Abnormality detection device, control device, abnormality detection method, program, and integrated abnormality detection system
Patent Information
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Filing Date
- 2023-08-21
- Publication Date
- 2026-05-21
Abstract
Description
Anomaly detection device, control device, anomaly detection method, program, and integrated anomaly detection system
[0001] The present disclosure relates to an anomaly detection device, a control device, an anomaly detection method, a program, and an integrated anomaly detection system that detect anomalies in feature quantities of industrial machinery based on variation patterns of the feature quantities.
[0002] Breakdowns of industrial machinery such as machine tools and robots cause a decrease in the availability rate at production sites. To prevent a decrease in availability rate and the generation of defective workpieces, there is a need to implement preventive maintenance for industrial machinery. To achieve this, methods have been proposed that collect information during operation (control signals, current values, vibration values, etc.) and model features using machine learning to detect anomalies. For example, see Patent Documents 1 to 3. These conventional technologies employ a method of setting a threshold value and issuing a notification when an evaluation value exceeds the threshold, indicating an anomaly.
[0003] JP 2020-123191 A JP 2022-128824 A International Publication No. 2023 / 007825
[0004] However, measurement data shows that behavioral changes in industrial machinery, such as machine tools, are not as simple as a dichotomy based on a threshold. For example, when changes in evaluation values are viewed over time, they can change stepwise or fluctuate up and down. In such situations, conventional methods that notify anomalies based on evaluation values exceeding a predefined threshold result in frequent notifications, preventing meaningful decision-making for operators. Adjusting the threshold does not fundamentally solve the difficulty of dichotomizing normal and abnormal behaviors. Measurement data also shows that the behavior of industrial machinery is unique. Therefore, it is difficult to create and apply general-purpose models and thresholds from large amounts of data. Because the general mechanisms of failure in industrial machinery, such as machine tools, have not been elucidated, the role required of anomaly detection systems is to match data anomalies with abnormalities in the industrial machinery as closely as possible.
[0005] Therefore, it is desirable to detect abnormalities by capturing collective data on changes in the condition of industrial machinery, rather than the data on each individual piece of industrial machinery, in order to present the minimum necessary inspection timing.
[0006] (1) One aspect of the anomaly detection device disclosed herein includes a feature acquisition unit that acquires a context in the production operation of an industrial machine and a state quantity related to the production operation state of the industrial machine, and acquires feature quantities related to the production operation state of the industrial machine based on the acquired context and the state quantity; a feature model operation unit that generates, updates, deletes, and adds feature models based on the context and the feature quantities; a feature model storage unit that stores the feature models generated by the feature model operation unit; an inference calculation unit that calculates an evaluation value of the operating state of the industrial machine based on the feature quantities and the feature models; a fluctuation pattern management unit that stores and manages predetermined fluctuation patterns of the evaluation value; a fluctuation pattern classification unit that classifies the fluctuation pattern of the evaluation value of the operating state of the industrial machine based on the calculation result of the inference calculation unit and the predetermined fluctuation pattern stored in the fluctuation pattern management unit; and a feature abnormality determination unit that determines whether a feature value of the industrial machine is abnormal based on the classification result of the fluctuation pattern classification unit.
[0007] (2) One aspect of the control device of the present disclosure is a control device that controls a motor that drives a mechanism of an industrial machine, and includes the abnormality detection device of (1).
[0008] (3) One aspect of the anomaly detection method disclosed herein is an anomaly detection method that causes a computer to operate as an anomaly detection device, and includes: a feature acquisition process that acquires a context of production operation of an industrial machine and a state quantity related to a production operation state of the industrial machine, and acquires feature quantities related to the production operation state of the industrial machine based on the acquired context and the state quantity; a feature model operation process that generates, updates, deletes, and adds a feature model based on the context and the feature quantities; a feature model storage process that stores the feature model generated in the feature model operation process; an inference calculation process that calculates an evaluation value of the operating state of the industrial machine based on the feature quantities and the feature model; a fluctuation pattern management process that stores and manages a predetermined fluctuation pattern of the evaluation value; a fluctuation pattern classification process that classifies the fluctuation pattern of the evaluation value of the operating state of the industrial machine based on a calculation result of the inference calculation process and the predetermined fluctuation pattern stored in the fluctuation pattern management process; and a feature anomaly determination process that determines whether a feature value of the industrial machine is abnormal based on the classification result of the fluctuation pattern classification process.
[0009] (4) One aspect of the program of the present disclosure is a program that causes a computer to realize the following: a feature acquisition process that acquires a context in the production operation of an industrial machine and a state quantity related to the production operation state of the industrial machine, and acquires feature quantities related to the production operation state of the industrial machine based on the acquired context and the state quantity; a feature model operation process that generates, updates, deletes, and adds a feature model based on the context and the feature quantities; a feature model storage process that stores the feature model generated in the feature model operation process; an inference calculation process that calculates an evaluation value of the operating state of the industrial machine based on the feature quantities and the feature model; a fluctuation pattern management process that stores and manages a predetermined fluctuation pattern of the evaluation value; a fluctuation pattern classification process that classifies a fluctuation pattern of the evaluation value of the operating state of the industrial machine based on a calculation result of the inference calculation process and the predetermined fluctuation pattern stored in the fluctuation pattern management process; and a feature abnormality determination process that determines whether a feature value of the industrial machine is abnormal based on the classification result of the fluctuation pattern classification process.
[0010] (5) One aspect of the integrated anomaly detection system disclosed herein comprises two or more anomaly detection devices (1), and the fluctuation pattern management unit is shared in part or in whole between the anomaly detection devices, and the fluctuation pattern detected by one anomaly detection device is used by the other anomaly detection devices.
[0011] 1 is a diagram illustrating an example of a hardware configuration when an anomaly detection device according to a first embodiment is connected to a numerical control device. FIG. 2 is a schematic functional block diagram when an anomaly detection device according to the first embodiment is connected to a numerical control device. FIG. 3 is a diagram illustrating an example of time-series data of the speed and torque of a spindle motor to which a tool is attached in a machine tool when not cutting, as state quantities acquired by a state quantity acquisition unit. FIG. 4 is a diagram illustrating an example of the operation of a feature quantity creation unit. FIG. 5 is a diagram illustrating an example of the relationship between a feature model and feature quantities. FIG. 6 is a diagram illustrating an example of the relationship between a feature model and feature quantities. FIG. 7 is a diagram illustrating an example of a predetermined fluctuation pattern. FIG. 8 is a diagram illustrating an example of a predetermined fluctuation pattern. FIG. 9 is a diagram illustrating an example of time-series data of evaluation values calculated by an inference calculation unit. FIG. 10 is a flowchart illustrating an anomaly detection process of the anomaly detection device. FIG. 11 is a schematic functional block diagram when an anomaly detection device according to a second embodiment is connected to a numerical control device. FIG. 12 is a diagram illustrating an example of how to select data used in machine learning in the case of a "struggle type". FIG. 13 is a diagram illustrating an example of how to select data used in machine learning in the case of a "rise and pause type". FIG. 14 is a flowchart illustrating an anomaly detection process of the anomaly detection device.
