Equipment diagnostic system, learning device, and method for generating trained models
The equipment diagnostic system uses datasets from normal and similar equipment states to generate a trained model for anomaly detection, addressing the need for accurate diagnosis without inducing abnormalities, thus maintaining efficiency and improving detection accuracy.
Patent Information
- Application Number
- JP2023580154
- Authority / Receiving Office
- JP · JP
- Patent Type
- Patents
- Current Assignee / Owner
- Priority Date
- 2022-02-09
- Filing Date
- 2023-01-24
- Publication Date
- 2026-11-11
- Estimated Expiration
- 2043-01-24
AI Technical Summary
Existing anomaly detection systems require inducing abnormalities in equipment for threshold setting, leading to potential inaccuracies due to user experience-based thresholds, which can reduce detection accuracy.
An equipment diagnostic system that utilizes datasets from normal and similar equipment states to generate a trained model for anomaly detection without causing intentional abnormalities, using equipment-independent features for accurate diagnosis.
Enables accurate equipment diagnosis without stopping the production line, maintaining high operational efficiency and improving anomaly detection accuracy beyond user experience-based thresholds.
Smart Images

Figure 0007948975000001 
Figure 0007948975000002 
Figure 0007948975000003
Abstract
Description
[Technical Field]
[0001] This disclosure relates to an equipment diagnostic system, a learning device, a trained model, and a method for generating a trained model. [Background technology]
[0002] Conventionally, anomaly detection devices are known that use the detection values of sensors installed in equipment to diagnose whether or not an abnormality has occurred in the equipment. For example, the anomaly detection device described in Japanese Patent Application Publication No. 2020-104257 (Patent Document 1) uses the detection results of a current sensor installed in the processing machine that is the target of anomaly detection to detect an abnormality in the processing machine.
[0003] More specifically, the abnormality detection device described in Japanese Patent Publication No. 2020-104257 (Patent Document 1) determines that an abnormality has occurred in the processing machine when the measured value of the current sensor exceeds a preset threshold. As methods for setting the threshold, Patent Document 1 describes a method of determining the threshold based on the actual measured value of the current sensor when an abnormality is actually occurring in the processing machine, and a method of determining the threshold based on the experience of the user who uses the processing machine. [Prior art documents] [Patent Documents]
[0004] [Patent Document 1] Japanese Patent Publication No. 2020-104257 [Overview of the Initiative] [Problems that the invention aims to solve]
[0005] However, in order to obtain actual current sensor readings when an abnormality is actually occurring in the processing machine, it is necessary to intentionally induce an abnormality in the processing machine. Furthermore, if thresholds are set based on user experience, the accuracy of the set thresholds may be low, which can reduce the accuracy of abnormality detection.
[0006] This disclosure was made to solve these problems, and its purpose is to provide an equipment diagnostic system that can appropriately diagnose the condition of equipment without intentionally causing abnormalities in the equipment being diagnosed. [Means for solving the problem]
[0007] The equipment diagnostic system in this disclosure is an equipment diagnostic system that diagnoses a first piece of equipment. The equipment diagnostic system comprises a storage device and a diagnostic device. The storage device stores a first dataset, a second dataset, and a third dataset. The first dataset includes first data representing the state of the first piece of equipment when it is operating normally. The second dataset includes second data representing the state of a second piece of equipment similar to the first piece of equipment when it is operating normally. The third dataset includes third data representing the state of the second piece of equipment when it is not operating normally. The diagnostic device diagnoses the state of the first piece of equipment based on the first dataset, the second dataset, and the third dataset. Each of the first dataset, the second dataset, and the third dataset includes at least one common first type of feature. The variance of the first type of feature in the first dataset and the second dataset is within a specified range. The diagnostic device compares the first dataset and the second dataset, selects the first type of feature from the types of features contained in the first and second datasets, uses the first type of feature in the second dataset and the first type of feature in the third dataset as training data to generate a trained model for diagnosing the state of the second equipment, inputs the first type of feature in the first dataset into the generated trained model, and diagnoses the state of the first equipment. [Effects of the Invention]
[0008] According to this disclosure, the condition of the equipment can be appropriately diagnosed without intentionally causing any abnormalities in the equipment being diagnosed. [Brief explanation of the drawing]
[0009] [Figure 1]This is a diagram showing the configuration of the equipment diagnosis system in Embodiment 1. [Figure 2] This is a conceptual diagram for explaining the flow of equipment diagnosis in the equipment diagnosis system in Embodiment 1. [Figure 3] This is a diagram showing an example of one cycle of vibration data for a machine tool that is a diagnosis target in a normal state. [Figure 4] This is a diagram showing an example of one cycle of vibration data for a test machine tool in a normal state. [Figure 5] This is a flowchart for generating a trained model of the equipment diagnosis system in Embodiment 1. [Figure 6] This is a scatter diagram of the maximum vibration acceleration value in 200 cycles of vibration data for each machine tool. [Figure 7] This is a scatter diagram of values obtained by dividing the maximum vibration acceleration value in 200 cycles of vibration data for each machine tool by the standard deviation. [Figure 8] This is a diagram showing an example of learning data. [Figure 9] This is a diagram for explaining generation of a trained model using the k-nearest neighbor method. [Figure 10] This is a diagram showing a machine tool that is a diagnosis target in Embodiment 2. [Figure 11] This is a diagram showing the configuration of the equipment diagnosis system in Embodiment 2. [Figure 12] This is a conceptual diagram for explaining the flow of equipment diagnosis in the equipment diagnosis system in Embodiment 2. [Mode for Carrying Out the Invention]
[0010] Hereinafter, embodiments of the technical idea according to the present disclosure will be described with reference to the drawings. In the following description, the same components are denoted by the same reference numerals. Their names and functions are also the same. Therefore, detailed descriptions thereof will not be repeated.
