Analysis device

Abstract

The analysis device of the present disclosure includes a data acquisition unit that acquires data detected by an industrial machine, an operation state extraction unit that extracts data detected by the industrial machine in operation from the data, an annotation unit that generates a plurality of data set groups to which a plurality of data sets cut from the extracted data in operation at predetermined reference points are given annotations indicating an operation state of the industrial machine at predetermined reference points, a feature quantity extraction unit that extracts feature quantities of data included in the data sets, a learning unit that generates decision tree models for the plurality of data set groups, respectively, and a display unit that sorts and displays the generated decision tree models in accordance with a prediction accuracy of the annotations based on predetermined training data.

Description

Technical Field

[0001] This invention relates to an analytical apparatus. Background Technology

[0002] In factories and other manufacturing sites, devices are introduced to monitor and manage the operating status of industrial machinery such as robots and machine tools installed on the production line.

[0003] As a device for managing the operating status of industrial machinery, for example, Patent Document 1 discloses the following device: when producing a product, it acquires data on normal production times and abnormal production times when an anomaly occurs in the produced product; based on the acquired data, it determines the correlation between the anomaly occurring in the produced product and the data, and selects data that is effective in predicting the anomaly.

[0004] Existing technical documents

[0005] Patent documents

[0006] Patent Document 1: Japanese Patent Application Publication No. 2018-116545 Summary of the Invention

[0007] The problem that the invention aims to solve

[0008] In the manufacturing process, it's desirable to understand what warning signs are present in the data collected before an anomaly occurs. To detect these signs, it's necessary to compare data collected during normal periods with data collected before the anomaly (defective processing, alarm activation) to identify any pre-existing indicators in the data.

[0009] However, such premonitory signs in the data sometimes occur only just before the anomaly, but can also appear several days in advance. Therefore, even comparing normal data with data before the anomaly requires comparing only specific portions of the massive dataset, placing an excessive burden on users. Furthermore, comparing such vast amounts of data to detect premonitory signs is inherently difficult. This kind of work is necessary not only in analyzing normal / abnormal data but also in analyzing the signs exhibited by data during various phenomena occurring in the manufacturing process.

[0010] Therefore, techniques are needed to assist in the analysis of data that indicate phenomena occurring on the manufacturing site.

[0011] Methods for solving problems

[0012] The operational data analysis apparatus disclosed herein extracts feature quantities from operational data obtained from industrial machinery. Then, it assists the user in analyzing the data by displaying the differences in the extracted feature quantities using a decision tree, thereby solving the aforementioned problem.

[0013] Furthermore, one aspect of this disclosure is an analysis apparatus that assists in the analysis of data collected from industrial machinery, comprising: a data acquisition unit that acquires the data detected by the industrial machinery; an operation state extraction unit that extracts data detected by the industrial machinery during operation from the data; an annotation unit that generates multiple dataset groups, each dataset group assigning annotations representing the operation state of the industrial machinery to multiple datasets extracted from the operation data based on a predetermined criterion; a feature extraction unit that extracts feature quantities of the data contained in each of the datasets; a learning unit that, for each of the multiple dataset groups, generates a decision tree model by treating the feature quantities as independent variables and the annotations representing the operation state of the industrial machinery as dependent variables; and a display unit that sorts and displays the multiple decision tree models according to the prediction accuracy of the annotations based on predetermined training data.

[0014] Invention Effects

[0015] According to one aspect of this disclosure, the conditions under which the state of industrial machinery (normal operation, before anomaly occurs, etc.) differs when the collected data is detected can be visualized as a decision tree model, making it easy to grasp the signs of state changes shown by the data. Attached Figure Description

[0016] Figure 1 This is a schematic hardware structure diagram of an analysis device according to one embodiment of the present invention.

[0017] Figure 2 This is a block diagram illustrating the general functions of an analysis apparatus according to one embodiment of the present invention.

[0018] Figure 3 This is a diagram illustrating the method of data set extraction and the method of assigning annotations to represent the operational states of industrial machinery.

[0019] Figure 4 This is a diagram showing an example of a display unit.

