Anomaly detection device, control device, anomaly detection method, program and integrated anomaly detection system

Abstract

The purpose of the present invention is to collect data relating to changes in the state of an industrial machine in aggregate, instead of collecting each individual data point relating to the industrial machine, and thereby detect anomalies in order to indicate the minimum required inspection time. This anomaly detection device is equipped with: a feature parameter acquisition unit that captures the context of a production process of an industrial machine and a state parameter relating to the operating state of the industrial machine, and, based on the context and the state parameter, captures a feature parameter relating to the operating state of the industrial machine; a feature model operating unit that creates, updates, deletes, and adds feature models using the context and the feature parameter; and a feature model storage unit that stores the feature models.a conclusion calculation unit that calculates an evaluation value of the operating state of the industrial machine using the characteristic variable and a characteristic model; a fluctuation pattern management unit that stores predefined 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 result of the conclusion calculation and the predefined fluctuation patterns; and a characteristic variable anomaly determination unit that determines, based on the classification result, whether the characteristic variable of the industrial machine is anomalous or not.

Description

TECHNICAL AREA

[0001] The present disclosure relates to an anomaly detection device that detects an anomaly in a feature size of an industrial machine based on a fluctuation pattern of the feature size, a control device, an anomaly detection method, a program and an integrated anomaly detection system. BACKGROUND OF THE TECHNOLOGY

[0002] The failure of an industrial machine, such as a machine tool or robot, leads to a reduction in the production rate at a production site. To prevent this reduction in production rate and the occurrence of defective workpieces, preventive maintenance of the industrial machine is necessary.

[0003] For this reason, a technique has been proposed for preventive maintenance in which information (control signal, current value, vibration value, etc.) is collected during operation and a feature metric is modeled using machine learning to detect anomalies. See, for example, patent documents 1 to 3. These previous techniques employ a method in which a threshold is set, an evaluation value exceeding the threshold is identified as an anomaly, and such an anomaly is reported. Citation list of patent specifications Patent specification 1: Japanese unexamined patent application, publication no. 2020-123191 Patent specification 2: Japanese unexamined patent application, publication no. 2022-128824 Patent Specification 3: International PCT Publication No. WO2023 / 007825 DISCLOSURE OF THE INVENTION Problems to be solved by the invention

[0004] However, measurement data show that a change in the behavior of an industrial machine, such as a machine tool, is not so simple as to be divided into two categories based on a threshold value. For example, if a change in an evaluation value is observed over time, the evaluation value may change gradually or in upward and downward movements.

[0005] Under such circumstances, a previous procedure, where an anomaly is reported based on an evaluation value exceeding a predefined threshold, frequently triggers notifications, preventing the operator from making a critical decision. Even adjusting the threshold does not fundamentally solve the difficulty of distinguishing between normality and anomalies. The measurement data also indicate that the industrial machine's behavior is individual.

[0006] For this reason, it is difficult to create versatile models and thresholds from extensive data in advance and to apply these models and thresholds.

[0007] A general mechanism for the failure of an industrial machine, such as a machine tool, is not known. Therefore, the required function of an anomaly detection device is to reconcile anomalies in the data with anomalies in the industrial machine as closely as possible.

[0008] Therefore, it was demanded that not every single data element of an industrial machine be recorded, but that data on a change in the state of the industrial machine as a whole be recorded so that anomaly detection can be carried out in order to reduce the time required for an inspection to a minimum. Means to solve the problems (1) An aspect of an anomaly detection device of the present disclosure comprises a feature size detection unit that detects a context in the production operation of an industrial machine and a state size that indicates the operating state of the production operation of the industrial machine, and, based on the detected context and state size, detects a feature size associated with the operating state of the production operation of the industrial machine; a feature model operating unit that generates, updates, deletes, or adds a feature model based on the context and the feature size; a feature model storage unit that stores the feature model generated by the feature model operating unit; a conclusion calculation unit that calculates an evaluation value of the operating state of the industrial machine based on the feature size and the feature model; and a fluctuation pattern management unit.which stores and manages a predetermined fluctuation pattern 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 a calculation result of the conclusion calculation unit and the predetermined fluctuation pattern stored in the fluctuation pattern management unit, and a feature size anomaly determination unit that determines, based on a classification result of the fluctuation pattern classification unit, whether the feature size of the industrial machine is anomalous or not. (2) One aspect of a control device of the present disclosure is a control device for controlling a motor which drives a mechanism of an industrial machine which includes the anomaly detection device of (1). (3) An aspect of an anomaly detection method of the present disclosure is an anomaly detection method that causes a computer to operate as an anomaly detection device, comprising a feature size acquisition step for acquiring a context in the production operation of an industrial machine and a state size indicating the state of the production operation of the industrial machine, and, based on the acquired context and the state size for acquiring a feature size associated with the state of the production operation of the industrial machine, a feature model operation step for generating, updating, deleting, or adding a feature model based on the context and the feature size, and a feature model storage step for storing the feature model generated by the feature model operation step.The process includes a conclusion calculation step for calculating an evaluation value of the operating state of the industrial machine based on the characteristic variable and the characteristic model, a fluctuation pattern management step for storing and managing a predetermined fluctuation pattern of the evaluation value, a fluctuation pattern classification step for classifying the fluctuation pattern of the evaluation value of the operating state of the industrial machine based on a calculation result of the conclusion calculation step and the predetermined fluctuation pattern stored in the fluctuation pattern management step, and a characteristic variable anomaly determination step for determining, based on a classification result of the fluctuation pattern classification step, whether the characteristic variable of the industrial machine is anomalous or not. (4) One aspect of a program of the present disclosure is a program that causes a computer to perform a feature size acquisition step to acquire a context in the production operation of an industrial machine and a state size indicating the state of the production operation of the industrial machine, and, based on the acquired context and state size, to acquire a feature size associated with the state of the production operation of the industrial machine; a feature model operation step to generate, update, delete, or add a feature model based on the context and the feature size; a feature model storage step to store the feature model generated by the feature model operation step; and a conclusion calculation step to calculate an evaluation value of the operating state of the industrial machine based on the feature size and the feature model.to implement a fluctuation pattern management step for storing and managing a predetermined fluctuation pattern of the evaluation value, a fluctuation pattern classification step for classifying the fluctuation pattern of the evaluation value of the operating state of the industrial machine based on a calculation result of the conclusion calculation step and the predetermined fluctuation pattern stored in the fluctuation pattern management step, and a feature size anomaly determination step for determining whether the feature size of the industrial machine is anomalous, based on a classification result of the fluctuation pattern classification step. (5) An aspect of an integrated anomaly detection system of the present disclosure comprises the anomaly detection device of (1) comprising two or more anomaly detection devices, wherein some or all of the fluctuation pattern management units are common between the anomaly detection devices and a fluctuation pattern detected by one of the anomaly detection devices is used in the other anomaly detection devices. BRIEF DESCRIPTION OF THE DRAWINGS Fig. Figure 1 is a diagram showing an example of a hardware configuration in a case where an anomaly detection device of a first embodiment and a numerical control device are connected; Fig. Figure 2 is a schematic functional block diagram in a case where the anomaly detection device according to a first embodiment and the numerical control device are connected to each other; Fig. Figure 3 is a diagram showing an example of time series data on the rotational speed and torque of a spindle motor attached to a tool of a machine tool in a non-cutting state, as a state variable captured by a feature size acquisition unit; Fig. Figure 4 is a diagram that shows an example to describe the operation of a feature variable calculation unit; Fig. Figure 5 is a diagram that shows an example of a relationship between a feature model and a feature size; Fig. Figure 6 is a diagram that shows an example of the relationship between the feature model and the feature size; Fig. 7A is a diagram showing an example of a predetermined fluctuation pattern; Fig. 7B is a diagram showing an example of the predetermined fluctuation pattern; Fig. 7C is a diagram showing an example of the predetermined fluctuation pattern; Fig. 7D is a diagram that shows an example of the predetermined fluctuation pattern; Fig. Figure 8 is a diagram showing an example of time series data for an evaluation value calculated by a conclusion calculation unit; Fig. Figure 9 is a flowchart describing the anomaly detection process of the anomaly detection device; Fig. Figure 10 is a schematic functional block diagram for the case in which an anomaly detection device according to a second embodiment and a numerical control device are connected to each other; Fig. Figure 11A is a diagram showing an example of the selection of data to be used for machine learning in the case of a “back and forth” pattern: Fig. Figure 11B is a diagram showing an example of the selection of data to be used for machine learning in the case of a “plateau pattern”; and Fig. Figure 12 is a flowchart describing the processing of the anomaly detection by the anomaly detection device. PREFERRED METHOD FOR IMPLEMENTING THE INVENTION<Erste Ausführungsform>

