State classification device and recording medium

Abstract

A state classification device according to the present invention comprises: a feature amount acquisition unit that acquires a feature amount set pertaining to a prescribed physical quantity detected during operation of industrial machinery; a state sensing unit that, on the basis of the acquired feature amount set, senses that the industrial machinery has transitioned out of a prescribed state; a clustering unit that executes a clustering process using a plurality of feature amount sets from when the transition of the state of the industrial machinery has been sensed to create a cluster; a label assignment assistance unit that assists in work performed by a user to assign a label to the cluster; a model-updating unit that, by using the feature amount sets belonging to the labeled cluster, updates a classification model used in classifying the state of the industrial machinery; a state classification unit that carries out a process for classifying the state of the industrial machinery using the classification model; and a classification result output unit that outputs the result of the classification process.

Description

Technical Field

[0001] The present invention relates to a state classification device and a recording medium.

Background Art

[0002] Many parts are used in industrial machines such as machine tools and robots. Therefore, the state changes that occur in industrial machines are diverse. For example, in the machining of workpieces by a machine tool, when a tool defect (such as wear or breakage) or a bearing defect (such as foreign matter intrusion, poor lubrication, or breakage) occurs, the machining accuracy deteriorates. As a result, defects occur in the workpiece. When defects occur in the workpiece, the yield rate deteriorates. In some cases, it may affect all processes of line production and result in high-cost scrap. Therefore, a method has been proposed to measure various physical quantities detected in industrial machines in real time, detect changes in the state of industrial machines based on the measured physical quantities, and predict the timing of state changes.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] Even if it is possible to detect changes in the state of industrial machinery, it is not always possible to estimate the type of state or predict when the state change will occur. In order to perform preventive maintenance for various state changes, it is necessary to be able to detect state changes as much as possible and then estimate the type of state that has changed. To solve this, one could consider creating a supervised learning model that has been trained using machine learning techniques and then classifying the state of industrial machinery using the created model. However, performing such machine learning requires collecting data on physical quantities detected in various industrial machine states in advance, which is very labor-intensive. Furthermore, when a new unknown state occurs, the user needs to label the data detected at that time and update the model sequentially, which also places a heavy burden on the user. Therefore, there is a need for a system that can reduce the workload required to classify the condition of industrial machinery. [Means for solving the problem]

[0005] The state classification device described in this disclosure detects unknown states by clustering the states of industrial machinery and enables labeling of data related to physical quantities detected in those states. The above-mentioned problems are then solved by updating the machine learning model used for state classification using the labeled data.

[0006] Furthermore, one aspect of the present disclosure is a state classification device for classifying the state of an industrial machine, comprising: a feature acquisition unit that acquires a set of features relating to a predetermined physical quantity detected during the operation of the industrial machine; a state detection unit that detects that the industrial machine has changed from a predetermined state based on the set of features acquired by the feature acquisition unit; a clustering unit that performs clustering processing using multiple set of features when the state detection unit detects a change in the state of the industrial machine, and creates at least one cluster of the set of features detected at the time of the predetermined state change; and an unknown cluster created by the clustering unit. By displaying the similarity matrix between the clusters created when a classification model was previously created,The state classification device comprises: a label assignment support unit that assists the user in assigning labels; a model update unit that updates a classification model used for classifying the state of the industrial machine using a set of features belonging to clusters to which labels have been assigned by the label assignment support unit; a state classification unit that performs state classification processing of the industrial machine using the classification model; and a classification result output unit that outputs the results of the classification processing.