[0012] First Embodiment An anomaly detection device according to a first embodiment will be described in detail below with reference to the drawings. Here, a case where a machine tool, a type of industrial machine, is connected as an anomaly detection target is illustrated. The present invention is also applicable to various industrial machines, such as machine tools, industrial robots, service robots, forging machines, laser processing machines, and injection molding machines. FIG. 1 illustrates an example of a hardware configuration when connecting an anomaly detection device according to the first embodiment to a numerical control device. As illustrated in FIG. 1 , the anomaly detection device 20 is connected to a numerical control device 10. The numerical control device 10 and the anomaly detection device 20 are directly connected to each other via a connection interface (not shown). The numerical control device 10 and the anomaly detection device 20 may be connected to each other and communicate via a network (not shown), such as a local area network (LAN) or the Internet. Although the numerical control device 10 and the anomaly detection device 20 are described as separate devices, the anomaly detection device 20 may be included in the numerical control device 10, as described below.
[0013] The numerical controller 10 is a numerical controller known to those skilled in the art for controlling the operation of, for example, a machine tool (not shown). If the machine tool (not shown) is a robot or the like, the numerical controller 10 may be a robot control device or the like. The numerical controller 10 includes a central processing unit (CPU) 11, a read-only memory (ROM) 12, a random access memory (RAM) 13, an axis control circuit 14, a servo amplifier 15, an interface (INT) 16, and a bus 17. The numerical controller 10 is also connected to a display device 130. The CPU 11 is a processor that controls the entire numerical controller 10. The CPU 11 reads a system program stored in the ROM 12 via the bus 17 and controls the entire numerical controller 10 in accordance with the system program. The RAM 13 temporarily stores temporary calculation data, display data, and various data input by an operator via an input device (not shown), such as a keyboard or touch panel. The axis control circuit 14 receives axis movement commands from the CPU 11 via the bus 17 and outputs the commands to the servo amplifier 15 to control the axes of the machine tool (not shown). Upon receiving the commands, the servo amplifier 15 drives the motor 120 that moves the axis of the machine tool (not shown). The axis motor 120 incorporates a position / speed detector (not shown), and feeds back position / speed feedback signals from the position / speed detector to the axis control circuit 14, thereby performing position / speed feedback control. While only one set of the axis control circuit 14, servo amplifier 15, and motor 120 is shown in the hardware configuration diagram of FIG. 1 , in reality, as many sets as there are axes in the machine tool (not shown). The INTs 16 and 24 are interfaces for connecting the numerical control device 10 and the abnormality detection device 20. The display 130 is, for example, a liquid crystal display, and is used to operate the machine tool (not shown) that includes the numerical control device 10 as a component. The display 130 may also display the detection results of abnormality detection in a machine tool (not shown) by the abnormality detection device 20 (described later).
[0014] The anomaly detection device 20 is, for example, a computer such as a personal computer, fog computer, or cloud server, and includes a processor 21, a ROM 22, a RAM 23, an INT 24, and a bus 25. The anomaly detection device 20 is also connected to a display 230. The processor 21 is, for example, a CPU, which reads a system program stored in the ROM 22 and controls the entire anomaly detection device 20 in accordance with the system program. The RAM 23 temporarily stores temporary calculation data, display data, and various data input by an operator via an input device (not shown), such as a keyboard or touch panel. The INT 24 transmits and receives various data to and from the numerical control device 10 via the INT 16 of the numerical control device 10. The display 230 is, for example, a liquid crystal display, and displays the results of processing by the anomaly detection device 20.
[0015] FIG. 2 is a schematic functional block diagram of an anomaly detection device 20 according to the first embodiment connected to a machine tool 30, which is a type of industrial machine. The functional blocks shown in FIG. 2 are realized by the CPU 11 of the numerical control device 10 shown in FIG. 1 and the processor 21 of the anomaly detection device 20 controlling the operation of each unit of the device in accordance with their respective system programs. Specifically, in FIG. 2 , for example, the numerical control device 10 includes a numerical control unit 110, a context acquisition unit 111, and a state quantity acquisition unit 112. The anomaly detection device 20 includes a feature quantity acquisition unit 210, a feature model operation unit 211, an inference calculation unit 212, a fluctuation pattern management unit 213, a fluctuation pattern classification unit 214, a feature quantity anomaly determination unit 215, and a feature model storage unit 216. The feature quantity acquisition unit 210 includes a state data extraction unit 2101 and a feature quantity creation unit 2102.
[0016] The numerical control unit 110 controls the machine tool 30 by executing a machining program stored in a memory (not shown). The numerical control unit 110 sequentially reads and analyzes blocks of the machining program stored in the memory (not shown), calculates the movement amount of the motor 31 for each control cycle based on the analysis results, and controls the motor 31 according to the calculated movement amount for each control cycle. The machine tool 30 drives a mechanical unit 32, including, for example, a ball screw used as a feed axis and a mechanism used as a spindle, by driving the motor 31. In this way, the machine tool 30 machines a workpiece. Note that, although one motor 31 is connected to one mechanical unit 32 of the machine tool 30 in FIG. 2 , the same number of motors 31 as the number of axes included in the mechanical unit 32 of the machine tool 30 may be provided. Furthermore, a single mechanical unit 32 may be driven by multiple motors 31.
[0017] The context acquisition unit 111 acquires contexts (production status, operating status, environmental status, etc.) during production operation executed by the numerical control unit 110 (and the machine tool 30 controlled by the numerical control unit 110) and outputs the acquired contexts to the anomaly detection device 20. Examples of the context during production operation include the operation pattern of the motor 31 during production operation (spindle rotation speed, feed rate, etc.), the purpose of the currently performed machining (rough machining, finish machining, etc.), the purpose of the currently performed drive of the moving part (rapid feed, cutting feed, etc.), the type of tool used for machining, and workpiece information indicating the hardness and material of the workpiece to be machined. The context acquisition unit 111 acquires the context during production operation, which is comprehensively determined based on the machining conditions commanded by the machining program, setting information set in the numerical control unit 110 by an operator via an input device (not shown), setting information set in the numerical control unit 110 by another computer connected via a network (not shown), information detected by devices such as sensors separately provided in the numerical control unit 110, signal values acquired from a programmable logic controller (PLC) (not shown), etc. The context acquisition unit 111 outputs the acquired context to a feature amount acquisition unit 210 and a feature model operation unit 211 of the anomaly detection device 20, which will be described later.