[0009] The following section describes in detail an anomaly detection device according to a first embodiment with reference to the figures. As an example, a case is described in which a machine tool, which is a type of industrial machine, is connected as the anomaly detection target. It should be noted that 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.

[0010] Fig. Figure 1 is a diagram showing an example of a hardware configuration in a case where the anomaly detection device of the first embodiment and a numerical control device are connected.

[0011] As in Fig. As shown in Figure 1, an anomaly detection device 20 is connected to a numerical control device 10.

[0012] The numerical control device 10 and the anomaly detection device 20 are directly connected to each other via a connection interface (not shown). It should be noted that the numerical control device 10 and the anomaly detection device 20 can also be connected and communicate with each other via a network (not shown), such as a local area network (LAN) or the internet.

[0013] The numerical control device 10 and the anomaly detection device 20 are different devices, but as described later, the numerical control device 10 can include the anomaly detection device 20.

[0014] The numerical control device 10 is, for example, a numerical control device known to those skilled in the art for controlling the operation of the machine tool (not shown). It should be noted that if the machine tool (not shown) is a robot or the like, the numerical control device 10 could be a robot control device or the like.

[0015] The numerical control device 10 has 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. In addition, the numerical control device 10 is connected to a display 130.

[0016] The CPU 11 is a processor that controls the numerical control device 10 as a whole. The CPU 11 reads a system program stored in the ROM 12 via the bus 17 and controls the entire numerical control device 10 according to the system program. The RAM 13 temporarily stores computational and display data, various types of data entered by an operator via an input device (not shown), such as a keyboard or touch panel, and the like.

[0017] To control an axis of the machine tool (not shown), the shaft control circuit 14 receives a shaft movement command from the CPU 11 via bus 17 and outputs this command to the servo amplifier 15. Upon receiving the command, the servo amplifier 15 drives a motor 120, which moves a shaft of the machine tool (not shown). The motor 120 for the shaft includes a position and velocity detector (not shown), provides feedback to the shaft control circuit 14 via a position and velocity feedback signal from the position and velocity detector, and performs position and velocity feedback control. Note that in the representation of the hardware configuration diagram in Fig. 1 only one group consisting of the axis control circuit 14, the servo amplifier 15 and the motor 120 is present, but in fact a number of groups corresponding to the number of shafts provided for the machine tool (not shown) is present.

[0018] The INTs 16, 24 are interfaces for connecting the numerical control device 10 and the anomaly detection device 20 to each other.

[0019] The display 130, for example, is a liquid crystal display or the like and is used to operate the machine tool (not shown), which includes the numerical control device 10 and the anomaly detection device 20 (described later). The display 130 can show a result of the anomaly detection or the like for the machine tool (not shown), which is obtained from the anomaly detection device 20.

[0020] The anomaly detection device 20 is, for example, a computer, such as a personal computer, a fog computer, or a cloud server, and has a processor 21, a ROM 22, a RAM 23, an INT 24, and a bus 25. In addition, the anomaly detection device 20 is connected to a display 230.

[0021] The processor 21, for example, is a CPU or similar device and reads a system program stored in ROM 22 and controls the entire anomaly detection device 20 according to the system program. RAM 23 temporarily stores computational and display data, various types of data entered by the operator via an input device (not shown), such as a keyboard or touch panel, and the like.

[0022] The INT 24 transmits various types of data to the numerical control device 10 and receives various types of data from the numerical control device 10 via the INT 16 of the numerical control device 10.

[0023] The display 230 is a liquid crystal display or the like and shows a result of the processing and the like performed by the anomaly detection device 20.

[0024] Fig. Figure 2 is a schematic functional block diagram for the case in which the anomaly detection device 20 according to the first embodiment is connected to a machine tool 30, which represents a type of industrial machine. Each in Fig. The functional block shown in Figure 2 is implemented such that the CPU 11 of the numerical control device 10 or the processor 21 of the anomaly detection device 20, as shown in Figure 2, can be used by the CPU 11 of the numerical control device 10 or the processor 21 of the anomaly detection device 20. Fig. Figure 1 shows how the operation of each unit of the device is controlled according to the system program.

[0025] In particular, the numerical control device comprises 10 in Fig. 2 For example, a numerical control unit 110, a context acquisition unit 111, and a state variable acquisition unit 112. The anomaly detection device 20 comprises a feature variable acquisition unit 210, a feature model operating unit 211, a conclusion calculation unit 212, a fluctuation pattern management unit 213, a fluctuation pattern classification unit 214, a feature variable anomaly determination unit 215, and a feature model storage unit 216. The feature variable acquisition unit 210 comprises a state data extraction unit 2101 and a feature variable calculation unit 2102.

[0026] 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 reads and analyzes a block of the machining program stored in the memory (not shown) sequentially, calculates the movement of a motor 31 per control period based on the machining program analysis result, and controls the motor 31 according to the calculated movement per control period. By driving the motor 31, the machine tool 30 drives a mechanism 32, which, for example, includes a ball screw used as a feed shaft and a mechanism used as a spindle in the machine tool 30. In this way, the machine tool 30 machines a workpiece. It should be noted that in Fig. 2. A motor 31 is connected to a mechanism 32 of the machine tool 30; however, the number of motors 31 can be arranged according to the number of shafts of the mechanism 32 of the machine tool 30. A single mechanism 32 can be driven by a plurality of motors 31.