[0011] Embodiment 1. <Configuration of Equipment Diagnosis System> FIG. 1 is a diagram showing a configuration of an equipment diagnosis system 100 according to Embodiment 1. The equipment diagnosis system 100 according to Embodiment 1 includes a machine tool 10 to be diagnosed and a test-purpose machine tool 20. The machine tools 10 and 20 in Embodiment 1 are, for example, machining centers. The machine tools 10 and 20 are not limited to machining centers, and may be other types of machines. For example, the machine tools 10 and 20 may be presses, lathes, or the like.
[0012] As shown in FIG. 1, the machine tool 10 that is the diagnosis target of the equipment diagnosis system 100 according to Embodiment 1 is incorporated into a production line. Products machined and molded by the machine tool 10 are sent to subsequent processes of the production line. On the other hand, the test-purpose machine tool 20 is not incorporated into the production line, and even if the machine tool 20 stops, the production efficiency of the production line is not affected.
[0013] The machine tool 20 does not need to be a machining center of the same model number as the machine tool 10. For example, the machine tool 20 may be a function used only for testing, and may have a function that the machine tool 10 actually incorporated in the production line does not have.
[0014] In this way, the machine tool 20 is not limited to a machine of the same model number as the machine tool 10, and it is sufficient that the machine tool 20 has similar machine characteristics to the machine tool 10. That is, it is sufficient that the use of the machine tool 10 and the use of the machine tool 20 are the same. When the machine tool 10 is a machining center, the machine tool 20 only needs to be a machining center, and when the machine tool 10 is a press, the machine tool 20 only needs to be a press. The machine tool 10 corresponds to the "first equipment" in the present disclosure. The machine tool 20 corresponds to the "second equipment" in the present disclosure.
[0015] As shown in FIG. 1, the equipment diagnosis system 100 includes, in addition to the machine tools 10 and 20, sensors 11 and 21, signal processing devices 12 and 22, a storage device 13, a diagnosis device 14, a control device 15, and a status indicator 16.
[0016] Sensors 11 and 21 are sensors that detect information representing the state of the machine tools 10 and 20, respectively. In Embodiment 1, sensors 11 and 21 are vibration sensors that detect vibrations generated in the bearings when the machine tools 10 and 20 are in operation. The operation of the machine tools 10 and 20 includes processing such as machining and molding of products.
[0017] The sensors 11 and 21 are not limited to vibration sensors, but may be of other types. For example, the sensors 11 and 21 may be microphones that detect vibration noise from bearings, current sensors that detect the current flowing to the drive motors included in the machine tools 10 and 20, load sensors that detect the load on a jig, image sensors that capture images of the state of the product being processed, temperature sensors, drive motor rotation speed sensors, and the like.
[0018] Sensor 11 in machine tool 10 is mounted in the same position as sensor 21 in machine tool 20. In other words, sensors 11 and 21 are mounted on components and parts that are common to both machine tools 10 and 20. More specifically, in Embodiment 1, sensor 11, which is a vibration sensor, is mounted on the outer wall of machine tool 10. Similarly, sensor 21, which is a vibration sensor, is mounted on the outer wall of machine tool 20. When sensors 11 and 21 are vibration sensors, it is desirable that sensors 11 and 21 be positioned near the workpiece to be cut.