[0007] Other aspects of the present disclosure include a computer-readable recording medium for recording a program executed in a state classification device for classifying the state of an industrial machine, comprising: a feature acquisition unit for acquiring a set of features relating to a predetermined physical quantity detected during the operation of the industrial machine; a state detection unit for detecting that the industrial machine has changed from a predetermined state based on the set of features acquired by the feature acquisition unit; a clustering unit for performing clustering processing using multiple set of features when the state detection unit detects a change in the state of the industrial machine, and creating at least one cluster of the set of features detected at the time of the predetermined state change; and an unknown cluster created by the clustering unit. By displaying the similarity matrix between the clusters created when a classification model was previously created, This is a computer-readable recording medium that records a program that operates a computer as follows: a label assignment support unit that assists the user in assigning labels; a model update unit that updates a classification model used for classifying the state of the industrial machine using the feature set belonging to the cluster to which the label assignment support unit has assigned labels; a state classification unit that performs state classification processing of the industrial machine using the classification model; and a classification result output unit that outputs the results of the classification processing. [Effects of the Invention]

[0008] In one aspect of this disclosure, labeling is performed on a cluster basis, allowing users to label and update models with minimal effort. [Brief explanation of the drawing]

[0009] [Figure 1] This is a hardware configuration diagram of a state classification device according to one embodiment. [Figure 2] Block diagram showing the functions of a state classification device according to one embodiment. [Figure 3] This figure shows an example of a label assignment screen displayed by the label assignment support unit. [Figure 4] This figure shows another example of the label assignment screen displayed by the label assignment support unit. [Figure 5] This figure shows another example of the label assignment screen displayed by the label assignment support unit. [Figure 6] This figure shows another example of the label assignment screen displayed by the label assignment support unit. [Figure 7] This figure shows another example of the label assignment screen displayed by the label assignment support unit. [Figure 8] This figure shows another example of the label assignment screen displayed by the label assignment support unit. [Figure 9] This figure shows another example of the label assignment screen displayed by the label assignment support unit. [Figure 10] This figure shows a modified version of the label assignment screen displayed by the label assignment support unit. [Modes for carrying out the invention]

[0010] Embodiments of the present invention will be described below with reference to the drawings. Figure 1 is a schematic hardware configuration diagram showing the main parts of a state classification device according to one embodiment of the present invention. The state classification device 1 of the present invention can be implemented, for example, as a control device for controlling industrial machinery such as machine tools and robots. Furthermore, the state classification device 1 of the present invention can be implemented on a personal computer attached to a control device for controlling industrial machinery, or on a personal computer, cell computer, fog computer 6, cloud server 7, or other computer connected to the control device via a wired / wireless network. In this embodiment, an example is shown in which the state classification device 1 is implemented on a personal computer connected to a control device for controlling industrial machinery via a network. In this embodiment, a state classification device 1 that detects abnormalities in industrial machinery 3 and classifies the abnormal state is described.

[0011] The CPU 11 included in the state classification device 1 according to this embodiment is a processor that controls the entire state classification device 1. The CPU 11 reads out the system program stored in the ROM 12 via the bus 22 and controls the entire state classification device 1 according to the system program. Temporary calculation data, display data, and various data input from the outside are temporarily stored in the RAM 13.

[0012] The non-volatile memory 14 is composed of, for example, a memory backed up by a battery not shown in the figure or an SSD (Solid State Drive), etc., and the stored state is retained even when the power of the state classification device 1 is turned off. In the non-volatile memory 14, programs and data read from the external device 72 via the interface 15, programs and data input via the input device 71, programs and data acquired from the industrial machine 3, and data related to physical quantities detected by the sensor 4 attached to the industrial machine 3 are stored. The data stored in the non-volatile memory 14 may be expanded in the RAM 13 during execution / use. Also, various system programs such as known analysis programs are pre-written in the ROM 12.

[0013] The interface 15 is an interface for connecting the CPU 11 of the state classification device 1 and an external device 72 such as a USB device. From the external device 72 side, for example, a control program stored in advance and data related to the operation of each industrial machine 3 can be read. Also, control programs and setting data edited within the state classification device 1 can be stored in external storage means via the external device 72.

[0014] The interface 20 is an interface for connecting the CPU of the state classification device 1 and a wired or wireless network 5. The industrial machine 3, fog computer 6, cloud server 7, etc. are connected to the network 5, and data is exchanged with the state classification device 1.

[0015] In the display device 70, data read onto the memory, data obtained as a result of execution of each program, etc. are output via the interface 17 and displayed. Also, an input device 71 composed of a keyboard, a pointing device, etc. passes commands, data, etc. based on operations by an operator to the CPU 11 via the interface 18.