[0020] Figure 5 This is a diagram showing other display examples of the display section. Detailed Implementation

[0021] The following is related to the appendix. Figure 1 This section describes the embodiments of the present invention.

[0022] Figure 1This is a schematic hardware structure diagram showing the main parts of the analysis device according to one embodiment of the present invention. The analysis device 1 of the present invention can be installed, for example, on a control device for controlling industrial machinery 3. Furthermore, the analysis device 1 of the present invention can be installed on a personal computer arranged alongside the industrial machinery 3, a personal computer connected to the industrial machinery 3 via a wired / wireless network, a unit computer, a fog computer 6, a cloud server 7, or a similar computer. In this embodiment, an example is shown where the analysis device 1 is installed on a personal computer connected to the industrial machinery 3 via a network.

[0023] The CPU 11 of the analysis device 1 in this embodiment is a processor that controls the analysis device 1 as a whole. The CPU 11 reads the system program stored in the ROM 12 via the bus 22 and controls the analysis device 1 as a whole according to the system program. The RAM 13 temporarily stores temporary calculation data, display data, and various data input from the outside.

[0024] The non-volatile memory 14 is composed of, for example, a battery-backed memory (not shown) or an SSD (Solid State Drive), and maintains its storage state even when the power supply to the analysis device 1 is disconnected. Data read from the external device 72 via the interface 15, data input via the input device 71, and data obtained from the industrial machinery 3 are stored in the non-volatile memory 14. The data stored in the non-volatile memory 14 can also be expanded in the RAM 13 during execution / use. Furthermore, various system programs, such as known parsing programs, are pre-written into the ROM 12.

[0025] Interface 15 is an interface for connecting to external devices 72 such as the CPU 11 of the analysis device 1 and a USB device. It can read, for example, pre-stored image data of the workpiece from the external device 72. Furthermore, setting data edited within the analysis device 1 can be stored in an external storage unit via the external device 72.

[0026] Interface 20 is used to connect the CPU 11 of the analysis device 1 to a wired or wireless network 5. Industrial machinery 3, fog computers 6, cloud servers 7, etc. are connected to the network 5, and they exchange data with the analysis device 1.

[0027] Industrial machinery 3 includes turning machines, electrical discharge machines, robots, and handling machines installed at the manufacturing site. The CPU 11 of the analysis device 1 obtains various data detected by the industrial machinery 3 during workpiece manufacturing operations, such as motor current / voltage, position and speed of moving parts, acceleration, images and sounds indicating processing status, temperature of the machine's periphery and various parts, signal status of each part, setting status of each part of the industrial machinery 3, operator input of the quality of the industrial machinery 3's actions, product processing results, and changeover adjustment information.

[0028] In the display device 70, data read into the memory and data obtained as a result of executing programs are output and displayed via the interface 17. In addition, the input device 71, which consists of a keyboard and indicator devices, transmits instructions and data based on the operator's operation to the CPU 11 via the interface 18.

[0029] Figure 2 The functions of the analysis apparatus 1 according to one embodiment of the present invention are illustrated in a schematic block diagram. The functions of the analysis apparatus 1 according to this embodiment are shown below. Figure 1 The analysis device 1 shown is implemented by having a CPU 11 that executes system programs and controls the operation of each part of the analysis device 1.

[0030] The analysis device 1 of this embodiment includes a data acquisition unit 110, an operation state extraction unit 120, a change point detection unit 125, an annotation unit 130, a feature extraction unit 135, a learning unit 145, and a display unit 155. Furthermore, the analysis device 1's RAM 13 or non-volatile memory 14 includes a pre-prepared area for storing data detected by the industrial machinery 3 (i.e., an operation data storage unit 115), a pre-prepared area for storing feature extraction patterns for feature extraction (i.e., a feature extraction pattern storage unit 140), an area for storing models generated as a result of learning (i.e., a model storage unit 150), and an area for storing training data used in training the generated model (i.e., a training data storage unit 160).