[0019] If sensor 11 is a current sensor and is attached to the power line of the spindle rotation axis of the machine tool 10, then sensor 21 is also a current sensor, similar to sensor 11, and is attached to the power line of the spindle rotation axis of the machine tool 20.
[0020] The signal processing devices 12 and 22 process the signals received from sensors 11 and 21, respectively. That is, the signal processing devices 12 and 22 convert the information detected by sensors 11 and 21 into a digital format. The signal processing devices 12 and 22 include, for example, amplifiers, filters, A / D converters, etc. The storage device 13 stores the information converted by each of the signal processing devices 12 and 22.
[0021] The diagnostic device 14 includes a data acquisition unit 14A and a model generation unit 14B. The data acquisition unit 14A acquires data stored in the storage device 13. The model generation unit 14B generates a trained model. The diagnostic device 14, using the data acquisition unit 14A and the model generation unit 14B, generates a trained model for diagnosing the state of the machine tool 10 based on the information stored in the storage device 13. The diagnostic device 14 that generates the trained model is an example of a "training device" in this disclosure. If an abnormality is detected in the machine tool 10, the diagnostic device 14 transmits information indicating that an abnormality has occurred in the machine tool 10 to the control device 15 and the status indicator 16.
[0022] The control device 15 controls the machining performed by the machine tool 10. When the control device 15 receives information from the diagnostic device 14 indicating that an abnormality has occurred in the machine tool 10, it stops the machine tool 10. However, the control device 15 may not completely stop the machine tool 10, but may restrict only a portion of the machine tool 10's functions depending on the degree of the abnormality.
[0023] The status indicator 16 is, for example, the operation screen of the machine tool 10. The status indicator 16 notifies the user of an abnormality by displaying information on the screen indicating that an abnormality has occurred. The status indicator 16 may also include a lamp, a speaker, etc. If an abnormality occurs, the status indicator 16 will light up a lamp and emit a warning sound from the speaker. Furthermore, the status indicator 16 may have a function to send an email to the user indicating that an abnormality has occurred.
[0024] In Figure 1, the equipment diagnostic system 100 is described as having a configuration with one storage device 13, but the number of storage devices that can be provided in the equipment diagnostic system 100 is not limited to one. For example, the equipment diagnostic system 100 may have separate storage devices for storing information processed by the signal processing device 12 and for storing information processed by the signal processing device 22.
[0025] <Flow of equipment diagnostics> Figure 2 is a conceptual diagram illustrating the flow of equipment diagnosis in the equipment diagnosis system 100 in Embodiment 1. As shown in Figure 2, in step S1, the diagnostic device 14 acquires data on the normal state of the machine tool 10 incorporated into the production line. The normal state is a state in which the processing and molding operations of the product can be performed without problems.
[0026] When the machine tool 10 is in a normal state and performing processing and forming of a product, the sensor 11 detects vibrations occurring in the machine tool 10. The vibration information detected by the sensor 11 is converted into digital vibration data by the signal processing device 12 and transmitted to the storage device 13. The storage device 13 stores vibration data that occurs in the bearings when the machine tool 10 is in a normal state. In Embodiment 1, the storage device 13 stores vibration data for each processing cycle performed by the machine tool 10. A processing cycle is one unit of processing performed by the machine tool 10. For example, one operation such as grinding or drilling is called "one cycle".
[0027] Figure 3 shows an example of vibration data for one cycle in the machine tool 10, which is the subject of diagnosis for normal conditions. Figure 3 shows the vibration acceleration (m / s²) when one cycle of drilling is performed. 2 The waveform of the vibration is shown as vibration data, and the storage device 13 stores the vibration data shown in Figure 3. The maximum value of vibration acceleration in the example of vibration data in Figure 3 is 0.17 m / s². 2 That is the case.
[0028] Returning to Figure 2, in the equipment diagnostic system 100 of Embodiment 1, in step S1, vibration data for 200 cycles is acquired. That is, the equipment diagnostic system 100 causes the machine tool 10, which is in a normal state, to perform 200 machining cycles. The sensor 11 detects vibration information for 200 cycles. Note that the number of cycles is not limited to 200 cycles, but may be, for example, tens of thousands to millions of cycles.
[0029] Next, in step S2, the diagnostic device 14 acquires vibration data of the test machine tool 20 under normal conditions. Figure 4 shows an example of vibration data for one cycle of the test machine tool 20 under normal conditions. Figure 4 shows the vibration acceleration (m / s²) when one cycle of drilling is performed, similar to Figure 3. 2 The waveform of the vibration is shown as vibration data. In the example of vibration data shown in Figure 4, the maximum value of the vibration acceleration is 0.10 m / s². 2 In step S2, the equipment diagnostic system 100 causes the machine tool 20 to perform 200 machining cycles, and has the sensor 21 detect vibration information for each processing cycle.