[0027] The context acquisition unit 111 acquires contexts (production status, operating status, environmental status, and the like) in the production operation performed by the numerical control device 110 (and the machine tool 30 controlled by the numerical control device 110) and outputs the acquired contexts to the anomaly detection device 20. The contexts in the production operation include, for example, an operating pattern (number of spindle revolutions, feed rate, or the like) of the motor 31 in the production operation, a purpose of the machining currently being performed (roughing, finishing, or the like), a purpose of the driving action currently being performed on a moving unit (rapid positioning, cutting feed, or the like), the type of tool used for machining, workpiece information indicating the hardness, material, and the like of the workpiece being machined, and the like.The context acquisition unit 111 comprehensively captures the contexts in the production operation based on a processing state specified by the processing program, information transmitted to the context acquisition unit 111 by the operator via the input device (not shown), information transmitted to the numerical control device 10 from another computer connected via a network (not shown) or the like, information captured by a device separately provided for the numerical control unit 110, such as a sensor, the value of a signal captured by a programmable logic controller (PLC) (not shown), or the like. The context acquisition unit 111 outputs the captured contexts to the feature size acquisition unit 210 and the feature model operating unit 211 of the anomaly detection device 20 (described later).

[0028] The state variable acquisition unit 112 acquires a state variable indicating the state of the production process being carried out by the numerical control unit 110 (and the machine tool 30 controlled by the numerical control unit 110). Examples of state variables indicating the state of the production process include the spindle load (current or torque), the feed shaft load (current or torque), the spindle speed, the feed shaft speed, the feed shaft position, a position feedback value, the temperature, the vibration value, and the noise of the motor 31, and the like. Furthermore, the state variable acquisition unit 112 acquires, for example, the rotational speed of the motor 31 driving the mechanism 32 of the machine tool 30 controlled by the numerical control unit 110, as well as the actual waveform of the torque value.The state variable acquisition unit 112 outputs the acquired state variable to the feature variable acquisition unit 210 of the anomaly detection device 20 (described later).

[0029] The feature size acquisition unit 210 acquires the contexts in the production operation of the machine tool 30 and the state size that indicates the state of the production operation of the machine tool 30, and acquires a feature size that is associated with the state of the production operation of the machine tool 30, based on the acquired contexts and the state size.

[0030] In particular, the feature size acquisition unit 210 includes, for example, the state data extraction unit 2101 and the feature size calculation unit 2102.

[0031] The state data extraction unit 2101 extracts state data from the state variable acquired by the state variable acquisition unit 112. This data is intended for use in the inference processing performed by the inference calculation unit 212 (described later), the processing of feature model updates by the feature model operating unit 211 (described later), or similar operations. The state data extraction unit 2101 extracts the state data to be used for inference processing or similar operations from the state variable acquired by the state variable acquisition unit 112 according to a predetermined extraction pattern based on the production operation contexts input by the context acquisition unit 111.The extraction pattern used by the State Data Extraction Unit 2101 is a predetermined data processing procedure in which a parameter is determined based on the contexts and includes, for example, data processing operations such as setting the interval for extracting time-series data obtained based on the contexts, selecting such data, and changing the scale of the state variable based on the contexts. The extraction pattern used by the State Data Extraction Unit 2101 can be pre-registered in memory by the operator.

[0032] The following describes the extraction of state data from the state variable according to the predetermined extraction pattern based on the contexts in the production operation by the state data extraction unit 2101.

[0033] Fig. Figure 3 is a diagram that, as a state variable acquired by the state variable acquisition unit 112, shows an example of time series data on the rotational speed and torque of a spindle motor attached to a tool of the machine tool 30 in a non-cutting state, in which no machining is performed. Fig. Figure 3 shows the time-series data for the spindle motor's speed and torque on the left and the torque state data extracted by the 2101 state data extraction unit on the right. It should be noted that the spindle motor's speed and torque time-series data, for example, are obtained by sampling at a sampling frequency of 0.001 seconds, and that they represent time-series data at a predetermined point in the spindle motor's operation, i.e., when the spindle motor is rotating at a constant speed of approximately 4000 rpm but is idling.

[0034] The 2101 state data extraction unit extracts a torque value as state data within a specific interval of the torque time series data, for example, an interval (e.g., 0.1 seconds) indicated by a dashed rectangle, during which the spindle motor rotates at a constant speed of approximately 4000 rpm but idles without any external force being applied. As shown on the right side in Fig. As shown in Figure 3, the torque representing the extracted state data does not show a constant value, as shown on the left side in Figure 3. Fig. 3 is shown, but it shows a minimal oscillation. The state data extraction unit 2101 extracts, for example, the torque as state data using the in Fig. The three tiny oscillations shown on the right are derived from a multitude of time series data on the speed and torque of the spindle motor when the spindle motor is rotating at a constant speed (e.g., about 4000 rpm or the like) but is idling, as in Fig. 3 shown on the left.

[0035] The state data extraction unit 2101 outputs the extracted state data to the feature size calculation unit 2102.

[0036] Based on the state data extracted by the state data extraction unit 2101, the feature size calculation unit 2102 calculates the feature size that indicates the characteristic of the operating state of the machine tool 30 controlled by the numerical control unit 110. The feature size calculated by the feature size calculation unit 2102 specifies the property of the operating state of the machine tool 30 and is useful information in the production process performed by the numerical control unit 110 (and the machine tool 30) for determining when an anomaly in the operating state of the machine tool 30 is detected. Furthermore, the feature size calculated by the feature size calculation unit 2102 is an input file when the conclusion calculation unit 212 (described later) performs a conclusion using the feature model.

[0037] For example, in Fig. As shown in Figure 4, the feature size calculation unit 2102 calculates the feature size by performing a Fourier transform for each frequency range of the state data (torque) extracted by the state data extraction unit 2101, standardizing the amplitude or power density of the state data, or calculating the statistics (average, dispersion, or the like) of the state data.

[0038] Based on the contexts in the production plant, which are entered by the context acquisition unit 111, and the feature size, which specifies the feature of the operating state of the machine tool 30 and is calculated by the feature size calculation unit 2102, the feature model operating unit 211 generates, updates, deletes or supplements the feature model stored in the feature model storage unit 216 (described later).

[0039] The feature model operating unit 211 selects the feature model to be generated, updated, deleted, or added based on the contexts in the production plant input by the context acquisition unit 111, and performs machine learning on the selected feature model using the feature metric calculated by the feature metric calculation unit 2102, which specifies the property of the state of the production plant. For example, if a feature model associated with the contexts (or combinations thereof) in the production process input by the context acquisition unit 111 is not stored in the feature model storage unit 216, the feature model operating unit 211 creates a new temporary standard (feature model) associated with the contexts (or combinations thereof), for example, from the frequency distribution of the minute vibrations of 50 state data points that are in Fig. 4 are shown.

[0040] If the feature model associated with the contexts (their combination) in the production process, which are input by the context capture unit 111, is stored in the feature model storage unit 216, and the state of the feature quantity, which is calculated by the feature quantity calculation unit 2102 and indicates the feature of the state of the production process, is closer to the range of such a feature model than by a predetermined distance, the feature model operating unit 211 can perform machine learning on the feature model including the feature quantity calculated by the feature quantity calculation unit 2102 and update the feature model.