[0018] The state quantity acquisition unit 112 acquires state quantities of the production operation state of the numerical control unit 110 (and the machine tool 30 controlled by the numerical control unit 110). Examples of the state quantities of the production operation state include the load (current value and torque) of the spindle, the load (current value and torque) of the feed axis, the spindle rotation speed, the feed axis speed, the feed axis position, the position feedback value, the temperature, vibration value, and sound of the motor 31. Furthermore, the state quantity acquisition unit 112 acquires, as state quantities, for example, the actual waveforms of the rotation speed and torque value of the motor 31 that drives the mechanism unit 32 of the machine tool 30 controlled by the numerical control unit 110. The state quantity acquisition unit 112 outputs the acquired state quantities to a feature quantity acquisition unit 210 of the abnormality detection device 20, which will be described later.
[0019] The feature quantity acquisition unit 210 acquires a context of production operation of the machine tool 30 and a state quantity related to the production operation state of the machine tool 30, and acquires a feature quantity related to the production operation state of the machine tool 30 based on the acquired context and state quantity. Specifically, the feature quantity acquisition unit 210 has, for example, a state data extraction unit 2101 and a feature quantity creation unit 2102. The state data extraction unit 2101 extracts state data from the state quantities acquired by the state quantity acquisition unit 112 to be used for inference processing by the inference calculation unit 212 (described later) and feature model update processing by the feature model operation unit 211 (described later), etc. The state data extraction unit 2101 extracts state data to be used for inference processing, etc., from the state quantities acquired by the state quantity acquisition unit 112 using a predetermined extraction pattern based on the context of production operation input from the context acquisition unit 111. The extraction pattern used by the state data extraction unit 2101 is a predetermined data processing method in which parameters are determined based on the context, and may be, for example, data processing such as setting an extraction interval of time-series data obtained based on the context, selecting data, or changing the scale of state quantities based on the context. Furthermore, the extraction pattern used by the state data extraction unit 2101 may be registered in advance in memory by an operator.
[0020] The following describes extraction of state data from state quantities by the state data extraction unit 2101 using a predetermined extraction pattern based on the context of production operation. Fig. 3 is a diagram showing an example of time-series data of the speed and torque of a spindle motor to which a tool is attached in machine tool 30 that is not performing machining or cutting, as state quantities acquired by the state quantity acquisition unit 112. In Fig. 3, the time-series data of the speed and torque of the spindle motor is shown on the left side, and the torque status data extracted by the status data extraction unit 2101 is shown on the right side. Note that the time-series data of the speed and torque of the spindle motor is data sampled at a sampling frequency of, for example, 0.001 seconds, and is time-series data at a predetermined timing when the spindle motor is driven, i.e., when the spindle motor is rotating at a constant speed of approximately 4000 rpm and running idle. The status data extraction unit 2101 extracts, as status data, torque values in a specific section of the torque time-series data, for example, a section (e.g., 0.1 seconds) indicated by a dashed rectangle in which the spindle motor rotates at a constant speed of approximately 4000 rpm and idles without any external force. As shown on the right side of FIG. 3 , the extracted status data, torque, is not a constant value but exhibits minute vibrations, as shown on the left side of FIG. 3 . Then, the status data extraction unit 2101 extracts, as status data, torques exhibiting minute vibrations, as shown on the right side of FIG. 3 , from multiple spindle motor speed and torque time-series data when the spindle motor rotates at a constant speed (e.g., approximately 4000 rpm) and idles, as shown on the left side of FIG. 3 . The status data extraction unit 2101 outputs the extracted status data to the feature creation unit 2102.
[0021] The feature quantity creation unit 2102 creates feature quantities that indicate characteristics of the operating state of the machine tool 30 controlled by the numerical control unit 110, based on the state data extracted by the state data extraction unit 2101. The feature quantities created by the feature quantity creation unit 2102 indicate characteristics of the operating state of the machine tool 30, and are useful information as material for making decisions when detecting an abnormality in the operating state of the machine tool 30 during production operations performed by the numerical control unit 110 (and the machine tool 30). The feature quantities created by the feature quantity creation unit 2102 are also input data when the inference calculation unit 212, described below, performs inference using a feature model. For example, as shown in FIG. 4 , the feature quantity creation unit 2102 creates feature quantities by performing a Fourier transform on each piece of state data (torque) extracted by the state data extraction unit 2101 into the frequency domain, normalizing the amplitude or power density, or calculating statistics (average, variance, etc.).
[0022] The feature model operation unit 211 generates, updates, deletes, and adds feature models stored in the feature model storage unit 216, which will be described later, based on the context of production operation input from the context acquisition unit 111 and feature amounts indicating features of the operating state of the machine tool 30 created by the feature amount creation unit 2102. The feature model operation unit 211 selects a feature model to be generated, updated, deleted, or added based on the context of production operation input from the context acquisition unit 111, and performs machine learning on the selected feature model using the feature amounts indicating features of the state of production operation created by the feature amount creation unit 2102. For example, if a feature model associated with (a combination of) the context of production operation input from the context acquisition unit 111 is not stored in the feature model storage unit 216, the feature model operation unit 211 newly generates a tentative reference (feature model) associated with the context (a combination of) the context from, for example, the frequency distribution of micro-vibrations of the 50 state data shown in FIG. Furthermore, when a feature model associated with (a combination of) contexts in production operation input from the context acquisition unit 111 is stored in the feature model storage unit 216, and a state of a feature quantity indicating a characteristic of the state of production operation created by the feature quantity creation unit 2102 is closer to a range of the feature model than a predetermined distance, the feature model operation unit 211 may perform machine learning on the feature model including the feature quantity created by the feature quantity creation unit 2102, and update the feature model. Furthermore, for example, as shown in FIG. 5 , when a feature model A associated with (a combination of) contexts in production operation input from the context acquisition unit 111 is stored in the feature model storage unit 216, and a state B of the feature quantity created by the feature quantity creation unit 2102 is away from the range of feature model A by a predetermined distance or more, the feature model operation unit 211 may perform machine learning using the feature quantity of state B including the feature quantity created by the feature quantity creation unit 2102, and generate and add a new feature model B.
