Data analysis device, data analysis system, data analysis method, data analysis program, learning device, and inference device

Abstract

A data analysis device (10) comprises: a data acquisition unit (11) that acquires operation data indicating the state of a processing device; a classification processing unit (12) that classifies the operation data into steady-state data in which values are constant or fluctuate regularly, and fluctuating data in which values fluctuate irregularly; a model generation unit (14) that generates a model for steady-state data by learning the steady-state data and a model for fluctuating data by learning the fluctuating data; and a prediction unit (16) that predicts the life of components used in the processing device by inputting the steady-state data into the model for steady-state data and the fluctuating data into the model for fluctuating data.

Description

Data analysis device, data analysis system, data analysis method, data analysis program, learning device, and inference device

[0001] The present disclosure relates to a data analysis device, a data analysis system, a data analysis method, a data analysis program, a learning device, and an inference device for analyzing data.

[0002] As one method for predicting the life of a consumable used in a processing device, there is a method of learning operation data, which is data related to the operation of the processing device, such as data on the load applied to the consumable, and predicting the life of the consumable based on a model generated by the learning.

[0003] Patent Document 1 discloses an abnormality determination device for determining an operation abnormality of an industrial machine. Among the observation data related to the operation of the industrial machine, data within a predetermined time width at a determined timing is extracted as partial time-series data, and machine learning is performed using the statistic calculated from the partial time-series data as learning data. The abnormality determination device according to Patent Document 1 acquires, as observation data, data when the industrial machine is operating normally in order to determine an operation abnormality of the industrial machine. Which timing of the observation data is used as the partial time-series data is set in consideration of the operation characteristics of the industrial machine.

[0004] Japanese Patent Application Laid-Open No. 2021-15573

[0005] The operating state of the processing device includes a steady state in which a constant value or a value that fluctuates regularly is acquired as operation data, and a fluctuating state in which a value that fluctuates irregularly is acquired as operation data. For example, when the processing device starts an operation according to a control signal, the processing device takes a fluctuating state until the operation stabilizes, and then transitions to a steady state and continues the operation. Not only the operation of the processing device in the steady state but also the operation of the processing device in the fluctuating state will affect the life of the consumable.

[0006] In the technology disclosed in Patent Document 1, a model for anomaly detection is generated by learning data from when the industrial machine is operating normally. The changes in operating data values ​​when the industrial machine is in a fluctuating state are completely different from the changes in operating data values ​​when the industrial machine is in a normal and steady state. Therefore, if operating data from when the industrial machine is in a fluctuating state is included in the training data, it becomes difficult to generate a model that enables highly accurate anomaly detection. For this reason, in the technology disclosed in Patent Document 1, operating data from when the industrial machine is in a fluctuating state is excluded as data unsuitable for training.

[0007] If the learning method disclosed in Patent Document 1 is applied to predicting the lifespan of consumables used in a processing apparatus, then operating data from when the processing apparatus is in a fluctuating state will be excluded from the learning process. As a result, the operation of the processing apparatus in a fluctuating state will not be considered in the lifespan prediction. Therefore, according to the conventional technology disclosed in Patent Document 1, there is a problem in that it is difficult to accurately predict the lifespan.

[0008] This disclosure is made in view of the above and aims to provide a data analysis device that enables accurate prediction of lifespan.

[0009] To solve the above-mentioned problems and achieve the objective, the data analysis device according to this disclosure includes: a data acquisition unit that acquires operational data indicating the state of the processing device; a classification processing unit that classifies the operational data into steady-state data, in which the value is constant or the value fluctuates regularly, and fluctuating data, in which the value fluctuates irregularly; a model generation unit that generates a model for steady-state data by learning steady-state data and generates a model for fluctuating data by learning fluctuating data; and a prediction unit that predicts the lifespan of parts used in the processing device by inputting steady-state data into the steady-state data model and fluctuating data into the fluctuating data model.

[0010] The data analysis device described in this disclosure has the effect of enabling accurate lifespan prediction.

[0011] A diagram showing an example configuration of the data analysis system according to Embodiment 1. A diagram for explaining the classification of operational data acquired in the data analysis system according to Embodiment 1. A first diagram for explaining the cumulative value of operational data acquired in the data analysis system according to Embodiment 1. A second diagram for explaining the cumulative value of operational data acquired in the data analysis system according to Embodiment 1. A diagram for explaining the model generated by the model generation unit of the data analysis system according to Embodiment 1. A flowchart showing an example of the processing procedure during learning by the data analysis device according to Embodiment 1. A flowchart showing an example of the processing procedure during life prediction by the data analysis device according to Embodiment 1. A diagram showing an example configuration of the data analysis system according to Embodiment 2. A diagram showing an example configuration of the control circuit according to Embodiment 1 or 2.

[0012] Below, a data analysis apparatus, data analysis system, data analysis method, data analysis program, learning apparatus, and inference apparatus according to the embodiment will be described in detail with reference to the drawings.

[0013] Embodiment 1. Figure 1 shows an example of the configuration of a data analysis system 1 according to Embodiment 1. The data analysis system 1 comprises a data analysis device 10 that analyzes the operation data of a processing apparatus and a display device 20 that displays the information. The data analysis device 10 and the display device 20 are connected to each other so as to be able to communicate with each other.