[0041] For example, in Fig. As shown in Figure 5, in a case where a feature model A, which is associated with the contexts (their combination) in the production process, which are input by the context capture unit 111, is stored in the feature model storage unit 216, and a state B of the feature size, which was calculated by the feature size calculation unit 2102, is a predetermined distance or more away from the area of ​​feature model A, the feature model operating unit 211 can perform machine learning using the feature size of state B, including the feature size calculated by the feature size calculation unit 2102, and generate and add a new feature model B.

[0042] For example, in Fig. As shown in Figure 6, in a case where feature models A, B, which are associated with the contexts (their combination) in the production plant and are input by the context acquisition unit 111, are stored in the feature model storage unit 216, and a state C, which was calculated by the feature size calculation unit 2102, lies in the middle between the areas of the feature models A, B, the feature model operating unit 211 can perform machine learning using the feature sizes in the areas of the feature models A, B and the feature size of the state C including the feature size calculated by the feature size calculation unit 2102, and generate a new feature model A'.The feature model operating unit 211 can delete feature models A, B from feature model storage unit 216, update the feature model to the newly generated feature model A' and store feature model A' in feature model storage unit 216.

[0043] It should be noted that a variational autoencoder (VAE), generative adversarial networks (GAN), or similar machine learning can be used in the feature model operating unit 211. Any method can be applied as long as an evaluation value for the operating state can be calculated in the inference computation unit 212 (described later).

[0044] Based on the feature model selected from the feature model storage unit 216 based on the contexts in the production operation of the machine tool 30, which are entered by the context acquisition unit 111, and the feature size calculated by the feature size calculation unit 2102, the conclusion calculation unit 212 calculates the evaluation value of the operating state of the machine tool 30, which is executed by the numerical control unit 110 (and the machine tool 30).

[0045] In particular, the inference calculation unit 212, for example, inputs the new characteristic value calculated by the characteristic value calculation unit 2102, which indicates the characteristic of the state of the production process, into the characteristic model (provisional standard) that was selected on the basis of the contexts in the production operation of the machine tool 30, which are entered by the context acquisition unit 111, and thereby calculates the evaluation value (degree of data anomaly) that indicates how similar the new characteristic value is to the characteristic model (provisional standard).The evaluation value obtained as a result of the calculation (inferred) by the inference calculation unit 212 can be, for example, a condition such as the classification of the normality / anomaly of the operating state of the machine tool 30, information indicating a location where the operating state of the machine tool 30 is anomalous (anomaly of a bearing of the motor 31, damage to a coupling section between the motor 31 and the mechanism 32, or the like), or a difference between the distribution of the current operating state of the machine tool 30 and the distribution of the normal operating state of the machine tool 30. The inference calculation unit 212 stores the calculated evaluation value in a memory unit (not shown).

[0046] The fluctuation pattern management unit 213 stores and manages a predetermined fluctuation pattern of the evaluation value.

[0047] In particular, the fluctuation pattern management unit 213, for example, stores four fluctuation patterns of the evaluation value, which are in the Fig. 7A to 7D are shown, in advance and these patterns are managed.

[0048] It should be noted that in the Fig. In sections 7A to 7D, the vertical axis shows the evaluation value and the horizontal axis shows the date. The fluctuation pattern of the evaluation value in Fig. 7A is a fluctuation pattern in which the evaluation value initially rises and then falls back to an original value, and is also referred to as a "circular pattern".

[0049] If the evaluation value exhibits this fluctuation pattern, it returns to its original state, and therefore the feature size anomaly determination unit 215 (described later) determines that machine tool 30 is normal. It should be noted that a dashed line indicating an evaluation value of "8.00" and a dotted line indicating an evaluation value of "6.00" are examples of a previous threshold and indicate that the threshold does not necessarily reflect the normality / anomaly of machine tool 30.

[0050] The fluctuation pattern of the evaluation value in Fig. 7B is a fluctuation pattern in which the evaluation value repeatedly rises and falls, and is also referred to as a "back-and-forth pattern". When the evaluation value exhibits this fluctuation pattern, it alternates between two states (i.e., a low evaluation value state and a high evaluation value state), and therefore the feature size anomaly determination unit 215 (described later) determines that the machine tool 30 is anomalous.

[0051] The fluctuation pattern of the evaluation value in Fig. 7C is a fluctuation pattern in which the evaluation value increases and remains high, and is also referred to as a "plateau pattern". When the evaluation value exhibits this fluctuation pattern, it is assumed that the evaluation value has transitioned to a new state, and the feature size anomaly detection unit 215 (described later) determines that the machine tool 30 is anomalous.

[0052] The fluctuation pattern of the evaluation value in Fig. 7D is a fluctuation pattern in which the evaluation value continuously increases, and is also referred to as a "rise pattern". When the evaluation value exhibits this fluctuation pattern, the state is continuously changing, and therefore the feature size anomaly determination unit 215 (which is described later) determines that the machine tool 30 has an anomaly.

[0053] It should be noted that the fluctuation pattern management unit 213 stores the four fluctuation patterns, but can also store, for example, three or five fluctuation patterns.

[0054] Based on the calculation result of the conclusion calculation unit 212 and the predetermined fluctuation patterns stored in the fluctuation pattern management unit 213, the fluctuation pattern classification unit 214 classifies the fluctuation pattern of the evaluation value of the operating state of the machine tool 30.

[0055] In particular, the fluctuation pattern classification unit 214 reads evaluation values ​​from a data storage unit (not shown) for example from three to ten days as a calculation result of the conclusion calculation unit 212.

[0056] Fig. Figure 8 shows an example of time series data for the evaluation value calculated by the inference calculation unit 212.

[0057] In particular, the fluctuation pattern classification unit 214 performs a comparison with the fluctuation patterns stored in the fluctuation pattern management unit 213. Fig. 7A to 7D and, for example, classifies the fluctuation pattern of the evaluation value over a period (1) as a "circular pattern". Alternatively, the fluctuation pattern classification unit 214 performs a comparison with the fluctuation patterns stored in the fluctuation pattern management unit 213. Fig. 7A to 7D and classifies the fluctuation pattern of the evaluation value over a period (2) as a "plateau pattern". The fluctuation pattern classification unit 214 outputs the classification result to the feature size anomaly determination unit 215.

[0058] Based on the classification result of the fluctuation pattern classification unit 214, the feature size anomaly determination unit 215 determines whether the feature size of the machine tool 30 is anomalous or not.

[0059] Specifically, the feature size anomaly determination unit 215 determines, for example, in a case where the fluctuation pattern of the evaluation value is the “circular pattern”. Fig. 7A, based on the risk assessment associated with the fluctuation pattern, indicates that the feature size of machine tool 30 is normal, as the evaluation value returns to its original state. If the fluctuation pattern of the evaluation value is the "back and forth pattern" from Fig. 7B, the feature size anomaly determination unit 215 determines, based on the risk assessment associated with the fluctuation pattern, that the feature size of the machine tool 30 is anomalous, since the evaluation value alternates between the two states. If the fluctuation pattern of the evaluation value is the "plateau pattern" from Fig. 7C determines, based on the risk assessment associated with the fluctuation pattern, that the feature size of the machine tool 30 is anomalous because the evaluation value has transitioned to the new state. If the fluctuation pattern of the evaluation value is the "rise pattern" from Fig. 7D determines, based on the risk assessment associated with the fluctuation pattern, that the feature size of the machine tool 30 is anomalous because the condition is continuously changing.