[0016] The interface 21 is an interface for connecting the CPU 11 and the machine learning device 100. The machine learning device 100 includes a processor 101 that controls the entire machine learning device 100, a ROM 102 that stores system programs, etc., a RAM 103 for temporarily storing each process related to machine learning, and a non-volatile memory 104 used for storing models, etc. The machine learning device 100 can observe each piece of information that can be acquired by the state classification device 1 via the interface 21. Also, the state classification device 1 acquires the processing result output from the machine learning device 100 via the interface 21, stores or displays the acquired result, or transmits it to other devices via the network 5 or the like.

[0017] FIG. 2 shows, as a schematic block diagram, the functions included in the state classification device 1 according to an embodiment of the present invention. Each function included in the state classification device 1 according to this embodiment is realized by the CPU 11 of the state classification device 1 shown in FIG. 1 and the processor 101 of the machine learning device 100 executing a system program and controlling the operations of each part of the state classification device 1 and the machine learning device 100.

[0018] The state classification device 1 of this embodiment includes a feature acquisition unit 110, a state detection unit 120, a clustering unit 130, a label assignment support unit 140, a model update unit 150, a state classification unit 160, and a classification result output unit 170. Furthermore, the RAM 103 to non-volatile memory 104 of the machine learning model 100 is pre-configured with a model storage unit 210, which is an area for storing a model that has learned to detect and classify the state of the industrial machine 3, a set of features relating to physical quantities that have been used to train the model in the past, and a feature storage unit 220, which is an area for storing a set of features relating to physical quantities detected when a predetermined state change occurs.

[0019] The feature acquisition unit 110 acquires a set of features related to predetermined physical quantities detected by the sensor 4 during the operation of the industrial machine 3. The set of features related to predetermined physical quantities may include, for example, the position of each motor that operates the industrial machine 3, the speed of each motor, the acceleration of each motor, the torque command of each motor, the temperature of each part, the vibration of each part, sound, optical information, ambient temperature, and ambient humidity. The features related to the predetermined physical quantities may be instantaneous values ​​or continuous time-series data. In addition, values ​​obtained by performing predetermined statistical processing on the acquired features, or values ​​obtained by frequency analysis, may be added to the set of features related to the predetermined physical quantities. The feature acquisition unit 110 may also acquire a set of features related to the physical quantities of the industrial machine 3 that has been acquired and stored by an external device 72, a fog computer 6, a cloud server 7, etc. The feature acquisition unit 110 outputs the acquired set of features related to physical quantities to the state detection unit 120.

[0020] The state detection unit 120 detects when the industrial machine 3 has changed from a predetermined state based on a set of feature quantities related to physical quantities input from the feature acquisition unit 110. Typically, the state detection unit 120 may detect when the industrial machine 3 has changed from a normal state to an abnormal state. When detecting the occurrence of an abnormality, for example, a normal model based on a set of feature quantities related to multiple physical quantities detected when the industrial machine 3 was operating normally is stored in the model storage unit 210 beforehand. The state detection unit 120 then calculates the degree of deviation of the set of feature quantities related to physical quantities input from the feature acquisition unit 110 compared to the distribution of feature quantities in the normal model. If the calculated degree of deviation exceeds a predetermined threshold, it can be determined that an abnormality has occurred in the industrial machine 3. The state detection unit 120 may also set the threshold automatically based on the distribution of the normal model. For example, the threshold may be set at a position separated by the mean value X × standard deviation σ from the mean value X of the normal model. Alternatively, the threshold may be set using the third quartile or the interquartile range. To detect such a change from a predetermined state, known methods such as Hotelling's theory or the k-nearest neighbor method can be used. When the state detection unit 120 detects a change in the state of the industrial machine 3 based on a set of features related to a physical quantity, it stores that set of features related to the physical quantity in the feature storage unit 220.