[0023] Furthermore, for example, as shown in FIG. 6 , when feature models A and B associated with the context (combination) of the production operation input from the context acquisition unit 111 are stored in the feature model storage unit 216, and state C of the feature quantities created by the feature quantity creation unit 2102 is in the middle of the ranges of feature models A and B, the feature model operation unit 211 may perform machine learning using the feature quantities in the ranges of feature models A and B and the feature quantities of state C including the feature quantities created by the feature quantity creation unit 2102 to generate a new feature model A′. The feature model operation unit 211 may delete feature models A and B from the feature model storage unit 216, update to a newly generated feature model A′, and store the updated feature model A′ in the feature model storage unit 216. Note that the machine learning in the feature model operation unit 211 can use a variational autoencoder (VAE), a generative adversarial network (GAN), or the like. Any technique may be used as long as the inference calculation unit 212, described later, can calculate an evaluation value of the operating state.
[0024] The inference calculation unit 212 calculates an evaluation value of the operating state of the machine tool 30 executed by the numerical control unit 110 (and the machine tool 30) based on a feature model selected from the feature model storage unit 216 based on the context of the production operation of the machine tool 30 input from the context acquisition unit 111, and on the feature amount created by the feature amount creation unit 2102. Specifically, the inference calculation unit 212 inputs a new feature amount indicating the characteristics of the state of production operation created by the feature amount creation unit 2102 into the feature model (provisional reference) selected based on the context of the production operation of the machine tool 30 input from the context acquisition unit 111, and calculates an evaluation value (degree of data abnormality) indicating how similar the new feature amount is to the feature model (provisional reference). The evaluation value obtained as a result of calculation (inference) by the inference calculation unit 212 may indicate, for example, a normal / abnormal classification of the operating state of the machine tool 30, information indicating an abnormal part of the operating state of the machine tool 30 (such as an abnormal bearing in the motor 31 or damage to the connection between the motor 31 and the mechanical unit 32), or the distance between the current operating state of the machine tool 30 and the distribution of normal operating states of the machine tool 30. The inference calculation unit 212 stores the calculated evaluation value in a storage unit (not shown).
[0025] The fluctuation pattern management unit 213 stores and manages predetermined fluctuation patterns of the evaluation value. Specifically, the fluctuation pattern management unit 213 pre-stores and manages, for example, four fluctuation patterns of the evaluation value shown in FIGS. 7A to 7D. In FIGS. 7A to 7D, the vertical axis indicates the evaluation value, and the horizontal axis indicates the date. The fluctuation pattern of the evaluation value in FIG. 7A is a fluctuation pattern in which the evaluation value increases and then settles back to its original value, also known as a "back and forth type." When the evaluation value exhibits this fluctuation pattern, the evaluation value has returned to its original state, and the feature abnormality determination unit 215, described below, determines that the machine tool 30 is normal. The dashed line for the evaluation value "8.00" and the dashed line for the evaluation value "6.00" are examples of conventional thresholds, indicating that the thresholds are not significantly related to whether the machine tool 30 is normal or abnormal. The fluctuation pattern of the evaluation value in FIG. 7B is a fluctuation pattern in which the evaluation value repeatedly fluctuates up and down, also known as a "flooding type." When the evaluation value shows this fluctuation pattern, the evaluation value fluctuates between two states (i.e., a low evaluation value state and a high evaluation value state), and therefore, characteristic amount abnormality determination unit 215, described later, determines that machine tool 30 is abnormal. The fluctuation pattern of the evaluation value in FIG. 7C is a fluctuation pattern in which the evaluation value rises and remains high, and is also referred to as a "rise and pause type." When the evaluation value shows this fluctuation pattern, it is considered that a transition has occurred to a new state, and characteristic amount abnormality determination unit 215, described later, determines that machine tool 30 is abnormal. The fluctuation pattern of the evaluation value in FIG. 7D is a fluctuation pattern in which the evaluation value continues to rise, and is also referred to as an "upward type." When the evaluation value shows this fluctuation pattern, the state continues to change, and therefore, characteristic amount abnormality determination unit 215, described later, determines that machine tool 30 is abnormal. Note that, although fluctuation pattern management unit 213 stores four fluctuation patterns, it may also store three, five, or other fluctuation patterns.
[0026] The fluctuation pattern classification unit 214 classifies the fluctuation pattern of the evaluation value of the operating state of the machine tool 30 based on the calculation results of the inference calculation unit 212 and the predetermined fluctuation pattern stored in the fluctuation pattern management unit 213. Specifically, the fluctuation pattern classification unit 214 reads out, for example, evaluation values for three to ten days from a storage unit (not shown) as the calculation results of the inference calculation unit 212. FIG. 8 is a diagram showing an example of time-series data of the evaluation value calculated by the inference calculation unit 212. Specifically, for example, the fluctuation pattern classification unit 214 matches the fluctuation pattern of the evaluation value in the period (1) with the fluctuation patterns of FIGS. 7A to 7D stored in the fluctuation pattern management unit 213 and classifies it as a "back and forth type." Alternatively, the fluctuation pattern classification unit 214 matches the fluctuation pattern of the evaluation value in the period (2) with the fluctuation patterns of FIGS. 7A to 7D stored in the fluctuation pattern management unit 213 and classifies it as an "upward and downward type." The fluctuation pattern classification unit 214 outputs the classification result to the feature amount abnormality determination unit 215 .
[0027] Feature quantity abnormality determination unit 215 determines whether the feature quantity of machine tool 30 is abnormal based on the classification result of fluctuation pattern classification unit 214. Specifically, for example, when the fluctuation pattern of the evaluation value is the "back and forth type" in FIG. 7A , the feature quantity abnormality determination unit 215 determines that the feature quantity of machine tool 30 is normal based on the risk assessment linked to the fluctuation pattern, since the feature quantity has returned to the original state. Also, when the fluctuation pattern of the evaluation value is the "struggle type" in FIG. 7B , the feature quantity abnormality determination unit 215 determines that the feature quantity of machine tool 30 is abnormal based on the risk assessment linked to the fluctuation pattern, since the feature quantity has fluctuated between two states. Also, when the fluctuation pattern of the evaluation value is the "rising pause type" in FIG. 7C , the feature quantity abnormality determination unit 215 considers that a transition has occurred to a new state, and determines that the feature quantity of machine tool 30 is abnormal based on the risk assessment linked to the fluctuation pattern. Furthermore, if the fluctuation pattern of the evaluation value is the "rising type" shown in Figure 7D, the characteristic abnormality determination unit 215 determines that the characteristic of the machine tool 30 is abnormal based on the risk assessment linked to the fluctuation pattern, since the state is continuously changing.
[0028] When it is determined that the feature of machine tool 30 is abnormal, feature amount abnormality determination unit 215 may display a message to the equipment manager on display 130 of numerical control device 10 and / or display 230 of abnormality detection device 20 indicating that inspection of machine tool 30 is necessary. Then, the equipment manager inspects machine tool 30 based on the displayed need for inspection. In this way, preventive maintenance can be performed.
