anomaly when the state of the industrial machinery 2 changes from the point in time when the feature model stored in the feature model storage unit 210 is generated. If the inspection result is determined to be without problems, a new normality feature model is generated using the feature quantity generated at that time. Therefore, in subsequent anomaly detections, the same state can be excluded from the anomaly detection targets. The inspection of the industrial machinery 2 can be performed at the moment when the unknown anomaly occurs (the moment the state of the machinery changes), thus improving the accuracy of anomaly detection with a minimal number of checks.

[0055] Furthermore, feature models can be generated for each state of industrial machinery 2. Therefore, by making each feature model meaningful, improved anomaly interpretability can be expected. In the case of detecting an unknown anomaly, it can be interpreted as a state where no similar machinery existed in the past. Therefore, by generating new feature models using feature quantities generated within the scope of the detected anomaly, highly interpretable feature models with a narrowed range of judgments can be generated. In the case of detecting a known anomaly, it can be interpreted as a state where machinery similar to the past exists. Even in the case where no anomaly is detected, it is possible to determine which state of machinery previously judged to be without problems is similar.

[0056] In the application of the anomaly detection device 1 of this embodiment with the above-described structure, a normality feature model is stored in the feature model storage unit 210 in the initial stage. This normality feature model can be a model generated based on feature quantities generated from data related to a relatively small number of states that can be obtained under normal operation of the industrial machine 2. Then, while the industrial machine 2 is running, a diagnosis is performed when certain unknown anomalies occur, and new normality feature models and anomaly feature models are added based on the results. By adding feature models in this way, the accuracy of determining the normal and abnormal states of the industrial machine 2 is increased. Therefore, the implementation cost can be suppressed, and a set of feature models that conform to the characteristics of the industrial machine 2 can be generated each time the industrial machine 2 is continuously used.

[0057] As a variation of the anomaly detection device 1 in this embodiment, multiple feature model groups can also be prepared for each environmental information of the industrial machinery 2. Figure 10 This is a schematic diagram illustrating an example of feature model groups stored in the feature model storage unit 210 of this modified example. In the feature model storage unit 210 of this modified example, at least one feature model group is stored for each group of environmental information of the industrial machine 2. Each feature model group includes at least one feature model. When the inference calculation unit 140 performs inference processing based on feature quantities generated from data related to the state obtained from the industrial machine 2, it selects the feature model group to be used based on the environmental information of the industrial machine 2. Then, inference is performed using at least one feature model contained in the selected feature model group. Furthermore, when instructed to generate a feature model, the feature model management unit 170 adds the generated feature model to the feature model group corresponding to the environmental information. By adopting this structure, feature models suitable for the environmental information can be used separately, and an improvement in the accuracy of anomaly detection can be expected.

[0058] As another variation of the anomaly detection device 1 of this embodiment, the feature model management unit 170 can also update or delete feature models stored in the feature model storage unit 210. The feature model management unit 170 performs updates or deletions of feature models stored in the feature model storage unit 210, for example, based on instructions from the operator. If an inappropriate model exists among the feature models stored in the feature model storage unit 210, the operator instructs the update or deletion of that feature model. This can be done, for example, when a diagnostic tendency different from other industrial machinery is observed when using a feature model used in other industrial machinery to diagnose the state of industrial machinery 2, or when an anomaly of industrial machinery 2 is discovered after verifying a normal feature model generated by the feature model management unit 170. By providing such a function, appropriate feature models for diagnosing industrial machinery 2 can be managed.

[0059] [Second Implementation]

[0060] The anomaly detection device of the second embodiment will be described below. The anomaly detection device 1 of this embodiment has the same hardware structure as the anomaly detection device 1 of the first embodiment.

[0061] Figure 11 This is a schematic block diagram illustrating the functions of the anomaly detection device 1 according to the second embodiment of this disclosure. Figure 1 The CPU 11 of the anomaly detection device 1 and the processor 101 of the machine learning device 100 executed the system / program to control the operation of each part of the anomaly detection device 1 and the machine learning device 100, thereby realizing the various functions of the anomaly detection device 1 of this embodiment.

[0062] In addition to the data acquisition unit 110, feature quantity generation unit 120, inference calculation unit 140, anomaly determination unit 150, state diagnosis unit 160, feature model management unit 170, and output unit 190, the anomaly detection device 1 of this embodiment also includes a feature quantity management unit 180. Furthermore, in the RAM 103 or non-volatile memory 104 of the machine learning device 100, in addition to the area storing at least one feature model (i.e., feature model storage unit 210) and the area storing feature quantities (i.e., feature quantity storage unit 220), an anomaly feature quantity storage unit 230 is also provided for storing feature quantities generated under abnormal conditions of the industrial machinery 2.