[0014] The data analysis device 10 includes a data acquisition unit 11 that acquires operational data indicating the state of the processing device; a classification processing unit 12 that classifies the operational data into steady-state data, where the value is constant or the value fluctuates regularly, and variable-state data, where the value fluctuates irregularly; a data storage unit 13 that stores the operational data classified into steady-state data and variable-state data; a model generation unit 14 that generates a model for steady-state data by learning steady-state data and generates a model for variable-state data by learning variable-state data; a model storage unit 15 that stores the generated models; and a prediction unit 16 that predicts the lifespan of parts used in the processing device by inputting steady-state data into the steady-state data model and variable-state data into the variable-state data model.

[0015] The display device 20 displays information indicating the lifespan predicted by the data analysis device 10. The display device 20 also functions as an output device that outputs information indicating the lifespan predicted by the data analysis device 10.

[0016] The data acquisition unit 11 acquires operational data, which is data indicating the state of the processing equipment. In Embodiment 1, the operational data is data indicating the state of parts used in the processing equipment, for example, data indicating the state of consumables used in the processing equipment. Examples of operational data include data on the load amount of consumables or data on the amount of wear of consumables. An example of processing equipment is a machine tool that cuts a workpiece by bringing a rotating tool into contact with the workpiece. An example of consumables is a tool. The operational data includes not only data when the processing equipment is in operation, but also data when the processing equipment is stopped. The operational data also includes data indicating the stopped state when the processing equipment is stopped periodically, or data indicating the stopped state when the processing equipment is stopped irregularly.

[0017] Figure 2 is a diagram illustrating the classification of operational data acquired in the data analysis system 1 according to Embodiment 1. Figure 2 shows four graphs with different waveforms. The vertical axis of each graph represents the instantaneous value. Here, the instantaneous value is the value of the operational data acquired by the data acquisition unit 11. The horizontal axis of each graph represents time. The instantaneous value is zero when the processing device is not operating, that is, when the processing device is stopped. The classification processing unit 12 divides the time series of operational data into a certain time width and classifies it into steady-state data and fluctuating data based on the waveform of the graph in the divided time width.

[0018] The graph in Figure 2(a) is an example of a graph where the instantaneous value is nearly constant. Operating data where the instantaneous value is nearly constant, as in this example, is classified as steady-state data.

[0019] The graph in Figure 2(b) is an example of a graph where the instantaneous value fluctuations are regular. In graph (b), a rectangular wave of almost the same shape is repeated, and the instantaneous value fluctuates regularly. Operating data when the instantaneous value fluctuates regularly, as in this example, is classified as steady-state data. Examples of regularly fluctuating instantaneous values ​​include cases where the same waveform shape is repeated, as shown in graph (b), as well as cases where the instantaneous value increases or decreases at an almost constant rate.

[0020] The graph in Figure 2(c) is an example of a graph that includes a period when the processing equipment is stopped. When the processing equipment is stopped, the instantaneous value is zero. Also, in graph (c), the instantaneous value is almost constant when the processing equipment is running. Graph (c) is a graph that shows the case when the processing equipment is in a steady state while it is running, and when the processing equipment is stopped periodically. For example, the graph when the processing equipment is running in a steady state on weekdays and stopped on holidays will be like graph (c). Even when the instantaneous value is like graph (c), the instantaneous value fluctuates regularly, so the operating data is classified as steady-state data. Note that when the processing equipment is stopped irregularly, or when the processing equipment is in a fluctuating state while it is running, the instantaneous value fluctuates irregularly, so the operating data is classified as fluctuating data.

[0021] The graph in Figure 2(d) is an example of a graph when instantaneous values ​​fluctuate irregularly. In graph (d), various waveforms appear irregularly. Operating data in which instantaneous values ​​fluctuate irregularly, as in this example, is classified as fluctuating data. An example of irregular fluctuations in instantaneous values ​​is when a load is applied to the shaft that rotates the tool, causing the instantaneous values ​​to fluctuate irregularly.

[0022] For example, the classification processing unit 12 compares several waveforms used as a reference for steady-state data with the waveforms of the graph over a defined time interval. The waveforms used as a reference for steady-state data are, for example, the waveform when the instantaneous value is constant, or the waveform when the instantaneous value fluctuates regularly. If the classification processing unit 12 finds a waveform among the reference waveforms that is similar to the waveform of the graph over a defined time interval, it classifies the data represented by that graph as steady-state data. The classification processing unit 12 classifies data other than the data classified as steady-state data as fluctuating data.

[0023] The method for determining whether the waveforms of the data to be classified are similar to a reference waveform is arbitrary. The classification processing unit 12 determines whether the waveforms are similar by calculating the distance between the data, for example, using the Euclidean distance or Manhattan distance. The classification processing unit 12 may also calculate the similarity of the waveforms using a method such as Dynamic Time Warping (DTW). The classification processing unit 12 may calculate the similarity after calculating features using a signal processing method such as spectral analysis. The classification processing unit 12 may also determine whether the waveforms are similar using a machine learning method such as the k-means method. The classification processing unit 12 may also compare the waveforms based on shapelets, which are multiple waveform patterns effective for time series classification.