[0031] The data acquisition unit 110 acquires data detected by the industrial machinery 3. This data includes, for example, current / voltage of the motor, position and speed of the moving parts, acceleration, images and sounds indicating processing status, temperature of the machinery's periphery and various parts, signal status of each part, setting status of each part of the industrial machinery 3, operator-inputted quality of the industrial machinery 3's operation, product processing results, and changeover adjustment information. The data acquisition unit 110 can also acquire data from the industrial machinery 3 via, for example, network 5. Additionally, pre-recorded data detected by the industrial machinery 3 can be acquired from external device 72. The data acquisition unit 110 stores the acquired data in the operation data storage unit 115.

[0032] The operation status extraction unit 120 extracts signal data indicating whether the industrial machine 3 is operating from the data detected by the industrial machine 3 stored in the operation data storage unit 115, and determines the portion of the data detected that the industrial machine 3 is operating. Normally, during the operation of the industrial machine 3, a predetermined signal indicating operation is turned on. The operation status extraction unit 120 extracts data representing the range in which this signal is turned on, as a set of data detected during the operation of the industrial machine 3.

[0033] The change point detection unit 125 detects change points in the data detected by the industrial machine 3 stored in the operation data storage unit 115, indicating a change in the trend of data variation. Examples of change points include changes that can be extracted from the data, such as changes in the processing program, tooling, or offset values; changes that can be extracted from signal data, such as rapid temperature changes or a decrease in the torque of the motor of the industrial machine 3; and changes in human operation, such as changes in changeover adjustment time or changes in the execution time of the processing program. Before and after the change point, no trend similar to the change in the detected data is usually observed. Therefore, the change point detection unit 125 detects change points as a benchmark for data analysis. In most cases, the data sets detected between change points are analyzed as a set.

[0034] The annotation unit 130 assigns annotations based on the data detected during operation of the industrial machine 3 extracted by the operation status extraction unit 120, specifying the operating state of the industrial machine 3 at the time the data was detected. Examples of annotations assigned by the annotation unit 130 for the operating state include "normal operation" and "before an abnormality occurs." Other examples of annotations assigned by the annotation unit 130 for the operating state include input from the operator regarding the operation of the industrial machine 3, such as "before a machining defect occurs." Furthermore, regarding abnormalities of the industrial machine 3, more detailed annotations can be assigned based on alarm signals detected before tool-related alarms or spindle-related alarms occur. Regarding machining defects, more detailed annotations corresponding to the defect can be assigned, such as before workpiece defects or before large surface roughness occurs. These annotations can be automatically assigned based on the content of the operating data or manually assigned by the user.

[0035] When a predetermined phenomenon is detected, it is not possible to directly know when the symptom was detected in the data. Therefore, the annotation unit 130, for example, after capturing the dataset each time the operation is run, generates a dataset group that traces back to the time the predetermined phenomenon was detected and assigns an annotation to the dataset from the first run, indicating the time before the phenomenon occurred. Conversely, it generates a dataset group that traces back to the time the predetermined phenomenon was detected and assigns an annotation to the dataset from the second run, indicating the time before the phenomenon occurred. Furthermore, it generates a dataset group that traces back to the time the predetermined phenomenon was detected and assigns an annotation to the dataset from the third run, indicating the time before the phenomenon occurred. In this way, the annotation unit 130 generates a predetermined set of n dataset groups. The number of dataset groups to be generated can be preset based on user specifications. The scope of the dataset capture is not limited to the number of runs. For example, it can also trace back to the time the predetermined phenomenon was detected, such as 30 minutes ago, 1 hour ago, etc., based on time. Additionally, the scope of the dataset capture can also use the change points detected by the change point detection unit 125. The scope of this dataset can be set based on user specifications.