[0029] The feature model storage unit 216 is, for example, a solid state drive (SSD) or a hard disk drive (HDD), and stores a plurality of feature models associated with combinations of contexts in production operation input from the context acquisition unit 111. The feature model storage unit 216 may be provided in an external numerical control device 10, a cell computer, a fog computer, a cloud server, a database server, or the like. The feature model storage unit 216 stores a plurality of feature models 1, 2, ..., n associated with combinations of contexts (machining status, operating status, environmental status, etc.) in production operation specified by the context acquisition unit 111. Here, the combination of contexts (machining status, operating status, environmental status, etc.) in machining operations refers to the combination of values, value ranges, and enumerations of values that each context can take. For example, if the combination of contexts is a combination of spindle speed, feed rate, cutting signal, tool type, and workpiece information, it can be "spindle speed: 500 to 1000 [rpm], feed rate: 200 to 300 [mm / min], cutting, drill tool, aluminum / steel."
[0030] The feature models stored in the feature model storage unit 216 are created by the feature model operation unit 211 and stored as information capable of configuring one feature model suitable for the inference processing in the inference calculation unit 212. The feature models stored in the feature model storage unit 216 may be created using a learning algorithm such as a multilayer neural network (VAE) or GAN, or may be created using a Bayesian network learning algorithm. Furthermore, the feature models stored in the feature model storage unit 216 may be created using the same learning algorithm or different learning algorithms, and may be created using any learning algorithm that can be used for the inference processing by the inference calculation unit 212. The feature model storage unit 216 may associate one feature model with a combination of contexts in one production operation and store the feature models, or may associate two or more feature models with each other and store the feature models. Furthermore, the feature model storage unit 216 may associate multiple feature models with a combination of contexts in one production operation using the same learning algorithm, or may associate feature models with two or more different learning algorithms and store the feature models. The feature model storage unit 216 may store feature models using different learning algorithms in association with each of context combinations in multiple production operations whose combination ranges overlap. In this case, the feature model storage unit 216 may further store usage conditions, such as required processing capacity and the type of learning algorithm, for feature models corresponding to context combinations in production operations. This allows, for example, the inference calculation unit 212 to select a feature model according to an executable inference process and processing capacity for a context combination in production operations. Note that the feature model storage unit 216 may encrypt and store the feature model generated by the feature model operation unit 211, and decrypt the encrypted feature model when the feature model is read out by the inference calculation unit 212.
[0031] <Abnormality detection processing of abnormality detection device 20> Next, the flow of abnormality detection processing of the abnormality detection device 20 will be described with reference to Fig. 9. Fig. 9 is a flowchart illustrating the abnormality detection processing of the abnormality detection device 20. The flow shown here is executed every time the numerical control device 10 executes a production operation on the machine tool 30.
[0032] In step S11, the feature acquisition unit 210 acquires the context of the production operation of the machine tool 30 and state quantities related to the production operation state of the machine tool 30, and acquires feature quantities related to the production operation state of the machine tool 30 based on the acquired context and state quantities.
[0033] In step S12, the inference calculation unit 212 determines whether or not a feature model corresponding to the context in production operation of the machine tool 30 is stored in the feature model storage unit 216. If a feature model corresponding to the context in production operation of the machine tool 30 is stored in the feature model storage unit 216, the process proceeds to step S14. On the other hand, if a feature model corresponding to the context in production operation of the machine tool 30 is not stored in the feature model storage unit 216, the process proceeds to step S13.
[0034] In step S13, the feature model operation unit 211 generates a new feature model of the context in production operation acquired in step S11, based on the context and feature quantities indicating the features of the operating state of the machine tool 30. The feature model operation unit 211 stores the generated feature model in the feature model storage unit 216.
[0035] In step S14, the inference calculation unit 212 infers (calculates) an evaluation value of the operating state of the machine tool 30 executed by the numerical control unit 110 (and the machine tool 30) based on the feature model selected from the feature model memory unit 216 and the feature acquired in step S11.
[0036] In step S15, the fluctuation pattern classification unit 214 classifies the fluctuation pattern of the evaluation value of the operating state of the machine tool 30 based on the inference result of step S14 and the predetermined fluctuation pattern stored in the fluctuation pattern management unit 213.
[0037] In step S16, feature amount abnormality determination unit 215 determines whether the feature amount of machine tool 30 is abnormal based on the classification result of step S15. If it is determined that the feature amount of machine tool 30 is abnormal, the processing proceeds to step S17. On the other hand, if it is determined that the feature amount of machine tool 30 is normal, abnormality detection device 20 ends the abnormality detection processing.
[0038] In step S17, the characteristic amount abnormality determination unit 215 outputs the necessity of inspecting the machine tool 30.
[0039] As described above, the anomaly detection device 20 according to the first embodiment can detect anomalies to suggest the minimum necessary inspection timing by capturing data on changes in the state of the machine tool 30 collectively, rather than the data for each individual point on the machine tool 30. In other words, the anomaly detection device 20 determines whether the risk of machine failure is high or low based on the fluctuation pattern, and if it determines that the risk is high, it detects an anomaly and prompts inspection of the machine tool 30. Furthermore, the anomaly detection device 20 can ignore transient changes that would be considered an anomaly if determined using a threshold value, thereby reducing unnecessary inspections. Furthermore, the anomaly detection device 20 can evaluate risk by checking the fluctuation pattern when there is a change in the evaluation value. The first embodiment has been described above.
[0040] Second Embodiment Next, a second embodiment will be described. In the first embodiment, the anomaly detection device 20 determines whether a feature value of the operating state of the machine tool 30 is abnormal based on the classification results of the fluctuation pattern of the evaluation value of the operating state of the machine tool 30 executed by the numerical control unit 110 (and the machine tool 30). If the feature value of the machine tool 30 is determined to be abnormal, the anomaly detection device 20 outputs a notification indicating the need for inspection of the machine tool 30. In contrast, in the second embodiment, when the feature value of the machine tool 30 is determined to be abnormal, the anomaly detection device 20A receives input from the maintenance management record or the equipment manager regarding the need for inspection of the machine tool 30, and updates the feature model based on the determination result and the inspection result. This differs from the first embodiment in that the anomaly detection device 20A according to the second embodiment can thus perform anomaly detection to present the minimum necessary inspection timing by capturing data on changes in the state of the industrial machine collectively, rather than the data for each individual point of the industrial machine. Furthermore, by updating the feature model, the anomaly detection device 20A can prevent the same anomaly from being detected multiple times, further reducing the number of inspections. The second embodiment will be described below.