[0024] The classification processing unit 12 classifies the operational data into steady-state data and variable-state data, and then stores the steady-state data and variable-state data in the data storage unit 13.

[0025] The model generation unit 14 reads steady-state data and fluctuating data from the data storage unit 13. The model generation unit 14 generates a model for steady-state data by learning the instantaneous values ​​of steady-state data, and generates a model for fluctuating data by learning the instantaneous values ​​of fluctuating data. The model generation unit 14 stores the steady-state data model and the fluctuating data model, which were generated by learning the instantaneous values, in the model storage unit 15.

[0026] The data storage unit 13 stores cumulative value data, which is a time series of cumulative values ​​obtained by accumulating the instantaneous values ​​of steady-state data and instantaneous values ​​of fluctuating data. The model generation unit 14 reads the cumulative value data stored in the data storage unit 13. The model generation unit 14 generates a model for steady-state data by learning the instantaneous values ​​of steady-state data, and generates a model for fluctuating data by learning the instantaneous values ​​of fluctuating data. After that, it accumulates the instantaneous values ​​of steady-state data and the instantaneous values ​​of fluctuating data into the cumulative value data.

[0027] The model generation unit 14 accumulates instantaneous values ​​with cumulative values, and then generates a model for cumulative values ​​by learning the cumulative values. The model generation unit 14 stores the model for cumulative values, which was generated by learning the cumulative values, in the model storage unit 15.

[0028] The model generation unit 14 stores the cumulative value data, including the cumulative value after accumulating instantaneous values, in the data storage unit 13. This updates the cumulative value data stored in the data storage unit 13.

[0029] Figure 3 is a first diagram illustrating the cumulative value of operational data acquired in the data analysis system 1 according to Embodiment 1. Figure 4 is a second diagram illustrating the cumulative value of operational data acquired in the data analysis system 1 according to Embodiment 1.

[0030] The graph shown in Figure 3 represents an example of cumulative value data stored in the data storage unit 13. In Figure 3, the vertical axis of the graph represents the cumulative value stored in the data storage unit 13. The horizontal axis of the graph represents time. In the example shown in Figure 3, the cumulative value stored in the data storage unit 13 is initialized to zero when the cumulative value reaches the upper limit.

[0031] The graph shown in Figure 4 represents the cumulative value assuming no initialization in the data storage unit 13. Here, the cumulative value assuming no initialization in the data storage unit 13 is referred to as the total cumulative value. In Figure 4, the vertical axis of the graph represents the total cumulative value. The horizontal axis of the graph represents time. The model generation unit 14 converts the cumulative value read from the data storage unit 13 into the total cumulative value and generates a model for the cumulative value by learning the total cumulative value. In the following explanation, "cumulative value" refers to the total cumulative value.

[0032] The cumulative value increases while the processing equipment is running. The increase in the cumulative value stops when the processing equipment is stopped.

[0033] When performing life prediction based on operational data, the prediction unit 16 reads steady-state data and fluctuating data from the data storage unit 13. The prediction unit 16 reads the cumulative values ​​stored in the data storage unit 13. The prediction unit 16 reads the models for instantaneous values, the steady-state data model and the fluctuating data model, from the model storage unit 15. The prediction unit 16 reads the cumulative value model from the model storage unit 15.

[0034] The prediction unit 16 predicts the lifespan by inputting instantaneous values ​​of steady-state data into a model for steady-state data, instantaneous values ​​of fluctuating data into a model for fluctuating data, and cumulative values ​​of operating data into a model for cumulative values. The prediction unit 16 outputs information indicating the predicted lifespan to the display device 20. The method of predicting the lifespan using these various models is arbitrary. For example, the prediction unit 16 may predict the lifespan using regression analysis, a representative statistical method. Alternatively, the prediction unit 16 may predict the lifespan using Gaussian process regression, ARIMA (Auto Regressive Integrated Moving Average) model, or a state-space model. The prediction unit 16 may also predict the lifespan using machine learning methods such as NeuralProphet or LSTM (Long Short Term Memory).

[0035] Instantaneous values, even when they fluctuate, are limited to a certain fixed numerical range. While instantaneous values ​​can be said to accurately represent the state of the processing equipment at each moment, it is difficult to understand the trend of operating data over a relatively long period from instantaneous values ​​alone. In contrast, cumulative values ​​show the trend of operating data over a relatively long period, but because the detailed fluctuations of the original instantaneous values ​​are not easily reflected, it is difficult to understand the state of the processing equipment at each moment from cumulative values ​​alone.

[0036] The data analysis device 10 generates models for instantaneous values, steady-state data, and fluctuating data, as well as a model for cumulative values. By predicting the lifespan based on each of these models, it becomes possible to predict the lifespan while taking into account the trend of instantaneous values ​​and the trend of cumulative values, thereby enabling more accurate lifespan predictions. Furthermore, by separating the models into those for steady-state data and those for fluctuating data, the data analysis device 10 enables accurate lifespan predictions based on operating data when the processing equipment is in a steady state and operating data when the processing equipment is in a fluctuating state.