[0021] The clustering unit 130 executes a known clustering process using feature sets related to multiple physical quantities stored in the feature memory unit 220, and creates a cluster of feature sets related to physical quantities detected when a predetermined state change occurs. The feature memory unit 220 stores multiple feature sets related to physical quantities when the state detection unit 120 detects that the industrial machine 3 has changed from a predetermined state. When this feature set related to physical quantities is detected, the state of the industrial machine 3 is unclassified. The clustering unit 130 may execute the clustering process when a predetermined number of feature sets related to physical quantities have been stored in the feature memory unit 220. As for the clustering method, known methods such as the k-nearest neighbor method, HDBSCAN (Hierarchical Density-Based Spatial Clustering of Applications with Noise), BIRCH (Balanced Iterative Reducing and Clustering using Hierarchies), and DPGMM (Dirichlet Process Gaussian Mixture Model) may be used. When performing clustering, it may be possible to display a screen to the user that shows the clustering results and allows them to adjust hyperparameters (such as the number of clusters) used in the clustering process.

[0022] The label assignment support unit 140 provides a user interface to assist the user in assigning labels to clusters of feature sets related to physical quantities created by the clustering unit 130. The label assignment support unit 140 displays the clusters created by the clustering unit 130 on a display device 70, for example, so that the user can see them. Figure 3 shows an example of a label setting screen displayed by the label assignment support unit 140. In the example in Figure 3, the clustered feature sets are plotted on a two-dimensional graph with features related to a predetermined physical quantity as axes, and the range of each cluster is displayed so that it can be understood. The range of the clusters can also be displayed by color coding, for example. In this way, the label assignment support unit 140 may display the clusters created by the clustering unit 130 on a two-dimensional or three-dimensional graph with features related to a predetermined physical quantity as axes. The user may be allowed to select which physical quantity features to use as axes as appropriate. Furthermore, for each plotted feature set, identification information identifying the machine that acquired the feature set, the date and time of acquisition, the operating status at the time of acquisition, any alerts that occurred, and subsequent actions may be obtained and referenced from the logs and maintenance records of industrial machine 3.

[0023] The label assignment support unit 140 may provide a user interface that, as illustrated in Figure 4, allows the user to specify predetermined conditions and perform a search, thereby highlighting feature sets that match the search conditions among the feature sets plotted on the graph. Furthermore, as illustrated in Figure 5, it may also allow the user to overlay and plot feature sets previously used for training on the graph, allowing comparison with feature sets that have not been labeled. In addition, the label assignment support unit 140 may provide a user interface that displays the relationships between multiple clusters. For example, as illustrated in Figure 6, the label assignment support unit 140 may provide the user with a value indicating the similarity between multiple created clusters. The similarity between clusters can be calculated using known methods, such as using the decrease in the log-likelihood when two clusters are combined into one as the similarity.

[0024] The label assignment support unit 140 may provide a user interface that displays a similarity matrix showing the similarity between each cluster and previously generated known clusters, as illustrated in Figure 7. In the similarity matrix in Figure 7, the clustering unit calculates the similarity between clusters A to E, created by the clustering unit based on a set of features obtained from a predetermined industrial machine 3, and each cluster "RD18 Clst_A" to "RD24 Clst_3," which was previously created based on a set of features obtained from another industrial machine 3, and displays the calculated similarity in a table format. As illustrated in Figure 7, by specifying a previously created cluster, information about the data belonging to that cluster and the labels assigned to that cluster may be displayed. It may also display the similarity between newly created clusters or the similarity between previously created clusters. For example, in the example in Figure 7, the similarity between clusters A to E may be displayed, or the similarity between "RD18 Clst_A" to "RD24 Clst_3" may be displayed. By displaying such a similarity matrix, the user can consider what labels to assign to a new cluster while referring to what labels have been assigned to similar clusters.

[0025] The label assignment support unit 140 may display clusters for a given machine on a graph with a predetermined feature quantity on the vertical axis and time on the horizontal axis, as illustrated in Figure 8. In the graph in Figure 8, a graph is drawn with motor vibration on the vertical axis and time on the horizontal axis for the feature quantities obtained from a given industrial machine 3, and clusters A and B created by the clustering unit based on the acquired feature quantity set are displayed on the graph in an identifiable manner. In the display illustrated in Figure 8, the deterioration of parts and the occurrence of failures of the given industrial machine 3 can be visually grasped as changes and developments over time. Each cluster may be displayed by enclosing it in a frame or by color-coding it, as illustrated in Figure 8, to distinguish it from other data. The calculated similarity is displayed in a table format. By displaying such a graph, the user can consider what labels to assign to a new cluster for a given industrial machine 3 while referring to what labels were assigned to similar clusters in the past.