[0063] The data acquisition unit 110, feature quantity generation unit 120, inference calculation unit 140, anomaly determination unit 150, feature model management unit 170, and output unit 190 of this embodiment have the same functions as those in the first embodiment.

[0064] In this embodiment, when the status diagnosis unit 160 diagnoses an unknown anomaly occurring in the industrial machine 2, the feature quantity management unit 180 stores the feature quantities generated by the feature quantity generation unit 120 during the period when the anomaly was diagnosed as feature quantities related to the anomaly state in the anomaly feature quantity storage unit 230 based on the diagnosis result. At this time, the feature quantity management unit 180 stores the information related to the content of the anomaly diagnosed by the status diagnosis unit 160 in association with the feature quantities related to the anomaly state.

[0065] Furthermore, when diagnosing an unknown anomaly in the industrial machinery 2, the status diagnosis unit 160 of this embodiment calculates the similarity between the feature quantity generated by the feature quantity generation unit 120 and each feature quantity stored in the anomaly feature quantity storage unit 230. For example, the status diagnosis unit 160 can consider the feature quantity as a vector with each value contained in that feature quantity as an element, and calculate the Euclidean distance between the feature quantities; the smaller the value, the higher the similarity. Then, the status diagnosis unit 160 can diagnose that the anomaly associated with the feature quantity of the anomaly state with the highest similarity is the currently occurring anomaly. If there is no feature quantity associated with an anomaly state within a predetermined distance of the feature quantity generated by the feature quantity generation unit 120, the status diagnosis unit 160 can diagnose the current anomaly as an unknown anomaly.

[0066] The anomaly detection device 1 of this embodiment, which has the above structure, stores the feature quantity generated when an anomaly occurred in the industrial machinery 2 in the past as the feature quantity involved in the anomaly state and the anomaly content. When a new anomaly occurs, if the anomaly feature quantity storage unit 230 stores the feature quantity involved in the anomaly state that is similar to the generated feature quantity, it is treated as a known anomaly. If the anomaly feature quantity storage unit 230 does not store the feature quantity involved in the anomaly state that is similar, it is diagnosed as an unknown anomaly.

[0067] As a variation of the anomaly detection device 1 in this embodiment, when the anomaly is diagnosed as a known anomaly based on the feature quantities related to the anomaly state, the state diagnosis unit 160 can instruct the feature model management unit 170 to update the existing anomaly feature model related to the anomaly state. Similar feature quantities are used because anomalies belonging to the same group have high accuracy. At this time, the state diagnosis unit 160 sends the feature quantities diagnosed as known anomalies and the feature quantities related to the anomaly state associated with the known anomaly content stored in the anomaly feature quantity storage unit 230 to the feature model management unit 170. The feature model management unit 170 uses the sent feature quantities to update the anomaly feature model associated with the anomaly state. Thus, when a known anomaly is diagnosed, the anomaly feature model is updated accordingly with the diagnosis result, thereby predictably improving the accuracy of anomaly detection.

[0068] The embodiments of this disclosure have been described in detail above, but this disclosure is not limited to the various embodiments described above. Various additions, substitutions, modifications, and partial deletions can be made to these embodiments without departing from the spirit of the invention, or without departing from the idea and spirit of this disclosure derived from the content described in the claimed scope and its equivalents. For example, in the embodiments described above, the order of each action and the order of each process are shown as an example and are not limited thereto. The same applies to the use of numerical values ​​or mathematical formulas in the description of the embodiments described above.

[0069] The following are notes regarding embodiments of this disclosure.

[0070] (Note 1)

[0071] An anomaly detection device (1) according to one aspect of this disclosure includes: a data acquisition unit (110) that acquires data related to the state of an industrial machine (2); a feature quantity generation unit (120) that generates feature quantities representing state characteristics of the industrial machine (2) based on the data related to the state; a feature quantity storage unit (220) that stores the feature quantities; a feature model storage unit (210) that stores at least one feature model; and an inference calculation unit (140) that uses the feature model to calculate an evaluation value of the state of the industrial machine (2) based on the feature quantities; anomaly detection device (1) The determination unit (150) determines that the state of the industrial machinery (2) has become abnormal based on the calculation result of the inference calculation unit (140); the state diagnosis unit (160) diagnoses the state of the industrial machinery (2) when the abnormality determination unit (150) determines that the industrial machinery (2) has become abnormal; and the feature model management unit (170) generates a feature model using the feature quantities stored in the feature quantity storage unit (220) based on the diagnosis result of the state diagnosis unit (160), and adds it to the feature model storage unit (210).