[0060] If the feature size of machine tool 30 is found to be anomalous, the feature size anomaly detection unit 215 can display a message to a plant manager on the display 130 of the numerical control device 10 and / or the display 230 of the anomaly detection device 20 indicating that an inspection of machine tool 30 is required. The plant manager then performs the inspection of machine tool 30 based on the indicated inspection requirement. In this way, preventive maintenance can be carried out.

[0061] The feature model storage unit 216 is, for example, a solid-state drive (SSD), a hard disk drive (HDD), or the like, and stores a multitude of feature models associated with combinations of contexts in the production operation and entered by the acquisition unit 111. It should be noted that the feature model storage unit 216 can be located in the external numerical control device 10, the cell computer, the fog computer, the cloud server, the database server, or the like.

[0062] The feature model storage unit 216 stores a multitude of feature models 1, 2, ..., n, which are associated with combinations of contexts (machining status, operating status, environmental status, and the like) in the production operation, specified by the context acquisition unit 111. Here, the combination of contexts (machining status, operating status, environmental status, and the like) in the production operation means a combination of a possible value for each context, a value range, and a value list. For example, if the combination of contexts includes the spindle speed, the feed rate, a cutting signal, the tool type, and the workpiece information, such a combination might be, for example, "Spindle speed: 500 to 1000 [rpm], feed rate: 200 to 300 [mm / min], cutting operation in progress, drilling tool, aluminum / ferrous steel".

[0063] The feature model stored in feature model storage unit 216 is created by feature model operating unit 211 and stored as information for forming a feature model that can be used for inference processing in inference processing unit 212. For example, the feature model stored in feature model storage unit 216 can be a model that uses a learning algorithm from a multilayer neural network such as VAE or GAN, or it can be a model that uses a learning algorithm from a Bayesian network. The feature models stored in feature model storage unit 216 can be feature models that use the same learning algorithm, feature models that use different learning algorithms, or feature models that use any learning algorithm suitable for inference processing in inference processing unit 212.

[0064] The feature model storage unit 216 can store one feature model in conjunction with a combination of contexts in the production operation, or two or more feature models in conjunction with a combination of contexts in the production operation. The feature model storage unit 216 can store a variety of feature models in conjunction with a combination of contexts in the production operation using the same learning algorithm, or feature models in conjunction with a combination of contexts in the production operation using two or more different learning algorithms. The feature model storage unit 216 can store feature models in conjunction with multiple combinations of contexts in the production operation that have an overlapping area, using different learning algorithms.In this case, the feature model storage unit 216 can additionally store a usage condition, such as required processing capacity or the type of learning algorithm, for the feature model corresponding to the combination of contexts in the production process. This allows the inference calculation unit 212, for example, to select the feature model for the combination of contexts in the production process based on the available inference processing or processing capacity.

[0065] It should be noted that the feature model storage unit 216 can encrypt the feature model generated by the feature model operating unit 211, store the encrypted feature model, and decrypt the encrypted feature model when the feature model is read by the inference calculation unit 212. <Anomalienerfassungsverarbeitung der Anomalienerfassungsvorrichtung 20>

[0066] Next, the anomaly detection processing sequence of the anomaly detection device 20 will be described with reference to Fig. 9 described.

[0067] Fig. Figure 9 is a flowchart describing the anomaly detection processing of the anomaly detection device 20. The sequence described here is executed each time the numerical control device 10 instructs the machine tool 30 to perform the machining operation.

[0068] In step S11, the feature size acquisition unit 210 captures the contexts in the production process of the machine tool 30 and the state size that indicates the state of the production process of the machine tool 30, and captures the feature size that is associated with the state of the production process of the machine tool 30, based on the captured contexts and the state size.

[0069] In step S12, the inference calculation unit 212 determines whether the feature model storage unit 216 stores the feature model that corresponds to the contexts in the production process of machine tool 30. If the feature model storage unit 216 stores the feature model that corresponds to the contexts in the production process of machine tool 30, processing continues with step S14. Conversely, if the feature model storage unit 216 does not store the feature model that corresponds to the contexts in the production process of machine tool 30, processing continues with step S13.

[0070] In step S13, the feature model operating unit 211 generates the new feature model for the contexts recorded in step S11 in the production plant, based on these contexts and the feature variable that indicates the operating state of the machine tool 30. The feature model operating unit 211 stores the generated feature model in the feature model storage unit 216.

[0071] In step S14, the inference calculation unit 212 performs a inference (calculation) regarding the evaluation value of the operating state executed by the numerical control unit 110 (and the machine tool 30) of the machine tool 30, based on the feature model selected from the feature model storage unit 216 and the feature size recorded in step S11.

[0072] 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 conclusion result of step S14 and the predetermined fluctuation pattern stored in the fluctuation pattern management unit 213.

[0073] In step S16, the feature size anomaly detection unit 215 determines, based on the classification result from step S15, whether the feature size of the machine tool 30 is anomalous or not. If the feature size of the machine tool 30 is determined to be anomalous, processing continues with step S17. Conversely, if the feature size of the machine tool 30 is determined to be normal, the anomaly detection device 20 terminates the anomaly detection.

[0074] In step S17, the feature size anomaly detection unit 215 indicates the need for an inspection of the machine tool 30.

[0075] With the configuration described above, the anomaly detection device 20, according to the first embodiment, does not record each individual data element on the machine tool 30, but rather records the data on changes in the state of the machine tool 30 as a whole, so that anomaly detection can be performed to minimize the time required for inspection. In other words, the anomaly detection device 20 determines, based on the fluctuation pattern, whether the risk of machine failure is high or low, and, if the risk is determined to be high, it records such a risk as an anomaly and requests an inspection of the machine tool 30.

[0076] The anomaly detection device 20 can ignore a temporary change that might be considered an anomaly when detected based on a threshold, thereby reducing unnecessary inspections.

[0077] If the evaluation value changes, the anomaly detection device 20 can assess the risk by checking the fluctuation pattern.

[0078] The first embodiment was described above. <Zweite Ausführungsform>

[0079] Next, a second embodiment is described. In the first embodiment, the anomaly detection device 20 determines, based on the classification result of the fluctuation pattern of the evaluation value of the operating state, which is executed by the numerical control device 110 (and the machine tool 30), whether the feature size of the operating state of the machine tool 30 is anomalous or not, and, in a case where it is determined that the feature size of the machine tool 30 is anomalous, indicates the need for an inspection of the machine tool 30.On the other hand, the second embodiment differs from the first embodiment in that, when it is determined that the feature size of the machine tool 30 is anomalous, an anomaly detection device 20A receives a result of the inspection of the machine tool 30, which is entered from a maintenance management protocol or by the plant manager, in response to the output of the need to inspect the machine tool 30, and updates the feature model based on the determination result and the inspection result.