[0026] The label assignment support unit 140 may display clusters with variable granularity. For example, normal data may be displayed with coarser granularity, while data closer to abnormality may be displayed with finer granularity. Also, when displaying a graph with time on the horizontal axis, as illustrated in Figure 8, the granularity of the displayed data may be changed by changing the time scale according to a predetermined operation. By making the granularity of the displayed data variable in this way, particularly important data sets can be displayed in detail, supporting the user's analysis of clusters.

[0027] The labeling support unit 140 may display representative features of each cluster in a comparable manner, as illustrated in Figure 9. In the example in Figure 9, representative values ​​of torque command (TCMD), torque command variance, sidebands, velocity, and one-turn components for the selected cluster are displayed overlaid on the radar chart, making each cluster comparable. Generally, sets of features obtained under abnormal conditions have different characteristics from sets of features obtained under normal conditions. Therefore, by displaying the features overlaid on a radar chart or line graph, as illustrated in Figure 9, users can more easily recognize clusters of data sets obtained under abnormal conditions.

[0028] By viewing this screen, users can determine what state each cluster represents for industrial machine 3 and assign a label related to that state. The labels assigned by the user to a cluster are then applied to all feature sets belonging to that cluster.

[0029] The model update unit 150 updates the model for classifying the state of the industrial machine 3 stored in the model memory unit 210 using the feature set related to the physical quantities that have been labeled by the labeling support unit 140. The model update unit 150 updates the model for classifying the state of the industrial machine 3 using a known supervised learning method. The model for classifying the state of the industrial machine 3 may be, for example, a known neural network, SVM, regression model, decision tree model, Bayesian classification model, etc. When updating the model, it is not necessary to use all of the feature set related to the physical quantities that have been labeled by the labeling support unit 140. For example, feature sets related to physical quantities that have a high similarity to the feature set related to the physical quantities that has been used to train the model in the past and are stored in the model memory unit 210 do not need to be used in the model update process. After updating the model, the model update unit 150 deletes the feature set stored in the feature memory unit 220. The feature set used to update the model is also stored in the model memory unit 210 as a feature set related to physical quantities that has been used to train the model in the past.

[0030] The state classification unit 160 performs a state classification process for the industrial machine 3 using a classification model stored in the model storage unit 210, based on the set of feature quantities related to physical quantities when the state detection unit 120 detects a change in the state of the industrial machine 3.

[0031] The classification result output unit 170 outputs the classification result of the state of the industrial machine 3 by the state classification unit 160 to the display device 700. The classification result output unit 170 may also transmit the classification result of the state of the industrial machine 3 to the industrial machine 3 via the network 5. Alternatively, the classification result output unit 170 may transmit the classification result of the state of the industrial machine 3 to a higher-level computer such as a fog computer 6 or a cloud server 7 via the network 5. It may also output to a log recording area pre-provided on a non-volatile memory 14 or the like.

[0032] By using the state classification device 1 according to this embodiment, which has the above configuration, labeling is performed on a cluster basis, allowing users to label and update the model with less effort. When labeling, it is possible to refer to the information at the time each feature was acquired and compare it with features used in past training, so it is expected that the effort required for the user to perform the labeling work will be reduced through these support measures.