[0041] The hardware configuration of the anomaly detection device 20A according to the second embodiment is the same as that of the first embodiment shown in FIG. 1 , and therefore a detailed description thereof will be omitted. FIG. 10 is a schematic functional block diagram of the anomaly detection device 20A according to the second embodiment connected to a numerical control device 10. Note that elements having the same functions as those of the anomaly detection device 20 shown in FIG. 2 are designated by the same reference numerals, and detailed description thereof will be omitted. As shown in FIG. 10 , the anomaly detection device 20A according to the second embodiment is connected to the numerical control device 10. The numerical control device 10 has the same functions as the numerical control device 10 according to the first embodiment. The numerical control unit 110, the context acquisition unit 111, and the state quantity acquisition unit 112 have the same functions as the numerical control unit 110, the context acquisition unit 111, and the state quantity acquisition unit 112 according to the first embodiment.
[0042] The anomaly detection device 20A includes a feature amount acquisition unit 210, a feature model operation unit 211a, an inference calculation unit 212, a variation pattern management unit 213, a variation pattern classification unit 214, a feature amount abnormality determination unit 215, a feature model storage unit 216, and a machine inspection result holding unit 217. The feature amount acquisition unit 210 includes a state data extraction unit 2101 and a feature amount creation unit 2102. The feature amount acquisition unit 210, the inference calculation unit 212, the variation pattern management unit 213, the variation pattern classification unit 214, the feature amount abnormality determination unit 215, and the feature model storage unit 216 have functions similar to those of the feature amount acquisition unit 210, the inference calculation unit 212, the variation pattern management unit 213, the variation pattern classification unit 214, the feature amount abnormality determination unit 215, and the feature model storage unit 216 in the first embodiment. The state data extraction unit 2101 and the feature creation unit 2102 have the same functions as the state data extraction unit 2101 and the feature creation unit 2102 in the first embodiment.
[0043] The machine inspection result holding unit 217 acquires inspection information from, for example, a system that manages external inspection records. The machine inspection result holding unit 217 may also be an input device such as a keyboard or touch panel (not shown). When the feature amount abnormality determination unit 215 determines that the feature amount of the machine tool 30 is abnormal and outputs the need for inspection of the machine tool 30, the machine inspection result holding unit 217 acquires inspection information on the results of a timely inspection of the machine tool 30 from a system that manages external inspection records. The machine inspection result holding unit 217 may also receive input of the inspection results from the equipment manager. The machine inspection result holding unit 217 outputs the received inspection results to the feature model operation unit 211a.
[0044] Similar to the feature model operation unit 211 in the first embodiment, the feature model operation unit 211a generates, updates, deletes, and adds feature models stored in the feature model storage unit 216 based on the context of production operation input from the context acquisition unit 111 and the feature amounts indicating the features of the operating state of the machine tool 30 created by the feature amount creation unit 2102. Furthermore, the feature model operation unit 211a updates the feature models stored in the feature model storage unit 216 based on the determination result of the feature amount abnormality determination unit 215 and the inspection results of the machine inspection result holding unit 217. Specifically, for example, when the feature amount abnormality determination unit 215 determines that the feature amount of the machine tool 30 is abnormal and the machine inspection result holding unit 217 receives an inspection result indicating that the machine tool 30 is normal, the feature model operation unit 211a performs machine learning including the feature amount determined to be normal in the inspection result, and updates the feature model of the context (combination) of the context of production operation input from the context acquisition unit 111. Furthermore, when the fluctuation pattern classification unit 214 determines that the fluctuation pattern is a "struggle type," the feature model operation unit 211a may perform machine learning using feature amounts including the two states enclosed in circles, as shown in FIG. 11A, to generate and add a new feature model m. Furthermore, when the fluctuation pattern classification unit 214 determines that the fluctuation pattern is a "rise pause type," the feature model operation unit 211a may perform machine learning using feature amounts including the states enclosed in circles, as shown in FIG. 11B, to generate and add a new feature model m. Furthermore, when the fluctuation pattern classification unit 214 determines that the fluctuation pattern is a "rise type," the feature model operation unit 211a may refrain from updating the feature model until the rise settles down.
[0045] <Abnormality detection process of abnormality detection device 20A> Next, the flow of the abnormality detection process of the abnormality detection device 20A will be described with reference to Fig. 12. Fig. 12 is a flowchart explaining the abnormality detection process of the abnormality detection device 20A. Note that the processes of steps S21 to S27 are similar to the processes of steps S11 to S17 in Fig. 9, and therefore description thereof will be omitted.
[0046] In step S28, the machine inspection result holding unit 217 acquires the inspection results of the machine tool 30 in a timely manner in response to the need for inspection output in step S27 from an external system that manages inspection records. The machine inspection result holding unit 217 may also receive input of the inspection results from the equipment manager.
[0047] In step S29, the feature model operation unit 211a updates the feature model based on the determination result in step S26 and the inspection result in step S28.
[0048] As described above, the anomaly detection device 20A according to the second embodiment can detect anomalies by capturing collective data on changes in the condition of industrial machinery, rather than the individual data of the industrial machinery, thereby detecting anomalies and suggesting the minimum necessary inspection timing. In other words, the anomaly detection device 20A determines whether the risk of machine failure is high or low based on the fluctuation pattern. If the risk is determined to be high, the anomaly detection device 20A detects an anomaly and prompts inspection of the machine tool 30. Furthermore, the anomaly detection device 20A can ignore transient changes that would be considered an anomaly based on a threshold, thereby reducing unnecessary inspections. Furthermore, the anomaly detection device 20A can evaluate risk by checking the fluctuation pattern when there is a change in the evaluation value. Furthermore, because the anomaly detection device 20A automatically determines whether to update the feature model based on the fluctuation pattern, there is no need to adjust the threshold, which avoids the problem of not knowing the appropriate threshold in the first place. The second embodiment has been described above.
[0049] From the above, the anomaly detection devices 20, 20A according to the first and second embodiments can detect anomalies in order to present the minimum necessary inspection timing by capturing data on changes in the condition of industrial machinery collectively, rather than the data on each individual point of industrial machinery.
[0050] <Modification 1> In the first and second embodiments, the anomaly detection devices 20 and 20A are illustrated as devices separate from the numerical control device 10. However, some or all of the functions of the anomaly detection devices 20 and 20A may be provided in the numerical control device 10. Alternatively, some or all of the feature amount acquisition unit 210, feature model operation unit 211, inference calculation unit 212, variation pattern management unit 213, variation pattern classification unit 214, feature amount anomaly determination unit 215, and feature model storage unit 216 of the anomaly detection device 20, or some or all of the feature amount acquisition unit 210, feature model operation unit 211a, inference calculation unit 212, variation pattern management unit 213, variation pattern classification unit 214, feature amount anomaly determination unit 215, feature model storage unit 216, and machine inspection result storage unit 217 of the anomaly detection device 20A may be provided in a server, for example. Furthermore, the functions of the anomaly detection devices 20 and 20A may be realized using a virtual server function or the like on a cloud. Furthermore, the anomaly detection devices 20 and 20A may be configured as a distributed processing system in which the functions of the anomaly detection devices 20 and 20A are appropriately distributed to a plurality of servers.