[0030] Returning to Figure 2, in step S3, the equipment diagnostic system 100 acquires data on the abnormal condition of the test machine tool 20. That is, in step S3, the user intentionally causes an abnormality in the test machine tool 20. The user, for example, replaces a part included in the test machine tool 20 with a damaged part.
[0031] As a result, the equipment diagnostic system 100 acquires vibration data from the machine tool 20 in an abnormal state, which is different from the vibration data of the normal state acquired in step S2. For example, in the vibration data of the abnormal state, the maximum value of the vibration acceleration may be larger or smaller compared to the vibration data of the normal state. In step S3, the equipment diagnostic system 100 also has the machine tool 20 in the abnormal state perform 200 machining cycles and has the sensor 21 detect vibration information for each processing cycle.
[0032] As a result, the storage device 13 stores vibration data for 200 cycles from the machine tool 10 in a normal state, vibration data for 200 cycles from the machine tool 20 in a normal state, and vibration data for 200 cycles from the machine tool 20 in an abnormal state. Hereafter, for example, the vibration data for 200 cycles will be referred to as a "dataset". As mentioned above, the number of vibration data points in the dataset is not limited to 200, but may range from tens of thousands to millions of data points.
[0033] The dataset for machine tool 10 in a normal state corresponds to the “first dataset” in this disclosure. The dataset for machine tool 20 in a normal state corresponds to the “second dataset” in this disclosure. The dataset for machine tool 20 in an abnormal state corresponds to the “third dataset” in this disclosure.
[0034] Furthermore, the order of steps S1 to S3 is not limited to the example shown in Figure 2. For example, the equipment diagnostic system 100 may first acquire a dataset of the machine tool 20 in an abnormal state, then acquire a dataset of the machine tool 20 in a normal state, and finally acquire a dataset of the machine tool 10 in a normal state.
[0035] Next, as shown in Figure 2, the equipment diagnostic system 100 generates a trained model in step S4. The generation of the trained model in step S4 is described in detail in Figure 5. Finally, in step S5, the equipment diagnostic system 100 uses the trained model generated in step S4 to diagnose the state of the machine tool 10 that is the target of the diagnosis.
[0036] <Steps for generating a pre-trained model> Figure 5 is a flowchart for generating a trained model of the equipment diagnostic system 100 in Embodiment 1. Figure 5 shows the processes of steps S100 to S105 performed by the diagnostic device 14.
[0037] The data acquisition unit 14A of the diagnostic device 14 acquires a dataset of the machine tool 10 in a normal state stored in the storage device 13 (step S100). That is, the diagnostic device 14 acquires multiple vibration data of the machine tool 10 in a normal state from the storage device 13. Subsequently, the data acquisition unit 14A of the diagnostic device 14 acquires a dataset of the machine tool 20 in a normal state stored in the storage device 13 (step S101). That is, the diagnostic device 14 acquires multiple vibration data of the machine tool 20 in a normal state from the storage device 13.
[0038] The diagnostic device 14 compares the dataset of the machine tool 10 in a normal state with the dataset of the machine tool 20 in a normal state (step S102). In other words, the diagnostic device 14 compares the dataset obtained in step S100 with the dataset obtained in step S101. The comparison process in step S102 is performed to select the types of features with small differences from among the types of features common to the dataset of the machine tool 10 and the dataset of the machine tool 20, as will be described later.
[0039] The vibration data shown in Figures 3 and 4 includes various features. For example, the vibration data in Figures 3 and 4 includes features such as the maximum value, average value, or standard deviation of vibration acceleration in one cycle, or the vibration time in one cycle, as well as various parameters that represent the characteristics of the vibration data, such as combinations of these features.
[0040] For example, the diagnostic device 14 may create new features by performing transformation processing such as Fourier transform and wavelet transform on the waveforms detected by sensors 11 and 21. Alternatively, the diagnostic device 14 may create new features using dimensionality reduction algorithms such as principal component analysis.
[0041] The diagnostic device 14 selects the type of feature whose change in change is small between the dataset of the machine tool 10 in a normal state and the dataset of the machine tool 20 in a normal state. The method for selecting the type of feature with small difference will be explained below using Figures 6 and 7.