[0036] use Figure 3 Example of operation of the operation status extraction unit 120 and the annotation unit 130. Figure 3 This is a diagram illustrating an example of data detected from industrial machinery 3 when a workpiece defect is detected. Furthermore, in this example, the data is segmented according to the number of times industrial machinery 3 operates. When a defect is detected... Figure 3 When the data is illustrated, the operation status extraction unit 120 refers to the operation signal within the detected data and extracts the data set of the interval where the operation signal is on as the operation detection dataset. Figure 3 In this example, at least three datasets 311, 312, and 313 are extracted. In these datasets, the operator reports that a workpiece processed when dataset 311 was detected suffered a machining defect. In this case, the annotation unit 130 generates a dataset group that assigns an annotation such as "before the machining defect occurred" to dataset 311 extracted by the operation status extraction unit 120, and assigns an annotation such as "normal operation" to datasets 312 and 313 respectively. Additionally, the annotation unit 130 generates a dataset group that assigns an annotation such as "before the machining defect occurred" to datasets 311 and 312 extracted by the operation status extraction unit 120, and assigns an annotation such as "normal operation" to dataset 313. Thus, the annotation unit 130 generates multiple dataset groups.

[0037] The feature extraction unit 135 extracts predetermined features for each data point in the annotated dataset. Examples of predetermined features include, for instance, the mean and slope of change, variance, maximum change, Fourier transform value, and outlier. The feature extraction unit 135 can extract multiple features from a single dataset. The feature extraction pattern storage unit 140 stores the features that are pre-extracted for each data point in the dataset. Referring to the feature extraction pattern storage unit 140, the feature extraction unit 135 determines the features to be extracted for each data point.

[0038] For each dataset group generated by the annotation unit 130, the learning unit 145 generates a decision tree model that uses the features extracted by the feature extraction unit 135 as independent variables and the annotations of the action state of the industrial machinery 3 assigned by the annotation unit 130 as dependent variables. The decision tree learning algorithm used by the learning unit 145 can be a well-known algorithm such as ID3, CART, or C4.5. The hyperparameters of the decision tree (tree depth, number of nodes, etc.) can be preset. Furthermore, the algorithm for generating the decision tree model is well-known, therefore a detailed description of the generation process is omitted in this specification. Each decision tree model generated by the learning unit 145 is stored in the model storage unit 150.

[0039] Display unit 155 evaluates the decision tree model generated by learning unit 145 using training data and displays the evaluation result on display device 70. Training data storage unit 160 may store, for example, multiple training data sets pre-generated by users. In this case, the training data is data for which the user has correctly annotated a predetermined dataset. Alternatively, the dataset annotated by annotation unit 130 may also be used as training data. Display unit 155 uses each decision tree model generated by learning unit 145 to predict annotations for the operating state of industrial machinery 3 for each training data set stored in training data storage unit 160. Then, it sets the predicted annotations as correct if they match the annotations assigned to the training data, and sets them as incorrect in other cases, calculating the overall accuracy of the decision tree model for the training data. Finally, it displays the result as an evaluation result on display device 70.

[0040] The display unit 155 can display parameters such as the accuracy and dataset truncation method of each decision tree model generated by the learning unit 145 on the display device 70. Figure 4 This is a display example showing the accuracy of each model in the display unit 155. For example... Figure 4 As illustrated, the display unit 155 displays, in a manner that allows for understanding of the correspondence, how the dataset is truncated for each model, how annotations are applied, and the accuracy when predictions are made on the training data. By displaying this information, it becomes immediately clear what data truncating methods and annotation techniques were used to generate a more accurate decision tree model.

[0041] In addition, the display unit 155 can also automatically display detailed information about the decision tree model and its accuracy as an evaluation result based on user specifications or for decision tree models with high accuracy. Figure 5 This is an example of the display of the evaluation results of the decision tree model by the display unit 155. Figure 5 The example illustrates generating a decision tree model for a dataset labeled with normal operation and a state before an anomaly occurs. Using a decision tree learning algorithm, the maximum spindle torque and the average servo motor temperature are treated as independent variables affecting the prediction results. By determining each independent variable, the model predicts that each dataset is either in a normal operation or a state before an anomaly occurs. The display unit 155 uses this decision tree model to predict the state based on pre-prepared training data. The prediction result is determined by... Figure 5 The pie chart is represented in [the context of the image / image]. Figure 5In the illustrated decision tree model, the accuracy is poor when it can be immediately determined that the maximum spindle torque is less than 100 N·m and the servo motor temperature is above 29°C. Furthermore, it is evident that the accuracy remains poor even when the maximum spindle torque is above 100 N·m and the average servo motor temperature is less than 26°C.