[0080] In this configuration, according to the second embodiment, the anomaly detection device 20A does not capture every single data element on the industrial machine, but rather captures data on changes in the state of the industrial machine as a whole. This allows anomaly detection to be performed to the greatest extent possible to determine the time of inspection. Furthermore, the anomaly detection device 20A updates the feature model, thus further reducing the number of inspections without detecting the same anomaly multiple times.

[0081] The second embodiment is described below.

[0082] The hardware configuration of the anomaly detection device 20A according to the second embodiment is similar to that in a case of the first embodiment of Fig. 1, and their description is therefore omitted.

[0083] Fig. Figure 10 is a schematic functional block diagram for the case where the anomaly detection device 20A according to the second embodiment is connected to a numerical control device 10. It should be noted that the same reference numerals are used to denote elements that have similar functions to the elements of the anomaly detection device 20 in Figure 10. Fig. 2, and a detailed description is omitted.

[0084] As in Fig. As shown in Figure 10, the anomaly detection device 20A according to the second embodiment is connected to the numerical control device 10.

[0085] The numerical control device 10 has a similar function to the numerical control device 10 in the first embodiment.

[0086] A numerical control unit 110, a context acquisition unit 111 and a state variable acquisition unit 112 have similar functions to the numerical control unit 110, the context acquisition unit 111 and the state variable acquisition unit 112 in the first embodiment.

[0087] The anomaly detection device 20A comprises a feature size acquisition unit 210, a feature model operating unit 211a, a conclusion calculation unit 212, a variation pattern management unit 213, a variation pattern classification unit 214, a feature size anomaly determination unit 215, a feature model storage unit 216, and a machine inspection result holding unit 217. The feature size acquisition unit 210 comprises a condition data extraction unit 2101 and a feature size calculation unit 2102.

[0088] The feature size acquisition unit 210, the inference calculation unit 212, the fluctuation pattern management unit 213, the fluctuation pattern classification unit 214, the feature size anomaly determination unit 215 and the feature model storage unit 216 have similar functions to the feature size acquisition unit 210, the inference calculation unit 212, the fluctuation pattern management unit 213, the fluctuation pattern classification unit 214, the feature size anomaly determination unit 215 and the feature model storage unit 216 in the first embodiment.

[0089] The state data extraction unit 2101 and the feature size calculation unit 2102 have similar functions to the state data extraction unit 2101 and the feature size calculation unit 2102 in the first embodiment.

[0090] The machine inspection result holding unit 217, for example, captures inspection information from an external system that manages an inspection record. The machine inspection result holding unit 217 can be, for example, an input device (not shown) such as a keyboard or a touch panel.

[0091] In response to the feature size anomaly detection unit 215 identifying an anomalous feature size of machine tool 30 and indicating the need for inspection, the machine inspection result holding unit 217 promptly retrieves the inspection result from the external system managing the inspection record as inspection information. The machine inspection result holding unit 217 can receive the inspection result input from the plant manager. The machine inspection result holding unit 217 then outputs the received inspection result to the feature model operating unit 211a.

[0092] Similar to the feature model operating unit 211 of the first embodiment, the feature model operating unit 211a generates, updates, deletes or supplements the feature model stored in the feature model storage unit 216 based on the contexts in the production operation that are entered by the context acquisition unit 111 and the feature size calculated by the feature size calculation unit 2102, which indicates the feature of the operating state of the machine tool 30.

[0093] In addition, based on the determination result of the feature size anomaly determination unit 215 and the inspection result of the machine inspection result holding unit 217, the feature model operating unit 211a updates the feature model stored in the feature model storage unit 216.

[0094] Specifically, in a case where the feature size anomaly detection unit 215 determines that the feature size of the machine tool 30 is anomalous, and the machine inspection result holding unit 217 receives the inspection result indicating that the machine tool 30 is normal, the feature model operating unit 211a performs machine learning using the feature size determined to be normal according to the inspection result and updates the feature model of the contexts (their combination) in the production operation that are input by the context acquisition unit 111.

[0095] If the fluctuation pattern classification unit 214 determines the fluctuation pattern as a "back and forth pattern", the feature model operating unit 211a can perform machine learning using the feature size, which includes two states surrounded by circles, as shown in Fig. As shown in 11A, execute, generate a new feature model m and add the new feature model m.

[0096] If the fluctuation pattern classification unit 214 determines the fluctuation pattern to be a "plateau pattern", the feature model operating unit 211a can perform machine learning using the feature size, which includes a state enclosed in a circle, as shown in Fig. As shown in Figure 11B, generate a new feature model m and add the new feature model m. If the fluctuation pattern classification unit 214 determines the fluctuation pattern to be an "increasing pattern", the feature model operating unit 211a may not update the feature model until such an increase slows down. <Anomalienerfassungsverarbeitung der Anomalienerfassungsvorrichtung 20A>

[0097] Next, the anomaly detection processing sequence of the anomaly detection device 20A will be described with reference to Fig. 12 described.

[0098] Fig. Figure 12 is a flowchart describing the processing of the anomaly detection of the anomaly detection device 20A.

[0099] It should be noted that the processing of steps S21 to S27 is the same as the processing of steps S11 to S17 in Fig. 9 is similar and therefore its description is omitted.

[0100] In step S28, the machine inspection result holding unit 217 promptly retrieves the inspection result of machine tool 30 from the external system that manages the inspection records, in response to the inspection output requirement in step S27. The machine inspection result holding unit 217 can receive the inspection result input from the plant manager.

[0101] In step S29, the feature model operating unit 211a updates the feature model based on the determination result from step S26 and the inspection result from step S28.

[0102] With the configuration described above, the anomaly detection device 20A, according to the second embodiment, does not record every single data element on the industrial machine, but rather records the data on changes in the state of the industrial machine as a whole. This allows the anomaly detection process for determining the time of inspection to be carried out with minimal effort. In other words, the anomaly detection device 20A determines, based on the fluctuation pattern, whether the risk of machine failure is high or low. If the risk is determined to be high, it records this risk as an anomaly and requests the inspection of the machine tool 30.

[0103] The anomaly detection device 20A can ignore a temporary change that might be considered an anomaly when detected based on a threshold, thereby reducing unnecessary inspections.

[0104] If the evaluation value changes, the anomaly detection device 20A can assess the risk by checking the fluctuation pattern.

[0105] The anomaly detection device 20A automatically determines, based on the fluctuation pattern, whether the feature model should be updated or not, so that it is not necessary to adjust a threshold and thus the problem of a suitable threshold being initially unknown can be avoided.

[0106] The second embodiment was described above.

[0107] With the configuration described above, the anomaly detection devices 20, 20A according to the first and second embodiments do not capture every single data element on the industrial machine, but capture the data on changes in the state of the industrial machine as a whole, so that anomaly detection can be carried out to represent the time of the inspection in the least possible way. <Modifikation 1>

[0108] In the first embodiment and the second embodiment, the anomaly detection device 20, 20A is a device that differs, for example, from the numerical control device 10, but the numerical control device 10 may include some or all functions of the anomaly detection device 20, 20A.