[0037] Next, we will explain the details of the models generated by the model generation unit 14. Here, we will mainly explain the generation of models for instantaneous values. The model generation unit 14 reads steady-state data and fluctuating data from the data storage unit 13. Assume that both the steady-state data and the fluctuating data contain data with different waveform patterns. The model generation unit 14 generates a model for each waveform pattern by learning each data group, which consists of data with similar waveform patterns, and stores the models for each waveform pattern in the model storage unit 15. That is, the model generation unit 14 generates a model for steady-state data by learning a data group of steady-state data with similar waveform patterns, and generates a model for fluctuating data by learning a data group of fluctuating data with similar waveform patterns. Hereinafter, "similar" refers to waveforms that are similar to each other.

[0038] Figure 5 is a diagram illustrating the models generated by the model generation unit 14 of the data analysis system 1 according to Embodiment 1. Each of the models A, B, and C shown in Figure 5 is a function that is a model for instantaneous values.

[0039] Model A and Model B are models for steady-state data. Model C is a model for fluctuating data. Models A and B are models generated from data sets consisting of similar data. The data set used to generate Model A is, for example, a data set summarizing steady-state data where the instantaneous value is approximately constant, as shown in Figure 2(a). The data set used to generate Model B is, for example, a data set summarizing steady-state data where a rectangular wave of approximately the same shape is repeated, as shown in Figure 2(b). Model C is a model generated from data sets consisting of similar data from the fluctuating data. In Figure 5, "Model A + Model B + Model C + ..." represents a group of models formed by accumulating instantaneous value models generated by the model generation unit 14.

[0040] The method for determining whether data is similar to each other is arbitrary. The model generation unit 14 may determine whether data is similar to each other by combining statistics such as the mean, variance, Pearson correlation coefficient, or MIC (Maximum Information Coefficient). Alternatively, the model generation unit 14 may determine whether data is similar to each other by calculating the distance between data, such as the Euclidean distance or Manhattan distance. The model generation unit 14 may also determine whether waveforms are similar to each other by using machine learning methods such as the k-means method. The classification processing unit 12 may also determine whether data is similar to each other based on shapelets, which are multiple waveform patterns effective for time series classification.

[0041] The data analysis device 10 generates a model for steady-state data by learning from a dataset of steady-state data with similar waveform patterns, and generates a model for fluctuating data by learning from a dataset of fluctuating data with similar waveform patterns. By generating both a steady-state model and a fluctuating data model, the data analysis device 10 can make life predictions that take into account the waveform patterns of the steady-state data acquired at the time of life prediction and the waveform patterns of the fluctuating data acquired at the time of life prediction. As a result, the data analysis device 10 can make more accurate life predictions.

[0042] The model generation unit 14 generates a new model by learning the new pattern data when it finds data with a different waveform pattern from the data previously used to generate a model among the steady-state data or fluctuating data read from the data storage unit 13, i.e., when it finds data with a new pattern. The model generation unit 14 generates a new model for steady-state data as the variation in waveform patterns in steady-state data increases, and generates a new model for fluctuating data as the variation in waveform patterns in fluctuating data increases. The model generation unit 14 stores the newly generated models in the model storage unit 15. That is, the newly generated models are stored in the "Model A + Model B + Model C + ..." structure shown in Figure 5.

[0043] In this way, the data analysis device 10 generates a new model for steady-state data as the variation in waveform patterns in steady-state data increases. Furthermore, the data analysis device 10 generates a new model for fluctuating data as the variation in waveform patterns in fluctuating data increases. As a result, the data analysis device 10 can make more accurate lifespan predictions that correspond to the waveforms of the operating data acquired during lifespan prediction, as the variation in waveforms increases.

[0044] Furthermore, the model generation unit 14 updates the model for stationary data by re-learning a data group obtained by grouping stationary data whose waveform patterns are similar to each other, and updates the model for non-stationary data by re-learning a data group obtained by grouping non-stationary data whose waveform patterns are similar to each other.

[0045] The model A1 shown in FIG. 5 is the model first generated as model A. The model generation unit 14 updates the model A1 to model A2 by re-learning the model A1 using the stationary data acquired by the data acquisition unit 11 after the generation of the model A1. FIG. 5 schematically shows that the model A1 is updated to the model A2 by re-learning, and then the model A2 is updated to the model A3 by further re-learning.

[0046] The model B1 shown in FIG. 5 is the model first generated as model B. FIG. 5 schematically shows that the model B1 is updated to the model B2 by re-learning, and then the model B2 is updated to the model B3 by further re-learning. The model C1 shown in FIG. 5 is the model first generated as model C. FIG. 5 schematically shows that the model C1 is updated to the model C2 by re-learning, and then the model C2 is updated to the model C3 by further re-learning.

[0047] As described above, the data analysis device 10 updates the model for stationary data by re-learning a data group obtained by grouping stationary data whose waveform patterns are similar to each other. Also, the data analysis device 10 updates the model for non-stationary data by re-learning a data group obtained by grouping non-stationary data whose waveform patterns are similar to each other. The data analysis device 10 can improve the accuracy of the model for stationary data and the accuracy of the model for non-stationary data by re-learning. Thereby, the data analysis device 10 can perform more accurate life prediction.