[0051] <Modification 2> For example, in the above-described embodiment, one numerical control device 10 is connected to one anomaly detection device 20, 20A, but this is not limiting. For example, one anomaly detection device 20, 20A may be connected to two or more numerical control devices 10. For example, an integrated anomaly detection system may have two or more anomaly detection devices 20, 20A, and each anomaly detection device 20, 20A may be connected to one or more numerical control devices 10. In this case, some or all of the functions of the variation pattern management unit 213 may be shared between the anomaly detection devices 20, 20A. This allows each anomaly detection device 20, 20A to detect anomalies using the same criteria.
[0052] <Modification 3> In the above-described embodiment, for example, anomaly detection device 20, 20A determines whether or not a feature value of machine tool 30 is abnormal from a classification result based on a comparison between the inference result and a predetermined fluctuation pattern stored in fluctuation pattern management unit 213. However, this is not limited to this. For example, anomaly detection device 20, 20A may store past fluctuation patterns in a storage unit (not shown) and perform risk assessment by comparing the classification result with the past fluctuation patterns. Alternatively, anomaly detection device 20, 20A may notify the equipment manager of the need for inspection when a fluctuation occurs, prompting the equipment manager to perform the inspection, and then evaluate the inspection results and risk by reviewing the fluctuation pattern several days later.
[0053] Note that the functions included in the anomaly detection devices 20 and 20A in the first and second embodiments can be realized by hardware, software, or a combination of these. Here, "realized by software" means that the functions are realized by a computer reading and executing a program.
[0054] The program can be stored and supplied to a computer using various types of non-transitory computer-readable media. Non-transitory computer-readable media include various types of tangible storage media. Examples of non-transitory computer-readable media include magnetic recording media (e.g., flexible disks, magnetic tapes, hard disk drives), magneto-optical recording media (e.g., magneto-optical disks), CD-ROMs (Read Only Memory), CD-Rs, CD-R / Ws, and semiconductor memories (e.g., mask ROMs, PROMs (Programmable ROMs), EPROMs (Erasable PROMs), flash ROMs, and RAMs). The program may be provided to the computer by various types of transient computer-readable media. Examples of transient computer-readable media include electrical signals, optical signals, and electromagnetic waves. The transient computer-readable media can provide the program to the computer via a wired communication path such as an electrical wire or optical fiber, or via a wireless communication path.
[0055] The step of writing a program to be recorded on a recording medium includes not only processes that are performed in chronological order, but also processes that are not necessarily performed in chronological order but are executed in parallel or individually. The step of writing the program may also be performed by cloud computing.
[0056] Although the present disclosure has been described in detail, the present disclosure is not limited to the individual embodiments described above. Various additions, substitutions, modifications, partial deletions, etc. are possible in these embodiments without departing from the gist of the present disclosure or the spirit of the present disclosure derived from the content of the claims and their equivalents. These embodiments can also be implemented in combination. For example, in the above-described embodiments, the order of each operation and the order of each process are shown as examples and are not limited to these. The same applies when numerical values or mathematical expressions are used in the description of the above-described embodiments.
[0057] The following supplementary note is further disclosed regarding the above embodiment and modified examples. (Supplementary note 1) The anomaly detection device (20) includes a feature quantity acquisition unit (210) that acquires a context in the production operation of the industrial machine (30) and a state quantity related to the production operation state of the industrial machine (30), and acquires feature quantities related to the production operation state of the industrial machine (30) based on the acquired context and state quantity, a feature model operation unit (211) that generates, updates, deletes, and adds feature models based on the context and the feature quantities, a feature model storage unit (216) that stores the feature models generated by the feature model operation unit (211), and a feature model storage unit (217) that stores the feature quantities and the feature models based on the feature quantities and the feature models. The abnormality detection device (20) according to claim 1, further comprising: an inference calculation unit (212) that calculates an evaluation value of the operating state of the industrial machine (30), a fluctuation pattern management unit (213) that stores and manages predetermined fluctuation patterns of the evaluation value, a fluctuation pattern classification unit (214) that classifies the fluctuation pattern of the evaluation value of the operating state of the industrial machine (30) based on the calculation result of the inference calculation unit (212) and the predetermined fluctuation pattern stored in the fluctuation pattern management unit (213), and a feature abnormality determination unit (215) that determines whether or not a feature value of the industrial machine (30) is abnormal based on the classification result of the fluctuation pattern classification unit (214). (Supplementary Note 2) In the abnormality detection device (20) according to Supplementary Note 1, the fluctuation pattern classification unit (214) classifies the evaluation value calculated by the inference calculation unit (212) into a fluctuation pattern in which the evaluation value at least continues to rise, a fluctuation pattern in which the evaluation value rises and remains high, a fluctuation pattern in which the evaluation value rises and remains high, and a fluctuation pattern in which the evaluation value rises and then settles to the original value. (Supplementary Note 3) The anomaly detection device (20A) of Supplementary Note 1 or Supplementary Note 2 further includes a machine inspection result storage unit (217) that accepts input of timely inspection results of the industrial machine (30), and the feature model operation unit (211 a) updates the feature model stored in the feature model storage unit (216) based on the determination result of the feature amount anomaly determination unit (215) and the inspection result of the machine inspection result storage unit (217). (Supplementary Note 4) In the anomaly detection device (20, 20A) of any of Supplementary Note 1 to Supplementary Note 3, the variation pattern management unit (213) is partly or entirely shared with other anomaly detection devices (20, 20A), and the variation pattern detected by the anomaly detection device (20, 20A) is used by the other anomaly detection devices (20, 20A).(Supplementary Note 5) The control device (10) is a control device that controls a motor (31) that drives a mechanism of an industrial machine (30), and includes any one of the abnormality detection devices (20, 20A) of Supplementary Note 1 to Supplementary Note 3. (Supplementary Note 6) The anomaly detection method is an anomaly detection method that causes a computer to operate as an anomaly detection device (20), and includes a feature amount acquisition step that acquires a context in production operation of the industrial machine (30) and a state amount related to the production operation state of the industrial machine (30), and acquires a feature amount related to the production operation state of the industrial machine (30) based on the acquired context and state amount; a feature model operation step that generates, updates, deletes, and adds a feature model based on the context and the feature amount; a feature model storage step that stores the feature model generated in the feature model operation step; an inference calculation step that calculates an evaluation value of the operating state of the industrial machine (30) based on the feature amount and the feature model; a fluctuation pattern management step that stores and manages a predetermined fluctuation pattern of the evaluation value; a fluctuation pattern classification step that classifies the fluctuation pattern of the evaluation value of the operating state of the industrial machine (30) based on a calculation result of the inference calculation step and the predetermined fluctuation pattern stored in the fluctuation pattern management step; and a feature amount abnormality determination step that determines whether or not the feature amount of the industrial machine (30) is abnormal based on the classification result of the fluctuation pattern classification step.(Supplementary Note 7) The program causes a computer to realize a feature acquisition step of acquiring a context in production operation of the industrial machine (30) and a state quantity related to the production operation state of the industrial machine (30), and acquiring a feature quantity related to the production operation state of the industrial machine (30) based on the acquired context and state quantity; a feature model operation step of generating, updating, deleting, and adding a feature model based on the context and the feature quantity; a feature model storage step of storing the feature model generated in the feature model operation step; an inference calculation step of calculating an evaluation value of the operating state of the industrial machine (30) based on the feature quantity and the feature model; a fluctuation pattern management step of storing and managing predetermined fluctuation patterns of the evaluation value; a fluctuation pattern classification step of classifying a fluctuation pattern of the evaluation value of the operating state of the industrial machine (30) based on a calculation result of the inference calculation step and the predetermined fluctuation pattern stored in the fluctuation pattern management step; and a feature abnormality determination step of determining whether or not a feature value of the industrial machine (30) is abnormal based on the classification result of the fluctuation pattern classification step. (Appendix 8) The integrated anomaly detection system includes two or more anomaly detection devices (20, 20A) of any one of Appendices 1 to 3, and the fluctuation pattern management unit (213) is shared in part or in whole between the anomaly detection devices (20, 20A), and the fluctuation pattern detected by one anomaly detection device (20, 20A) is used by the other anomaly detection devices (20, 20A).