[0072] (Note 2)

[0073] The inference calculation unit (140) of the anomaly detection device (1) of other embodiments of the present disclosure uses the various feature models stored in the feature model storage unit (210) to calculate the evaluation value of the state of the industrial machinery (2).

[0074] (Note 3)

[0075] The anomaly determination unit (150) of the anomaly detection device (1) of the other manner disclosed herein uses a predetermined determination formula for at least one evaluation value calculated by the inference calculation unit (140) to determine that the state of the industrial machinery (2) has become abnormal.

[0076] (Note 4)

[0077] When the anomaly detection device (1) of other embodiments of the present disclosure is equipped with a status diagnosis unit (160), when the anomaly determination unit (150) determines that an unknown anomaly has occurred, it accepts the inspection information of the industrial machinery (2) and diagnoses whether there is an anomaly based on the accepted inspection information.

[0078] (Note 5)

[0079] In the anomaly detection device (1) of the other manner disclosed herein, the state diagnosis unit (160) diagnoses the state of the industrial machinery (2) when the anomaly determination unit (150) determines that an anomaly has occurred in the state of the industrial machinery for a predetermined time or a predetermined number of times. The feature model management unit (170) generates a feature model using the feature quantity generated during the period when the anomaly determination unit (150) determines that an anomaly has occurred.

[0080] (Note 6)

[0081] The anomaly detection device (1) of other embodiments of the present disclosure further includes: an anomaly feature quantity storage unit (230) which stores feature quantities generated when the industrial machinery (2) is in an abnormal state; and a feature quantity management unit (180) which, when the state diagnosis unit (160) diagnoses the industrial machinery (2) as being in an abnormal state, adds the feature quantities used by the feature model management unit (170) to generate a feature model to the anomaly feature quantity storage unit (230).

[0082] (Note 7)

[0083] The status diagnosis unit (160) of the anomaly detection device (1) of other embodiments of the present disclosure diagnoses whether the anomaly is a known anomaly based on the feature quantity generated by the feature quantity generation unit (120) and the feature quantity stored in the anomaly feature quantity storage unit (230).

[0084] (Note 8)

[0085] The anomaly detection device (1) of other embodiments of the present disclosure further includes: an output unit (190) that displays and outputs the determination result of the anomaly determination unit (150).

[0086] (Note 9)

[0087] The feature model management unit (170) of the anomaly detection device (1) of other embodiments of the present disclosure updates or deletes the feature model stored in the feature model storage unit (210) based on the operator's operation.

[0088] (Postscript 10)

[0089] One aspect of this disclosure includes a computer-readable recording medium containing a program that enables a computer to operate as the following components: a data acquisition unit (110) that acquires data related to the state of industrial machinery (2); a feature quantity generation unit (120) that generates feature quantities representing state characteristics of the industrial machinery (2) based on the data related to the state; a feature quantity storage unit (220) that stores the feature quantities; a feature model storage unit (210) that stores at least one feature model; and an inference calculation unit (140) that uses the feature model to calculate the state characteristics of the industrial machinery (2) based on the feature quantities. The evaluation value of the state; the anomaly determination unit (150), which determines that the state of the industrial machinery (2) has become abnormal based on the calculation result of the inference calculation unit (140); the state diagnosis unit (160), which diagnoses the state of the industrial machinery (2) when the anomaly determination unit (150) determines that the industrial machinery (2) has become abnormal; and the feature model management unit (170), which generates a feature model using the feature quantity stored in the feature quantity storage unit (220) based on the diagnosis result of the state diagnosis unit (160), and adds it to the feature model storage unit (210).