[0048] Next, the processing procedure by the data analysis device 10 will be described. Here, the processing procedure by the data analysis device 10 will be described separately for the processing at the time of learning and the processing at the time of life prediction.

[0049] Figure 6 is a flowchart showing an example of the processing procedure during learning by the data analysis device 10 according to Embodiment 1.

[0050] In step S1, the data acquisition unit 11 acquires operational data indicating the state of the processing device during operation. The data acquisition unit 11 outputs the acquired operational data to the classification processing unit 12.

[0051] In step S2, the classification processing unit 12 classifies the operational data into steady-state data and variable-state data. The classification processing unit 12 stores the operational data classified as steady-state data and the operational data classified as variable-state data in the data storage unit 13.

[0052] The model generation unit 14 reads steady-state data and fluctuating data from the data storage unit 13. In step S3, the model generation unit 14 generates a model for steady-state data by learning the steady-state data and generates a model for fluctuating data by learning the fluctuating data.

[0053] In step S4, the model generation unit 14 stores the model for steady-state data and the model for variable-state data in the model storage unit 15.

[0054] The model generation unit 14 reads the cumulative value data stored in the data storage unit 13 and adds the instantaneous values ​​of the steady-state data and the instantaneous values ​​of the fluctuating data to the cumulative value data. In step S5, the model generation unit 14 generates a model for the cumulative value by learning the cumulative value.

[0055] In step S6, the model generation unit 14 stores the model for cumulative values ​​in the model storage unit 15. With this, the data analysis device 10 completes its processing during the learning phase.

[0056] Figure 7 is a flowchart showing an example of the processing procedure when predicting lifespan using the data analysis device 10 according to Embodiment 1.

[0057] In step S11, the data acquisition unit 11 acquires operational data indicating the state of the processing device during operation. The data acquisition unit 11 outputs the acquired operational data to the classification processing unit 12.

[0058] In step S12, the classification processing unit 12 classifies the operational data into steady-state data and variable-state data. The classification processing unit 12 stores the operational data classified as steady-state data and the operational data classified as variable-state data in the data storage unit 13.

[0059] In step S13, the prediction unit 16 reads out the model for steady-state data, the model for variable data, and the model for cumulative values ​​from the model storage unit 15.

[0060] The prediction unit 16 reads steady-state data, variable-state data, and cumulative values ​​from the data storage unit 13. In step S14, the prediction unit 16 predicts the lifespan of consumables by inputting steady-state data into the steady-state data model, variable-state data into the variable-state data model, and cumulative values ​​into the cumulative value model.

[0061] In step S15, the prediction unit 16 outputs information indicating the lifespan predicted in step S14 to the display device 20. With this, the data analysis device 10 completes the processing for lifespan prediction.

[0062] According to Embodiment 1, the data analysis device 10 includes a data acquisition unit 11 that acquires operational data, a classification processing unit 12 that classifies the operational data into steady-state data and variable-state data, a model generation unit 14 that generates a model for steady-state data and a model for variable-state data, and a prediction unit 16 that predicts the lifespan of parts used in the processing equipment by inputting steady-state data into the steady-state data model and variable-state data into the variable-state data model. By separating the models into those for steady-state data and those for variable-state data, the data analysis device 10 can accurately predict the lifespan based on operational data when the processing equipment is in a steady state and operational data when the processing equipment is in a variable state. As a result, the data analysis device 10 has the effect of enabling accurate lifespan prediction.

[0063] Furthermore, the model generation unit 14 generates a model for steady-state data by learning the instantaneous values ​​of steady-state data, and generates a model for fluctuating data by learning the instantaneous values ​​of fluctuating data. The model generation unit 14 generates a model for cumulative values ​​by learning the cumulative values, which are the cumulative values ​​obtained by accumulating the instantaneous values ​​of steady-state data and the instantaneous values ​​of fluctuating data. The prediction unit 16 predicts the lifespan by inputting the instantaneous values ​​of steady-state data into the steady-state data model, the instantaneous values ​​of fluctuating data into the fluctuating data model, and the cumulative values ​​into the cumulative value model. The data analysis device 10 generates models for instantaneous values, such as the steady-state data model and the fluctuating data model, and predicts the lifespan based on each model, thereby enabling lifespan prediction that takes into account the trend of instantaneous values ​​and lifespan prediction that takes into account the trend of cumulative values. As a result, the data analysis device 10 can make more accurate lifespan predictions.

[0064] Furthermore, the model generation unit 14 generates a model for steady-state data by learning from a data set of steady-state data whose waveform patterns are similar to each other, and generates a model for fluctuating data by learning from a data set of fluctuating data whose waveform patterns are similar to each other. The data analysis device 10 can then predict the lifespan by taking into account the waveform patterns of the steady-state data acquired at the time of life prediction and the waveform patterns of the fluctuating data acquired at the time of life prediction. As a result, the data analysis device 10 can make more accurate lifespan predictions.

[0065] Furthermore, the model generation unit 14 generates new models for steady-state data as the variation in waveform patterns in steady-state data increases, and generates new models for fluctuating data as the variation in waveform patterns in fluctuating data increases. The data analysis device 10 can increase the number of models for each waveform as the variation in waveforms increases. As a result, the data analysis device 10 can make more accurate lifespan predictions according to the waveforms of the operating data acquired during lifespan prediction.