[0109] Alternatively, for example, a server may include some or all of the following units: the feature size acquisition unit 210, the feature model operating unit 211, the inference calculation unit 212, the variation pattern management unit 213, variation pattern classification unit 214, feature size anomaly determination unit 215, and feature model storage unit 216 of the anomaly detection device 20, or some or all of the feature size acquisition unit 210, feature model operating unit 211a, inference calculation unit 212, variation pattern management unit 213, variation pattern classification unit 214, feature size anomaly determination unit 215, feature model storage unit 216, and machine inspection result holding unit 217 of the anomaly detection device 20A. Each function of the anomaly detection device 20, 20A can be implemented in the cloud using a virtual server function or the like.

[0110] The anomaly detection device 20, 20A can be a distributed processing system in which the functions of the anomaly detection device 20, 20A are distributed across a large number of servers as needed. <Modifikation 2>

[0111] In the embodiments described above, for example, a numerical control device 10 is connected to an anomaly detection device 20, 20A, but the present disclosure is not limited thereto. For example, an anomaly detection device 20, 20A can be connected to two or more numerical control devices 10. For example, two or more anomaly detection devices 20, 20A can be provided as an integrated anomaly detection system, and each anomaly detection device 20, 20A can be connected to one or more numerical control devices 10. In this case, some or all functions of the fluctuation pattern management unit 213 can be common to the anomaly detection devices 20, 20A.

[0112] With this configuration, the anomaly detection devices 20, 20A can detect the anomaly according to the same standard. <Modifikation 3>

[0113] In the embodiments described above, the anomaly detection device 20, 20A determines, for example, whether the feature size of the machine tool 30 is anomalous or not, based on the classification result, which is based on the comparison between the conclusion result and the predetermined fluctuation pattern stored in the fluctuation pattern management unit 213. However, the present disclosure is not limited to this. For example, the anomaly detection device 20, 20A can store a previous fluctuation pattern in a storage unit (not shown) and assess the risk by comparing the classification result with the previous fluctuation pattern.

[0114] Alternatively, the anomaly detection device 20, 20A can notify the need for inspection at the time of the fluctuation, prompting the plant manager to carry out the inspection, and review the fluctuation pattern after several days to assess the inspection result and the risk.

[0115] It should be noted that each function of the anomaly detection devices 20, 20A in the first embodiment and the second embodiment can be implemented by hardware, software, or a combination thereof. Here, implementation by software means implementation by reading and executing a program by a computer.

[0116] The program can be stored and delivered to the computer using various types of non-volatile, computer-readable media. Non-volatile, computer-readable media include various types of tangible storage media. Examples of non-volatile, computer-readable media are magnetic recording media (e.g., flexible floppy disks, magnetic tapes, and hard disks), magneto-optical recording media (e.g., magneto-optical floppy disks), CD-ROMs (CD-Read Only Memory), CD-Rs, CD-R / Ws, and semiconductor memory (e.g., mask ROM, programmable ROM (PROM), erasable PROM (EPROM), flash ROM, and RAM). Alternatively, the program can be delivered to the computer via various types of volatile, computer-readable media. Examples of volatile, computer-readable media include electrical signals, optical signals, and electromagnetic waves.The volatile, computer-readable medium can deliver the program to the computer via a wired communication path such as an electrical wire or optical fiber, or a wireless communication path.

[0117] It should be noted that the steps for describing the program recorded on the recording medium include not only processing operations executed in chronological order, but also parallel or separately executed processing operations. These program description steps can be implemented using cloud computing.

[0118] The present disclosure has been described in detail, but is not limited to the embodiments described above. Various additions, substitutions, modifications, partial omissions, and the like may be made to these embodiments without deviating from the core of the present disclosure or from the core of the present disclosure as derived from the content of the claims and their equivalents. These embodiments may be implemented in combination. For example, the sequence of each type of operation and the sequence of each type of processing have been described as examples in the embodiments described above, and the present disclosure is not limited to such examples. The same applies to numerical values ​​or mathematical expressions when used in the description of the embodiments described above.

[0119] Furthermore, the following additional remarks on the above-mentioned embodiments and modifications are disclosed. (Additional Note 1)

[0120] An anomaly detection device (20) comprises a feature size acquisition unit (210) that acquires a context in the production operation of an industrial machine (30) and a state size that indicates the state of the production operation of the industrial machine (30), and acquires a feature size related to the state of the production operation of the industrial machine (30) based on the acquired context and state size; a feature model operating unit (211) that generates, updates, deletes, or adds a feature model based on the context and the feature size; a feature model storage unit (216) that stores the feature model generated by the feature model operating unit (211); a conclusion calculation unit (212) that calculates an evaluation value of the operating state of the industrial machine (30) based on the feature size and the feature model; and a fluctuation pattern management unit (213).which stores and manages a predetermined fluctuation pattern of the evaluation value, a fluctuation pattern classification unit (214) which classifies the fluctuation pattern of the evaluation value of the operating state of the industrial machine (30) based on a calculation result of the conclusion calculation unit (212) and the predetermined fluctuation pattern stored in the fluctuation pattern management unit (213), and a feature size anomaly determination unit (215) which determines, based on a classification result of the fluctuation pattern classification unit (214), whether the feature size of the industrial machine (30) is anomalous or not. (Additional Note 2)

[0121] In the anomaly detection device (20) from Additional Note 1, the fluctuation pattern classification unit (214) classifies the fluctuation pattern into a fluctuation pattern in which the evaluation value calculated by the inference calculation unit (212) increases at least continuously, a fluctuation pattern in which the evaluation value increases and remains high, a fluctuation pattern in which the evaluation value repeatedly increases and decreases, and a fluctuation pattern in which the evaluation value increases and then decreases to an initial value. (Additional Note 3)

[0122] The anomaly detection device (20A) from additional note 1 or 2 further comprises a machine inspection result holding unit (217) which promptly receives the input of an inspection result of the industrial machine (30), and on the basis of a determination result of the feature size anomaly determination unit (215) and the inspection result of the machine inspection result holding unit (217), the feature model operating unit (211a) updates the feature model stored in the feature model storage unit (216). (Additional note 4)

[0123] In the anomaly detection device (20, 20A) from any of Additional Notes 1 to 3, part or all of the fluctuation pattern management unit (213) is common to other anomaly detection devices (20, 20A), and the fluctuation pattern detected by the anomaly detection device (20, 20A) is used in the other anomaly detection devices (20, 20A). (Additional note 5)

[0124] A control device (10) is a control device for controlling a motor (31) which drives a mechanism of an industrial machine (30) which includes the anomaly detection device (20, 20A) of any of the additional notes 1 to 3. (Additional Note 6)

[0125] An anomaly detection procedure is an anomaly detection procedure that causes a computer to act as an anomaly detection device (20), comprising: a feature size acquisition step for acquiring a context in the production operation of an industrial machine (30) and a state size indicating the state of the production operation of the industrial machine (30), and, based on the acquired context and state size, the acquisition of a feature size associated with the state of the production operation of the industrial machine (30), a feature model operation step for generating, updating, deleting, or adding a feature model based on the context and feature size, and a feature model storage step for storing the feature model generated by the feature model operation step.a conclusion calculation step for calculating an evaluation value of the operating state of the industrial machine (30) based on the feature size and the feature model, a fluctuation pattern management step for storing and managing a predetermined fluctuation pattern of the evaluation value, a fluctuation pattern classification step for classifying the fluctuation pattern of the evaluation value of the operating state of the industrial machine (30) based on a calculation result of the conclusion calculation step and the predetermined fluctuation pattern stored in the fluctuation pattern management step, and a feature size anomaly determination step for determining whether the feature size of the industrial machine (30) is anomalous, based on a classification result of the fluctuation pattern classification step. (Additional note 7)