[0090] Explanation of reference numerals in the attached figures

[0091] 1. Anomaly detection device

[0092] 2 Industrial Machinery

[0093] 5 Networks

[0094] 6 Fog Computer

[0095] 7 cloud servers

[0096] 11 CPU

[0097] 12 ROM

[0098] 13 RAM

[0099] 14 Non-volatile memory

[0100] 15 interfaces

[0101] 16 PLC

[0102] 17 I / O Units

[0103] Interfaces 18, 19, 20, and 21

[0104] 22 bus

[0105] 30-axis control circuit

[0106] 40 servo amplifier

[0107] 50 servo motors

[0108] 60 spindle control circuit

[0109] 61 spindle amplifier

[0110] 62 spindle motor

[0111] 63-position encoder

[0112] 70 display devices

[0113] 71 Input Device

[0114] 72 External Devices

[0115] 100 machine learning devices

[0116] 101 processor

[0117] 102 ROM

[0118] 103 RAM

[0119] 104 non-volatile memory

[0120] 110 Data Acquisition Department

[0121] 120 Feature Generation Unit

[0122] 140 Inference Calculation Department

[0123] 150 Anomaly Detection Department

[0124] 160 Status Diagnostic Department

[0125] 170 Feature Model Management Department

[0126] 180 Feature Quantity Management Department

[0127] 190 Output Section

[0128] 210 Feature Model Storage Section

[0129] 220 Feature Quantity Storage Unit

[0130] 230 Abnormal Feature Quantity Storage Department.

Claims

1. An anomaly detection device, characterized in that, The anomaly detection device includes: The data acquisition department acquires data related to the status of industrial machinery; The feature quantity generation unit generates feature quantities representing the state characteristics of the industrial machinery based on data related to the state. A feature storage unit that stores the feature values; A feature model storage unit, which stores at least one feature model; The inference calculation unit uses the feature model to calculate an evaluation value of the state of the industrial machinery based on the feature quantities; The anomaly determination unit determines, based on the calculation results of the inference calculation unit, that the state of the industrial machinery has become abnormal. The status diagnosis unit diagnoses the status of the industrial machinery when the anomaly determination unit determines that an anomaly has occurred in the industrial machinery; and The feature model management unit generates a feature model using the feature quantities stored in the feature quantity storage unit based on the diagnosis results of the status diagnosis unit, and appends it to the feature model storage unit.

2. The anomaly detection device according to claim 1, characterized in that, The inference calculation unit uses each feature model stored in the feature model storage unit to calculate the evaluation value of the state of the industrial machinery.

3. The anomaly detection device according to claim 1, characterized in that, The anomaly determination unit uses a predetermined determination formula based on at least one evaluation value calculated by the inference calculation unit to determine that the state of the industrial machinery has become abnormal.

4. The anomaly detection device according to claim 1, characterized in that, When the anomaly determination unit determines that an unknown anomaly has occurred, the status diagnosis unit receives the inspection information of the industrial machinery and diagnoses whether there is an anomaly based on the received inspection information.

5. The anomaly detection device according to claim 1, characterized in that, If the anomaly determination unit determines that an anomaly has occurred in the state of the industrial machinery for a predetermined time or a predetermined number of times, the state diagnosis unit diagnoses the state of the industrial machinery. The feature model management unit generates a feature model that uses the feature quantity generated during the period when the anomaly determination unit determines that the feature is abnormal.

6. The anomaly detection device according to claim 1, characterized in that, The anomaly detection device also includes: An abnormal feature quantity storage unit stores feature quantities generated when the industrial machinery experiences an abnormal state; and The feature quantity management unit, when the status diagnosis unit diagnoses the industrial machinery as having an abnormal status, adds the feature quantities used by the feature model management unit to generate the feature model to the abnormal feature quantity storage unit.

7. The anomaly detection device according to claim 6, characterized in that, The status diagnosis unit diagnoses whether the anomaly is a known anomaly based on the feature quantity generated by the feature quantity generation unit and the feature quantity stored in the anomaly feature quantity storage unit.

8. The anomaly detection device according to claim 1, characterized in that, The anomaly detection device further includes an output unit that displays and outputs the determination result of the anomaly determination unit.

9. The anomaly detection device according to claim 1, characterized in that, The feature model management department updates or deletes feature models stored in the feature model storage department based on operator operations.

10. A computer-readable recording medium, characterized in that, Its records contain programs that enable the computer to function as the following components: The data acquisition department acquires data related to the status of industrial machinery; The feature quantity generation unit generates feature quantities representing the state characteristics of the industrial machinery based on data related to the state. A feature storage unit that stores the feature values; A feature model storage unit, which stores at least one feature model; The inference calculation unit uses the feature model to calculate an evaluation value of the state of the industrial machinery based on the feature quantities; The anomaly determination unit determines, based on the calculation results of the inference calculation unit, that the state of the industrial machinery has become abnormal. The status diagnosis unit diagnoses the status of the industrial machinery when the anomaly determination unit determines that an anomaly has occurred in the industrial machinery; and The feature model management unit generates a feature model using the feature quantities stored in the feature quantity storage unit based on the diagnosis results of the status diagnosis unit, and appends it to the feature model storage unit.

Images