[0033] As one modification of the state classification device 1 according to this embodiment, the label assignment support unit 140 may instruct the state classification unit 160 to classify the feature sets related to physical quantities stored in the feature storage unit 220, and assign a provisional label to each cluster based on the classification result. In this case, for example, for each cluster, when a predetermined percentage of feature sets exceeding a certain threshold are classified under the same label, that classification can be assigned as the provisional label for that cluster. Figure 10 shows an example of the label assignment screen displayed by the label assignment support unit 140 according to this modification. In the example in Figure 10, a predetermined percentage or more of feature sets belonging to cluster B are classified as tool breakage abnormalities. Also, a predetermined percentage or more of feature sets belonging to cluster E are classified as motor bearing abnormalities. Therefore, the label assignment support unit 140 automatically assigns a provisional label of tool breakage abnormalities to cluster B and a provisional label of motor bearing abnormalities to cluster E. By implementing this configuration, users can assign labels to each cluster while referring to classification results using existing classification models, which is expected to further reduce the effort required for users to perform labeling tasks.

[0034] Although embodiments of the present invention have been described above, the present invention is not limited to the examples of embodiments described above, and can be implemented in various forms by making appropriate modifications. [Explanation of Symbols]

[0035] 1. State classification device 3. Industrial Machinery 4 sensors 5 Network 6. Fog Computer 7 Cloud Server 11 CPU 12 ROM 13 RAM 14 Non-volatile memory 15 Interfaces 17,18,20,21 Interface 22 buses 70 Display device 71 Input device 72 External equipment 100 machine learning machines 101 Processors 102 ROM 103 RAM 104 Non-volatile memory 110 Feature acquisition unit 120 State detection unit 130 Clustering section 140 Labeling Support Department 150 Model Update Section 160 State Classification Section 170 Classification result output section 210 Model Memory Unit 220 Feature Memory Unit

Claims

1. A condition classification device for classifying the condition of industrial machinery, A feature quantity acquisition unit acquires a set of feature quantities related to a predetermined physical quantity detected during the operation of the industrial machine, A state detection unit detects when the industrial machine has changed from a predetermined state based on the feature set acquired by the feature acquisition unit, A clustering unit that performs clustering processing using multiple feature sets when the state detection unit detects a change in the state of the industrial machine, and creates at least one cluster of feature sets detected at a predetermined state change, A label assignment support unit assists the user in assigning labels by displaying a similarity matrix between the unknown clusters created by the clustering unit and the clusters created when a classification model was previously created. A model update unit updates the classification model used for classifying the state of the industrial machine using the feature set belonging to the cluster to which labels have been assigned by the label assignment support unit, A state classification unit performs state classification processing of the industrial machine using the classification model, A classification result output unit that outputs the results of the classification process, A state classification device equipped with the following features.

2. The state detection unit automatically sets a threshold used for determining state changes based on the distribution of data from the model used for detecting state changes in the industrial machine. The state classification device according to claim 1.

3. The label assignment support unit compares and displays the features belonging to the cluster with the features used when creating the classification model in the past. The state classification device according to claim 1.

4. The label assignment support unit highlights the feature quantities belonging to the cluster that meet predetermined conditions. The state classification device according to claim 1.

5. The label assignment support unit assigns a provisional label to the cluster using a classification model created in the past. The state classification device according to claim 1.

6. The label assignment support unit displays the data at a variable level when displaying clusters. The state classification device according to claim 1.

7. The label assignment support unit displays representative features of each cluster created by the clustering unit in a comparable manner. The state classification device according to claim 1.

8. A computer-readable recording medium for recording a program executed in a state classification device that classifies the state of industrial machinery, A feature acquisition unit acquires a set of feature quantities related to a predetermined physical quantity detected during the operation of the industrial machine. A state detection unit detects when the industrial machine has changed from a predetermined state, based on the set of features acquired by the feature acquisition unit. A clustering unit performs clustering processing using multiple feature sets when the state detection unit detects a change in the state of the industrial machine, and creates at least one cluster of the feature sets detected at a predetermined state change. A label assignment support unit assists the user in assigning labels by displaying a similarity matrix between the unknown clusters created by the clustering unit and the clusters created when a classification model was previously created. A model update unit updates the classification model used for classifying the state of the industrial machine using the feature set belonging to the cluster to which labels have been assigned by the label assignment support unit. A state classification unit that performs state classification processing of the industrial machine using the classification model described above. A classification result output unit that outputs the results of the classification process, A computer-readable recording medium that stores programs used to operate a computer.

Images