[0066] Furthermore, the model generation unit 14 updates the model for steady-state data by relearning a data set of steady-state data whose waveform patterns are similar to each other, and updates the model for fluctuating data by relearning a data set of fluctuating data whose waveform patterns are similar to each other. The data analysis device 10 can improve the accuracy of both the steady-state data model and the fluctuating data model through relearning. As a result, the data analysis device 10 can make more accurate life predictions.

[0067] According to Embodiment 1, the data analysis system 1 comprises a data analysis device 10 and an output device that outputs information indicating the lifespan predicted by the data analysis device 10. This makes it possible for the data analysis system 1 to show the user the accurate lifespan predicted by the data analysis device 10.

[0068] Embodiment 2. Figure 8 shows an example of the configuration of the data analysis system 1A according to Embodiment 2. In Embodiment 2, the same reference numerals are used for the same components as in Embodiment 1, and the configuration that differs from Embodiment 1 will be described in detail.

[0069] The data analysis system 1A shown in Figure 8 comprises a data analysis device 10A that analyzes the operating data of a processing machine, a display device 20 that displays the information, and a learning device 30 that generates a model for predicting the lifespan of consumables through learning. In Embodiment 2, the data analysis device 10A functions as an inference device that predicts the lifespan of consumables using the model. The data analysis device 10A and the display device 20 are connected to each other so as to be able to communicate with each other. The data analysis device 10A and the learning device 30 are connected to each other so as to be able to communicate with each other.

[0070] The data analysis device 10A, like the data analysis device 10 shown in Figure 1, includes a data acquisition unit 11, a classification processing unit 12, a data storage unit 13, a model storage unit 15, and a prediction unit 16. The data analysis device 10A further includes a model acquisition unit 17 that acquires the model generated by the learning device 30.

[0071] The learning device 30 includes a learning data acquisition unit 31 that acquires steady-state data and fluctuating data, and a model generation unit 32 that generates a model for steady-state data by learning steady-state data and generates a model for fluctuating data by learning fluctuating data.

[0072] The data acquisition unit 11 of the data analysis device 10A acquires operational data that indicates the state of the processing device during operation. The classification processing unit 12 stores the operational data, which has been classified into steady-state data and variable-state data, in the data storage unit 13.

[0073] The learning data acquisition unit 31 of the learning device 30 acquires learning data, which consists of steady-state data and fluctuating data, by reading them from the data storage unit 13 of the data analysis device 10A. The learning data acquisition unit 31 outputs the acquired steady-state data and fluctuating data to the model generation unit 32.

[0074] The model generation unit 32 generates a model for steady-state data by learning instantaneous values ​​of steady-state data, and generates a model for fluctuating data by learning instantaneous values ​​of fluctuating data. The model generation unit 32 generates a model for steady-state data and a model for fluctuating data, similar to the model generation unit 14 of the data analysis device 10 according to Embodiment 1. The model generation unit 32 outputs the steady-state data model and the fluctuating data model, which were generated by learning instantaneous values, to the data analysis device 10A.

[0075] The data storage unit 13 stores cumulative value data, which is a time series of cumulative values ​​obtained by accumulating the instantaneous values ​​of steady-state data and instantaneous values ​​of fluctuating data. The learning data acquisition unit 31 reads the cumulative value data from the data storage unit 13. The model generation unit 32 accumulates the instantaneous values ​​of steady-state data and instantaneous values ​​of fluctuating data into the cumulative value data. The model generation unit 32 generates a model for cumulative values ​​by learning the cumulative values, similar to the model generation unit 14 of the data analysis device 10 according to Embodiment 1. The model generation unit 32 outputs the model for cumulative values ​​generated by learning the cumulative values ​​to the data analysis device 10A.

[0076] The model acquisition unit 17 of the data analysis device 10A acquires models for steady-state data, models for fluctuating data, and models for cumulative values ​​sent from the learning device 30. The model acquisition unit 17 stores the models for steady-state data, models for fluctuating data, and models for cumulative values ​​in the model storage unit 15.

[0077] When performing life prediction based on operational data, the prediction unit 16 predicts the life by inputting instantaneous values ​​of steady-state data into a model for steady-state data, instantaneous values ​​of fluctuating data into a model for fluctuating data, and cumulative values ​​of operational data into a model for cumulative values, similar to the prediction unit 16 of the data analysis device 10 according to Embodiment 1. The prediction unit 16 outputs information indicating the predicted life to the display device 20.

[0078] Next, the processing procedure by the data analysis device 10A will be described. The data analysis device 10A according to Embodiment 2 performs the same processing as steps S1, S2, S4, and S6 shown in Figure 6 during learning. The processing similar to steps S3 and S5 shown in Figure 6 is performed by the learning device 30.

[0079] The data analysis device 10A according to Embodiment 2 performs the same processing as steps S11-S15 shown in Figure 7 when predicting the lifespan.