[0042] Figure 6 illustrates the difference in the feature "maximum vibration acceleration" between machine tool 10 and machine tool 20. Figure 6 is a scatter plot of the maximum vibration acceleration values for 200 cycles of vibration data for each machine tool 10 and 20. In Figure 6, the circular plots represent the maximum vibration acceleration values for each vibration data set over 200 cycles for machine tool 10 in a normal state. In Figure 6, the triangular plots represent the maximum vibration acceleration values for each vibration data set over 200 cycles for machine tool 20 in a normal state. As shown in Figure 6, the average of the maximum vibration acceleration values for each vibration data set, shown as circular plots, is approximately 0.14 m / s². 2 That is the case.
[0043] On the other hand, the average of the maximum vibration acceleration values in each vibration data point, shown as a triangular plot in Figure 6, is approximately 0.09 m / s². 2 Therefore, the difference between the average of the maximum vibration acceleration values in machine tool 10 and the average of the maximum vibration acceleration values in machine tool 20 is 0.05 m / s². 2 That is the case.
[0044] Next, Figure 7 explains the difference in the feature "value obtained by dividing the maximum value of vibration acceleration by the standard deviation" between machine tool 10 and machine tool 20. Figure 7 is a scatter plot of the value obtained by dividing the maximum value of vibration acceleration by the standard deviation in the vibration data for 200 cycles for each machine tool 10 and 20.
[0045] In Figure 7, the circular plots represent values obtained by dividing the standard deviation by the maximum value of vibration acceleration in each piece of vibration data for 200 cycles of the machine tool 10 in a normal state. The standard deviation for the circular plots in Figure 7 is the standard deviation of the maximum values of vibration acceleration over 200 cycles for the machine tool 10 in a normal state.
[0046] In Figure 7, the triangular plots represent values obtained by dividing the standard deviation by the maximum value of vibration acceleration in each piece of vibration data for 200 cycles of the machine tool 20 in a normal state. The standard deviation for the triangular plots in Figure 7 is the standard deviation of the maximum values of vibration acceleration over 200 cycles for the machine tool 20 in a normal state.
[0047] As shown by the circular plots in Figure 7, for the machine tool 10, the average of the values obtained by dividing the standard deviation by the maximum value of vibration acceleration of one cycle of vibration data is approximately 1.45 m / s 2 . Further, as shown by the triangular plots in Figure 7, for the machine tool 20, the average of the values obtained by dividing the standard deviation by the maximum value of vibration acceleration of one cycle of vibration data is approximately 1.44 m / s 2 . That is, the difference between the average of the values obtained by dividing the standard deviation by the maximum value of vibration acceleration for the machine tool 10 and the average of the values obtained by dividing the standard deviation by the maximum value of vibration acceleration for the machine tool 20 is 0.01 m / s 2 .
[0048] As described above, the feature quantity "value obtained by dividing the standard deviation by the maximum value of vibration acceleration" shown in Figure 7 has a smaller difference between the machine tool 10 and the machine tool 20 than the feature quantity "maximum value of vibration acceleration" shown in Figure 6.
[0049] That is, when the diagnostic device 14 acquires the "maximum value of vibration acceleration" shown in Figure 6 from certain vibration data, if the maximum value is close to 0.14 m / s 2 or a value close to 0.9 m / s 2Depending on how close the value is to the specified value, it is easier to predict whether the vibration data is from machine tool 10 or machine tool 20. On the other hand, even if the diagnostic device 14 obtains the "value obtained by dividing the maximum value of vibration acceleration by the standard deviation" shown in Figure 7 from a given vibration data, it is difficult to predict whether the vibration data is from machine tool 10 or machine tool 20 because the difference between machine tool 10 and machine tool 20 is small.
[0050] In the following, features that show a large difference between machine tool 10 and machine tool 20, such as the "maximum value of vibration acceleration" shown in Figure 6, will be referred to as "equipment-dependent features." Conversely, features that show a small difference between machine tool 10 and machine tool 20, such as the "value obtained by dividing the maximum value of vibration acceleration by the standard deviation" shown in Figure 7, will be referred to as "equipment-independent features."
[0051] Equipment-independent features may consist of multiple features. Specifically, equipment-independent features may be the result of dividing features together. Furthermore, equipment-independent features may be dimensionless features. The diagnostic device 14 may also reduce the difference in sensor output signals between machine tool 10 and machine tool 20 by connecting an amplifier to the sensor and amplifying the electrical signal output from the sensor. Specifically, the diagnostic device 14 uses an amplifier to amplify the sensor output signal with the smaller output between machine tool 10 and machine tool 20. This reduces the difference in sensor output signals between machine tool 10 and machine tool 20, and also reduces the difference in features.