[0042] In addition, Figure 5 When selecting nodes in a decision tree model in the display, the datasets classified under that node can also be displayed simultaneously. This side-by-side display makes it easier for users to interpret the classification rules within each decision tree model.

[0043] In this way, while observing the decision tree model displayed on the display unit 155 and the prediction results for the training data, the user can change the method of truncating the dataset, or append / delete features stored in the feature extraction mode storage unit 140 that are the objects of extraction. By repeating this process, a decision tree model that makes more appropriate predictions for the training data can be obtained. Regarding the decision tree model that makes appropriate predictions, by checking how annotations are given, the user can easily determine at what time signs of state change are displayed.

[0044] The analysis device 1 of this embodiment, equipped with the aforementioned structure, can conditionally visualize data indicating changes in the state (normal operation, pre-abnormality, etc.) of industrial machinery 3 as a decision tree model. By referring to this decision tree model, users can easily grasp the signs of state changes exhibited by the data. By visualizing the decision tree models generated for each data set extraction method, it is possible to determine when the various signs of state changes appearing in the data begin to appear.

[0045] The embodiments of the present invention have been described above, but the present invention is not limited to the examples of the described embodiments and can be implemented in various ways by applying appropriate modifications.

[0046] Explanation of reference numerals in the attached figures

[0047] 1. Analytical apparatus

[0048] 3. Industrial machinery

[0049] 5. Network

[0050] 6 Fog Computer

[0051] 7 cloud servers

[0052] 11 CPUs

[0053] 12ROM

[0054] 13 RAM

[0055] 14. Non-volatile memory

[0056] 15, 17, 18, 20 interfaces

[0057] 22 bus,

[0058] 70 display devices

[0059] 71 Input Device

[0060] 72 external devices

[0061] 110 Data Acquisition Department

[0062] 115 Operation Data Storage Department

[0063] 120 Operation Status Extraction Unit

[0064] 125 Change Point Detection Department

[0065] 130. Notes Section

[0066] 135 Feature Extraction Unit

[0067] 140 Feature Extraction Pattern Storage Unit

[0068] 145 Study Department

[0069] 150 model storage unit

[0070] 155 Display Unit

[0071] 160 training data storage unit

[0072] Data sets 311-313.

Claims

1. An analytical apparatus for assisting in the analysis of data collected from industrial machinery, characterized in that, The analytical device has the following features: The data acquisition unit acquires the data detected by the industrial machinery. The operation status extraction unit extracts data from the data indicating that the industrial machinery is in operation. The annotation section extracts multiple datasets from the extracted operational data based on a predetermined benchmark, and generates multiple dataset groups by assigning annotations to the extracted datasets based on the predetermined benchmark. The annotations indicate the operational status of the industrial machinery. The feature extraction unit extracts the feature quantities of the data contained in each of the datasets; The learning department generates a decision tree model for each of the multiple dataset groups, using the feature quantity as an independent variable and the annotation representing the operating state of the industrial machinery as a dependent variable. as well as The display unit sorts and displays multiple decision tree models based on the accuracy of predictions made using the annotations from predetermined training data.

2. The analytical apparatus according to claim 1, characterized in that, The predetermined benchmark for extracting the dataset from the extracted operational data is the number of times or the duration of operation of the industrial machinery.

3. The analytical apparatus according to claim 1, characterized in that, The analysis device also includes a change point detection unit, which detects the change points in the operation of the industrial machinery based on the data acquired by the data acquisition unit. The predetermined benchmark for extracting the dataset from the extracted, operational data is the change point detected by the change point detection unit.

4. The analytical apparatus according to claim 1, characterized in that, Multiple datasets were annotated at different times as the operating state of the industrial machinery had changed.

5. The analytical apparatus according to claim 1, characterized in that, The display unit shows the decision tree model and displays the accuracy based on the training data for each terminal node of the decision tree model.

6. The analytical apparatus according to claim 1, characterized in that, When the terminal node of the decision tree model is selected, the display unit displays the training data classified at that terminal node.

Images