[0080] According to Embodiment 2, the learning device 30 includes a learning data acquisition unit 31 that acquires steady-state data and fluctuating data, and a model generation unit 32 that generates a model for steady-state data and a model for fluctuating data. The learning device 30 can generate a model that enables accurate lifetime prediction based on steady-state data and fluctuating data.

[0081] The data analysis device 10A includes a data acquisition unit 11 that acquires operational data, a classification processing unit 12 that classifies the operational data into steady-state data and variable-state data, and a prediction unit 16 that predicts the lifespan of parts used in the processing equipment by inputting steady-state data into a model for steady-state data and variable-state data into a model for variable-state data. By separating the models into those for steady-state data and those for variable-state data, the data analysis device 10A can accurately predict the lifespan based on operational data when the processing equipment is in a steady state and operational data when the processing equipment is in a variable state. As a result, the data analysis device 10A has the effect of enabling accurate lifespan prediction. Furthermore, the data analysis system 1A can show the user the accurate lifespan predicted by the data analysis device 10A.

[0082] Next, the hardware that realizes the data analysis devices 10 and 10A according to Embodiments 1 and 2 will be described. The data analysis devices 10 and 10A are realized by using a processing circuit. The processing circuit is a circuit in which a processor executes software. The processing circuit is, for example, the control circuit shown in Figure 9.

[0083] Figure 9 shows an example configuration of a control circuit 40 according to Embodiment 1 or 2. The control circuit 40 comprises an input unit 41, a processor 42, a memory 43, and an output unit 44. The input unit 41 is an interface circuit that receives data from outside the control circuit 40 and provides it to the processor 42. The output unit 44 is an interface circuit that sends data from the processor 42 or the memory 43 to the outside of the control circuit 40.

[0084] The functions of the processing units in the data analysis device 10 shown in Figure 1, namely the classification processing unit 12, the model generation unit 14, and the prediction unit 16, are realized by software, firmware, or a combination of software and firmware. The functions of the processing units in the data analysis device 10A shown in Figure 8, namely the classification processing unit 12 and the prediction unit 16, are realized by software, firmware, or a combination of software and firmware. The software or firmware is written as a program and stored in memory 43.

[0085] The processing circuit realizes the processing unit of the data analysis devices 10 and 10A by having the processor 42 read and execute a program stored in the memory 43. In other words, the processing circuit includes a memory 43 for storing a program that will ultimately be executed by the data analysis devices 10 and 10A. The program stored in the memory 43 is a data analysis program that causes the computer to execute the processing procedures and methods of the data analysis devices 10 and 10A. The data acquisition unit 11 and the model acquisition unit 17 of the data analysis devices 10 and 10A are realized using the input unit 41. The data storage unit 13 and the model storage unit 15 of the data analysis devices 10 and 10A are realized using the memory 43. The memory 43 is also used as temporary memory when the processor 42 executes various processes.

[0086] The processor 42 is a CPU (Central Processing Unit). The processor 42 may also be a central processing unit, processing unit, arithmetic unit, microprocessor, microcomputer, processor, or DSP (Digital Signal Processor). The memory 43 may be, for example, a non-volatile or volatile semiconductor memory such as RAM (Random Access Memory), ROM (Read Only Memory), flash memory, EPROM (Erasable Programmable Read Only Memory), EEPROM® (Electrically Erasable Programmable Read Only Memory), magnetic disk, flexible disk, optical disk, compact disk, minidisc, or DVD (Digital Versatile Disc).

[0087] The display device 20 of the data analysis system 1,1A is implemented with a hardware configuration similar to that shown in Figure 9. The hardware configuration for implementing the display device 20 includes a display device for displaying information. The display device is, for example, an LCD (Liquid Crystal Display) or an organic EL (Electro-Luminescence) display.

[0088] The learning device 30 of the data analysis system 1A is implemented with a configuration similar to the hardware configuration shown in Figure 9. The function of the model generation unit 32 is realized by the processor 42 reading and executing a program stored in the memory 43. The learning data acquisition unit 31 is realized by using the input unit 41.

[0089] The processing circuits of the data analysis devices 10, 10A, the display device 20, and the learning device 30 may each be dedicated circuits. Dedicated circuits include single circuits, composite circuits, programmed processors, parallel programmed processors, ASICs (Application Specific Integrated Circuits), FPGAs (Field Programmable Gate Arrays), or circuits combining these.

[0090] Each component of the data analysis system 1,1A does not need to be physically configured as shown in the diagram. The specific forms of distribution and integration of each component are not limited to those shown in the diagram. Each component may be configured functionally or physically distributed in any unit, or it may be configured as an integrated unit.

[0091] At least a portion of the data analysis systems 1 and 1A may be hosted on a cloud server. A cloud server is a server built in a cloud environment that includes computing resources provided on a cloud service platform. For example, the learning device 30 of the data analysis system 1A may be hosted on a cloud server.