[0052] Returning to Figure 5, the diagnostic device 14 selects the type of feature with small differences (step S103). Specifically, the diagnostic device 14 obtains the type of common feature from both the machine tool 10 and 20 datasets in a normal state and calculates the variance with both machine tool 10 and 20 datasets as the population. If the calculated variance is smaller than a predetermined range, the diagnostic device 14 selects that type of feature as a feature with small differences. In short, the diagnostic device 14 uses the dataset of machine tool 10 in a normal state and the dataset of machine tool 20 in a normal state as a single population and calculates the variance of features in that population.
[0053] In step S103 of Embodiment 1, the variance of the population for "the value obtained by dividing the maximum value of vibration acceleration by the standard deviation" shown in Figure 7 is within the specified range. Therefore, the diagnostic device 14 selects the feature "the value obtained by dividing the maximum value of vibration acceleration by the standard deviation" as a feature with small difference. On the other hand, the variance of the population for "the maximum value of vibration acceleration" shown in Figure 6 is not within the specified range. Therefore, the diagnostic device 14 does not select the feature "the maximum value of vibration acceleration" as a feature with small difference. In this way, the diagnostic device 14 selects features with small difference using variance. The diagnostic device 14 may select multiple features with small difference in step S103.
[0054] In Embodiment 1, in addition to the feature "value obtained by dividing the maximum value of vibration acceleration by the standard deviation," "vibration time per cycle" is selected as a feature with small difference. That is, the diagnostic device 14 in Embodiment 1 selects two feature quantities, "value obtained by dividing the maximum value of vibration acceleration by the standard deviation" and "vibration time per cycle," as feature quantities with small difference. The diagnostic device 14 may select three or more feature quantities as feature quantities with small difference. The type of feature quantity selected as a feature quantity with small difference corresponds to the "first type" in this disclosure.
[0055] Next, the data acquisition unit 14A of the diagnostic device 14 acquires a dataset of the abnormal state of the machine tool 20 stored in the storage device 13 (step S104). That is, the diagnostic device 14 acquires multiple vibration data of the abnormal state of the machine tool 20 from the storage device 13.
[0056] The diagnostic device 14 generates a trained model (step S105) using the features selected in step S103 from the dataset of machine tools 20 in normal and abnormal states as training data, and then terminates the process. Figure 8 shows an example of training data. Figure 8 shows the columns "Max / σ", "T", and "State Label".
[0057] The column "Feature Max / σ" shows the feature selected in step S103, which is "the value obtained by dividing the maximum value of vibration acceleration by the standard deviation." The column "Feature T" shows the feature selected in step S103, which is "the vibration time of one cycle." The column "State Label" is a state variable that indicates whether the machine tool 20 is in an abnormal state or a normal state. The state variable is associated with each vibration data.
[0058] As shown in Figure 8, the training data is numbered from 1 to 400 cycle numbers to uniquely identify each vibration data. Hereafter, the variable used to uniquely identify each vibration data will be referred to as the "independent variable." Note that the independent variable is not limited to the cycle number as shown in Figure 8, but could also be, for example, the processing start time. In Figure 8, the specific numerical values of the columns "Feature Max / σ" and "Feature T" for each vibration data are not shown for the sake of simplicity.
[0059] The rows from cycle 1 to cycle 200 show vibration data from the dataset for the machine tool 20 under normal conditions. The rows from cycle 201 to cycle 400 show vibration data from the dataset for the machine tool 20 under abnormal conditions.
[0060] Returning to Figure 5, in step S105, the model generation unit 14B of the diagnostic device 14 generates a trained model using the training data in Figure 8, which consists of features with small differences, as training data. As a result, the equipment diagnostic system 100 generates a trained model for inferring the state of the machine tool 10 from features that represent the state of the machine tool 10. The trained model is generated, for example, using the k-nearest neighbors method (k-NN). Figure 9 is a diagram illustrating the generation of a trained model using the k-nearest neighbors method.
[0061] Figure 9 shows a graph with "Feature Max / σ" on the vertical axis and "Feature T" on the horizontal axis. A portion of the vibration data from the training data in Figure 8 is plotted. In Figure 9, circular plots represent vibration data where the state variable is in a "normal state". The group of circular plots is called "Cluster A". That is, the data belonging to Cluster A are any of the data shown in the rows from cycle 1 to cycle 200 in Figure 8. In Figure 9, rectangular plots represent vibration data where the state variable is in an "abnormal state". The group of rectangular plots is called "Cluster B". That is, the data belonging to Cluster B are any of the data shown in the rows from cycle 201 to cycle 400 in Figure 8.
[0062] In the diagnostic process for diagnosing the condition of the machine tool 10, the diagnostic device 14 acquires new vibration data from the machine tool 10. The diagnostic device 14 determines whether the newly acquired vibration data from the machine tool 10 belongs to cluster A or cluster B.