[0126] A program is a program that causes a computer to perform a feature size acquisition step to acquire a context in the production operation of an industrial machine (30) and a state size indicating the state of the production operation of the industrial machine (30), and, based on the acquired context and the acquired state size, to acquire a feature size associated with the state of the production operation of the industrial machine (30), a feature model operation step to create, update, delete, or add a feature model based on the context and the feature size, a feature model storage step to store the feature model created by the feature model operation step, and a conclusion calculation step to calculate an evaluation value of the operating state of the industrial machine (30) based on the feature size and the feature model.to implement a fluctuation pattern management step for storing and managing a predetermined fluctuation pattern of the evaluation value, a fluctuation pattern classification step for classifying the fluctuation pattern of the evaluation value of the operating state of the industrial machine (30) based on a calculation result of the conclusion calculation step and the predetermined fluctuation pattern stored in the fluctuation pattern management step, and a feature size anomaly determination step for determining, based on a classification result of the fluctuation pattern classification step, whether the feature size of the industrial machine (30) is anomalous or not. (Additional Note 8)

[0127] An integrated anomaly detection system comprises the anomaly detection device (20, 20A) of one of Additional Notes 1 to 3, comprising two or more anomaly detection devices (20, 20A), wherein some or all of the fluctuation pattern management units (213) are shared by the anomaly detection devices (20, 20A) and a fluctuation pattern detected by one of the anomaly detection devices (20, 20A) is used in the other anomaly detection devices (20, 20A). EXPLANATION OF DRAWING REFERENCES 10 Numerical control device 110 Numerical Control Unit 111 Context Capture Unit 112 State variable acquisition unit 20, 20A Anomaly Detection Device 210 Feature size-capture unit 2101 State Data Extraction Unit 2102 Feature size - Calculation unit 211, 211a Feature model - operating unit 212 Conclusion - Calculation Unit 213 Fluctuation Pattern Management Unit 214 Fluctuation Pattern Classification Unit 215 Feature size-anomaly detection unit 216 Feature model storage unit 217 Machine Inspection Result Holding Unit 30 machine tool 31 Engine 32 mechanism QUOTES INCLUDED IN THE DESCRIPTION

[0000] This list of documents cited by the applicant was automatically generated and is included solely for the reader's convenience. The list is not part of the German patent or utility model application. The DPMA accepts no liability for any errors or omissions. Cited patent literature

[0000] JP 2020-123191

[0003] JP 2022-128824

[0003] WO 2023 / 007825

[0003]

Claims

[1] An anomaly detection device comprising: a feature size acquisition unit that captures a context in the production operation of an industrial machine and a state size that indicates an operating state of the production operation of the industrial machine, and on the basis of the captured context and the captured state size captures a feature size that is associated with the operating state of the production operation of the industrial machine; a feature model operating unit that creates, updates, deletes, or adds a feature model based on the context and feature size; a feature model storage unit that stores the feature model generated by the feature model operating unit; a conclusion calculation unit that calculates an evaluation value for the operating state of the industrial machine based on the characteristic size and the characteristic model; a fluctuation pattern management unit that stores and manages a predetermined fluctuation pattern 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 a calculation result of the conclusion calculation unit and the predetermined fluctuation pattern stored in the fluctuation pattern management unit; and a feature size anomaly determination unit that determines, based on a classification result of the fluctuation pattern classification unit, whether the feature size of the industrial machine is anomalous or not. [2] The anomaly detection device according to claim 1, wherein the fluctuation pattern classification unit classifies the fluctuation pattern into a fluctuation pattern in which the evaluation value calculated by the inference calculation unit increases at least continuously, a fluctuation pattern in which the evaluation value increases and remains high, a fluctuation pattern in which the evaluation value repeatedly increases and decreases, and a fluctuation pattern in which the evaluation value increases and then decreases to an initial value. [3] The anomaly detection device according to claim 2, further comprising: a machine inspection result holding unit that promptly receives an input of an inspection result from the industrial machine, wherein the feature model operating unit updates the feature model stored in the feature model storage unit based on a determination result of the feature size anomaly determination unit and the inspection result of the machine inspection result holding unit. [4] The anomaly detection device according to any one of claims 1 to 3, wherein a part or all of the fluctuation pattern management unit is common to those of other anomaly detection devices and the fluctuation pattern detected by the anomaly detection device is used in the other anomaly detection devices. [5] A control device for controlling a motor that drives a mechanism of an industrial machine, comprising: the anomaly detection device according to one of claims 1 to 3. [6] An anomaly detection method that causes a computer to act as an anomaly detection device, comprising: a feature size acquisition step for capturing a context in the production operation of an industrial machine and a state size that indicates a state of the production operation of the industrial machine, and for capturing a feature size that is associated with the state of the production operation of the industrial machine, based on the captured context and the captured state size; a feature model operation step to create, update, delete or add a feature model based on the context and feature size; a feature model storage step for storing the feature model generated by the feature model operation step; a conclusion calculation step to calculate an evaluation value of the operating state of the industrial machine based on the characteristic variable and the characteristic model; a fluctuation pattern management step for storing and managing a predetermined fluctuation pattern of the evaluation value; a fluctuation pattern classification step for classifying the fluctuation pattern of the evaluation value of the operating state of the industrial machine based on a calculation result of the conclusion calculation step and the predetermined fluctuation pattern stored in the fluctuation pattern management step; and a feature size anomaly determination step in which, based on a classification result of the step to classify the fluctuation pattern, it is determined whether the feature size of the industrial machine is anomalous or not. [7] A program that a computer a feature size acquisition step to capture a context in the production operation of an industrial machine and a state size that indicates a state of the production operation of the industrial machine, and to capture a feature size associated with the state of the production operation of the industrial machine on the basis of the captured context and the state size, a feature model operation step for creating, updating, deleting, or adding a feature model based on the context and feature size; a feature model storage step for storing the feature model created by the feature model operation step; a conclusion calculation step to calculate an evaluation value of the operating state of the industrial machine based on the characteristic variable and the characteristic model, a fluctuation pattern management step for storing and managing a predetermined fluctuation pattern of the evaluation value, a fluctuation pattern classification step for classifying the fluctuation pattern of the evaluation value of the operating state of the industrial machine based on a calculation result of the conclusion calculation step and the predetermined fluctuation pattern stored in the fluctuation pattern management step, and a feature size anomaly determination step to determine whether the feature size of the industrial machine is anomalous, based on a classification result of the fluctuation pattern classification step. [8] An integrated anomaly detection system, further comprising: the anomaly detection device according to one of claims 1 to 3, comprising two or more anomaly detection devices, where some or all of the fluctuation pattern management units are common to the anomaly detection devices, and A fluctuation pattern detected by one of the anomaly detection devices is used in the other anomaly detection devices.

Images