[0092] The configurations of the data analysis devices 10, 10A or data analysis systems 1, 1A according to Embodiments 1 and 2 may be applied to devices or systems other than those that predict the lifespan of parts used in processing equipment. For example, the configurations of the data analysis devices 10, 10A or data analysis systems 1, 1A according to Embodiments 1 and 2 may be applied to devices or systems that predict the quality of processed products manufactured by processing equipment. For example, in a device or system that predicts the quality of processed products manufactured by electrical discharge machining equipment, voltage value data of electrical discharge machining may be acquired as operational data, and a model for predicting the quality of processed products may be generated by learning the voltage value data. In addition, the configurations of the data analysis devices 10, 10A or data analysis systems 1, 1A according to Embodiments 1 and 2 may be applied to devices or systems that predict the battery life of devices that are powered by charging, such as EVs (Electric Vehicles).

[0093] The configurations shown in each of the embodiments described above are examples of the content of this disclosure. The configurations of each embodiment can be combined with other known technologies. The configurations of each embodiment may be combined with each other as appropriate. It is possible to omit or modify parts of the configurations of each embodiment without departing from the gist of this disclosure.

[0094] 1, 1A Data analysis system, 10, 10A Data analysis device, 11 Data acquisition unit, 12 Classification processing unit, 13 Data storage unit, 14, 32 Model generation unit, 15 Model storage unit, 16 Prediction unit, 17 Model acquisition unit, 20 Display device, 30 Learning device, 31 Learning data acquisition unit, 40 Control circuit, 41 Input unit, 42 Processor, 43 Memory, 44 Output unit.

Claims

1. A data analysis device comprising: a data acquisition unit that acquires operational data indicating the state of a processing device; a classification processing unit that classifies the operational data into steady-state data, in which the value is constant or the value fluctuates regularly, and fluctuating data, in which the value fluctuates irregularly; a model generation unit that generates a model for steady-state data by learning the steady-state data and generates a model for fluctuating data by learning the fluctuating data; and a prediction unit that predicts the lifespan of parts used in the processing device by inputting the steady-state data into the steady-state data model and inputting the fluctuating data into the fluctuating data model.

2. The data analysis apparatus according to claim 1, characterized in that the model generation unit generates a model for steady-state data by learning the instantaneous values ​​of the steady-state data, generates a model for fluctuating data by learning the instantaneous values ​​of the fluctuating data, and further generates a model for cumulative values ​​by learning cumulative values ​​which are the sum of the instantaneous values ​​of the steady-state data and the instantaneous values ​​of the fluctuating data, and the prediction unit predicts the lifespan by inputting the instantaneous values ​​of the steady-state data into the model for steady-state data, inputting the instantaneous values ​​of the fluctuating data into the model for fluctuating data, and further inputting the cumulative values ​​into the model for cumulative values.

3. The data analysis apparatus according to claim 2, characterized in that the model generation unit generates a model for steady-state data by learning a data set of steady-state data whose waveform patterns are similar to each other, and generates a model for fluctuating data by learning a data set of fluctuating data whose waveform patterns are similar to each other.

4. The data analysis apparatus according to claim 3, characterized in that the model generation unit generates a new model for the steady-state data in response to an increase in the variation of waveform patterns in the steady-state data, and generates a new model for the fluctuating data in response to an increase in the variation of waveform patterns in the fluctuating data.

5. The data analysis apparatus according to claim 3 or 4, characterized in that the model generation unit updates the model for steady-state data by relearning a data group of steady-state data whose waveform patterns are similar to each other, and updates the model for fluctuating data by relearning a data group of fluctuating data whose waveform patterns are similar to each other.

6. A data analysis system comprising: a data analysis device according to any one of claims 1 to 5; and an output device that outputs information indicating the lifespan predicted by the data analysis device.

7. A data analysis method characterized by comprising the steps of: acquiring operational data indicating the state of a processing device; classifying the operational data into steady-state data, where the value is constant or the value fluctuates regularly, and fluctuating data, where the value fluctuates irregularly; generating a model for steady-state data by learning the steady-state data, and generating a model for fluctuating data by learning the fluctuating data; and predicting the lifespan of the parts used in the processing device by inputting the steady-state data into the steady-state data model and inputting the fluctuating data into the fluctuating data model.

8. A data analysis program characterized by causing a computer to perform the following steps: acquiring operational data indicating the state of a processing device; classifying the operational data into steady-state data, where the value is constant or the value fluctuates regularly, and variable-state data, where the value fluctuates irregularly; generating a model for steady-state data by learning the steady-state data, and generating a model for variable-state data by learning the variable-state data; and predicting the lifespan of the parts used in the processing device by inputting the steady-state data into the steady-state data model and inputting the variable-state data into the variable-state data model.

9. A learning device comprising: a learning data acquisition unit that acquires steady-state data, which is a constant value or whose values ​​fluctuate regularly, and fluctuating data, which is a variable value, from among the operating data that indicates the state of the processing equipment; and a model generation unit that generates a model for steady-state data by learning the steady-state data and generates a model for fluctuating data by learning the fluctuating data.

10. An inference device comprising: a data acquisition unit that acquires operational data indicating the state of a processing device; a classification processing unit that classifies the operational data into steady-state data, in which the value is constant or the value fluctuates regularly, and fluctuating data, in which the value fluctuates irregularly; and a prediction unit that predicts the lifespan of the parts used in the processing device by inputting the steady-state data into a model for steady-state data generated by learning the steady-state data, and inputting the fluctuating data into a model for fluctuating data generated by learning the fluctuating data.

Images