[0063] The diagnostic device 14 acquires the following features from the new vibration data of the machine tool 10 to be diagnosed: "the value obtained by dividing the maximum value of vibration acceleration by the standard deviation" and "the vibration time per cycle," which are the features with small differences selected in step S103 of Figure 5. The diagnostic device 14 plots the new vibration data to be diagnosed as shown in Figure 9. The new vibration data q is shown as a star-shaped plot.
[0064] The diagnostic device 14 extracts k vibration data points in the vicinity of vibration data point q, which is shown as a star shape. In the example in Figure 9, k is 5, and 6 data points, including vibration data point q, are enclosed by dashed lines. As shown in Figure 9, among the 5 data points enclosed by dashed lines, the number of square-shaped plots is greater than the number of circular plots. Specifically, there are 3 square-shaped plots compared to 2 circular plots.
[0065] The diagnostic device 14 determines that cluster B, to which the most numerous rectangular plots belong among the k plots in the vicinity of vibration data q, is the cluster to which vibration data q belongs. This allows the diagnostic device 14 to classify the state of newly acquired vibration data based on a trained model generated from the dataset. It is preferable that the number of k be odd.
[0066] Furthermore, the diagnostic device 14 may use methods other than the k-nearest neighbors (k-NN) method to generate a trained model for determining the cluster to which vibration data q belongs. For example, the diagnostic device 14 may simply define the cluster to which vibration data q belongs as the cluster to which vibration data q belongs, which contains vibration data with the closest value to it. Alternatively, the diagnostic device 14 may classify vibration data q using methods such as decision trees or support vector machines.
[0067] Furthermore, the state variables may include other states in addition to "normal state" and "abnormal state." For example, "abnormal state" may be further subdivided into "machining defect state," "tool missing state," etc. "Machine defect state" refers to a state in which a defect occurs in the product processed by the machine tool 20. "Tool missing state" refers to a state in which an abnormality occurs in the tool included in the machine tool 20.
[0068] These subdivisions may be performed manually by the user, or the diagnostic device 14 may subdivide the "abnormal state" based on the characteristics of the vibration data. That is, the diagnostic device 14 extracts the characteristics of each vibration data that is in an "abnormal state" and further classifies each vibration data using the k-means method or the like. Similarly, the diagnostic device 14 may extract the characteristics of each vibration data that is in a "normal state" and subdivide the "normal state". In this way, the diagnostic device 14 further classifies the state of the machine tool 20 based on the dataset of machine tool 20 in a normal state and the dataset of machine tool 20 in an abnormal state.
[0069] Thus, in the first embodiment, the equipment diagnostic system 100 can diagnose the state of the machine tool 10 without stopping the machine tool 10 incorporated into the production line, by using vibration data detected from a machine tool 20 similar to the machine tool 10 as training data. As a result, the equipment diagnostic system 100 can appropriately diagnose the state of the machine tool 10 without intentionally causing an abnormality in the machine tool 10 to be diagnosed. In other words, since the equipment diagnostic system 100 does not stop the machine tool 10, it suppresses a decrease in the operating rate of the machine tool 10, and furthermore, by using a trained model, it can perform more accurate anomaly detection than when setting an anomaly detection threshold based on the user's experience, thus suppressing a decrease in the accuracy of anomaly detection.
[0070] <Modified example in Embodiment 1> The machine tool 20 may not be an actual machine tool, but a simulated machine tool. The simulation parameters of the simulated machine tool 20 are adjusted so that its specifications, model, and mechanical configuration match those of the actual machine tool 10 incorporated into the production line.
[0071] Embodiment 2. In Embodiment 1, the equipment diagnostic system 100 was described in an example where machine tool 10 and machine tool 20 are separate machine tools. In Embodiment 2, a configuration for diagnosing the condition of a single machine tool to which multiple tools are attached will be described. In Embodiment 2, configurations that overlap with those of Embodiment 1 will not be repeated in the equipment diagnostic system 200.
[0072] Figure 10 shows a machine tool 30 to be diagnosed in Embodiment 2. The machine tool 30 has tools 38A to 38F stored in a holder 37. As shown in Figure 10, the machine tool 30 mounts one of the tools 38A to 38F in a mounting section 39 to process the workpiece 36. In the example shown in Figure 10, tool 38A is mounted in the mounting section 39. The machine tool 30 processes the workpiece 36 by using different tools from the holder 37 according to the processing program. The machine tool 30 can correspond to the "processing equipment" in this disclosure.