[0058] REFERENCE SIGNS LIST 10 Numerical control device 110 Numerical control unit 111 Context acquisition unit 112 State quantity acquisition unit 20, 20A Abnormality detection device 210 Feature quantity acquisition unit 2101 State data extraction unit 2102 Feature quantity creation unit 211, 211a Feature model operation unit 212 Inference calculation unit 213 Fluctuation pattern management unit 214 Fluctuation pattern classification unit 215 Feature quantity abnormality determination unit 216 Feature model storage unit 217 Machine inspection result retention unit 30 Machine tool 31 Motor 32 Mechanism unit
Claims
1. A feature quantity acquisition unit acquires the context of the production operation of an industrial machine and state quantities related to the production operation state of the industrial machine, and acquires feature quantities related to the production operation state of the industrial machine based on the acquired context and state quantities. A feature model manipulation unit that generates, updates, deletes, and adds feature models based on the aforementioned context and the aforementioned features, A feature model storage unit that stores the feature model generated by the feature model operation unit, An inference calculation unit that calculates an evaluation value of the operating state of the industrial machine based on the aforementioned feature quantities and the aforementioned feature model, A variation pattern management unit that stores and manages predetermined variation patterns of the evaluation value, A variation pattern classification unit classifies the variation patterns of the evaluation value of the operating state of the industrial machine based on the calculation results of the inference calculation unit and the predetermined variation patterns stored in the variation pattern management unit, A feature anomaly determination unit that determines whether the feature quantities of the industrial machine are abnormal based on the classification results of the variation pattern classification unit, An anomaly detection device equipped with the following features.
2. The anomaly detection device according to claim 1, wherein the fluctuation pattern classification unit classifies the evaluation value calculated by the inference calculation unit into fluctuation patterns that continue to rise, fluctuation patterns that rise and remain high, fluctuation patterns that fluctuate up and down, and fluctuation patterns that rise and then settle back to the original value.
3. The machine inspection result holding unit further includes a unit that receives timely input of inspection results for the aforementioned industrial machine. An anomaly detection device according to claim 1 or 2, wherein the feature model operation unit updates the feature model stored in the feature model storage unit based on the determination result of the feature quantity anomaly determination unit and the inspection result of the machine inspection result holding unit.
4. The aforementioned fluctuation pattern management unit is partially or entirely shared with other anomaly detection devices. The abnormality detection device according to claim 1 or 2, wherein the fluctuation pattern detected by the abnormality detection device is used in the other abnormality detection device.
5. A control device for controlling a motor that drives the mechanism of an industrial machine, A control device comprising the abnormality detection device according to claim 1 or claim 2.
6. An anomaly detection method that operates a computer as an anomaly detection device, A feature acquisition step involves acquiring the context of the production operation of an industrial machine and state quantities related to the production operation state of the industrial machine, and acquiring feature quantities related to the production operation state of the industrial machine based on the acquired context and state quantities. A feature model manipulation process that generates, updates, deletes, and adds to a feature model based on the aforementioned context and the aforementioned features, A feature model storage step for storing the feature model generated in the feature model manipulation step, An inference calculation step for calculating an evaluation value of the operating state of the industrial machine based on the aforementioned feature quantities and the aforementioned feature model, A variation pattern management step for storing and managing a predetermined variation pattern of the evaluation value, A variation pattern classification step classifies the variation patterns of the evaluation value of the operating state of the industrial machine based on the calculation results of the inference calculation step and the predetermined variation patterns stored in the variation pattern management step, A feature anomaly determination step that determines whether the feature quantities of the industrial machine are abnormal or not based on the classification results of the variation pattern classification step, An anomaly detection method comprising the following features.
7. On the computer, A feature acquisition step involves acquiring the context of the production operation of an industrial machine and state quantities related to the production operation state of the industrial machine, and acquiring feature quantities related to the production operation state of the industrial machine based on the acquired context and state quantities. A feature model manipulation process that generates, updates, deletes, and adds to a feature model based on the aforementioned context and the aforementioned features, A feature model storage step for storing the feature model generated in the feature model manipulation step, An inference calculation step for calculating an evaluation value of the operating state of the industrial machine based on the aforementioned feature quantities and the aforementioned feature model, A variation pattern management step for storing and managing a predetermined variation pattern of the evaluation value, A variation pattern classification step classifies the variation patterns of the evaluation value of the operating state of the industrial machine based on the calculation results of the inference calculation step and the predetermined variation patterns stored in the variation pattern management step, A feature anomaly determination step that determines whether the feature quantities of the industrial machine are abnormal or not based on the classification results of the variation pattern classification step, A program that makes this possible.
8. The device comprises two or more abnormality detection devices according to claim 1 or claim 2, The aforementioned fluctuation pattern management unit is shared in part or in whole among the aforementioned anomaly detection devices. The fluctuation patterns detected by one anomaly detection device are used in an integrated anomaly detection system by other anomaly detection devices.