[0073] In Embodiment 2, the diagnostic device 14 diagnoses the state of the machine tool 30 using tool 38A and acquires learning data from the machine tool 30 using tool 38B. Tools 38A and 38B are tools of the same type with the same cutting characteristics. For example, in Embodiment 2, both tools 38A and 38B are drills. That is, tools 38A and 38B are similar tools. Tools 38A and 38B may be end mills, reamers, turning inserts, or various other tools, but it is desirable that tools 38A and 38B are of the same type.
[0074] For example, if tool 38A is an end mill, it is desirable that tool 38B is also an end mill. Tools 38A and 38B differ in tool diameter, tool length, tool material, or cutting conditions such as rotational speed and feed rate. Tool 38A corresponds to the "first tool" in this disclosure, and each of tools 38B to 38F corresponds to the "second tool" in this disclosure.
[0075] <Configuration of the equipment diagnostic system> Figure 11 shows the configuration of the equipment diagnostic system 200 in Embodiment 2. As shown in Figure 11, in Embodiment 2, the equipment diagnostic system 200 includes a sensor 31 and a signal processing device 32. The sensor 31 in Embodiment 2 is, for example, a current sensor. The sensor 31, which is a current sensor, is preferably attached to the wire from the power supply of the spindle motor that rotates one of the tools 38A to 38F of the machine tool 30 to the inverter, or to the wire from the inverter to the motor. Each of the sensor 31, signal processing device 32, storage device 13, diagnostic device 14, control device 15, and status indicator 16 may be located inside the machine tool 30 or may be provided separately from the machine tool 30. If the sensor 31 is a current sensor or a vibration sensor, it is preferable that the sampling period be the same for measuring tool 38A and tool 38B.
[0076] <Flowchart of equipment diagnosis in Embodiment 2> Figure 12 is a conceptual diagram illustrating the flow of equipment diagnosis in the equipment diagnosis system 200 in Embodiment 2. The equipment diagnosis system 200 acquires data on the normal state of the machine tool 30 using tool 38A (step S21). Subsequently, the equipment diagnosis system 200 changes the tool used by the machine tool 30 from tool 38A to tool 38B.
[0077] The equipment diagnostic system 200 acquires data on the machine tool 30 using the tool 38B in a normal state (step S22). Subsequently, similar to Embodiment 1, an abnormality is intentionally induced in the tool 38B, and the equipment diagnostic system 200 acquires data on the machine tool 30 using the tool 38B in an abnormal state (step S23).
[0078] The equipment diagnostic system 100 generates a trained model based on the data acquired in steps S21 to S23 (step S24). Specifically, the equipment diagnostic system 200 selects features with small differences from multiple data points of the machine tool 30 in a normal state using tool 38A and multiple data points of the machine tool 30 in a normal state using tool 38B. It is desirable to select dimensionless features. The equipment diagnostic system 200 generates a trained model using the features with small differences from multiple data points of the machine tool 30 using tool 38B in both normal and abnormal states.
[0079] Subsequently, the equipment diagnostic system 200 uses the generated trained model to diagnose the state of the machine tool 30 using tool 38A (step S25). In this way, the equipment diagnostic system 200 of Embodiment 2 can diagnose whether or not an abnormality has occurred in tool 38A using only data from cases where an abnormality has occurred in tool 38B. The equipment diagnostic system 200 may also use the generated trained model to diagnose the state of the machine tool 30 using other tools similar to tool 38B, not just tool 38A.
[0080] In other words, in the equipment diagnostic system 200, when generating a trained model to diagnose the state of each machine tool 30 when using tools 38A to 38F, it is not necessary to intentionally cause an abnormality with respect to tool 38A, which is similar to tool 38B. As a result, in the equipment diagnostic system 200 of Embodiment 2, a trained model can be generated by acquiring data on the abnormal and normal states when using one tool 38B, without acquiring data on the abnormal states when using other tools.
[0081] Furthermore, in the equipment diagnostic system 200 of Embodiment 2, the condition of the machine tool 30 using the tool 38A can be appropriately diagnosed without intentionally causing an abnormality in the machine tool 30 using the tool 38A to be diagnosed. In other words, the equipment diagnostic system 200 does not stop the machine tool 30 in order to cause an abnormal condition by attaching each tool to it, thus suppressing a decrease in the operating rate of the machine tool 30. Moreover, by using a trained model, more accurate abnormality detection can be achieved than when setting thresholds for abnormality detection based on the user's experience, thus suppressing a decrease in the accuracy of abnormality detection.
[0082] <How to obtain abnormal data> In the equipment diagnostic system 200, when a durability test is performed, a dataset of the normal and abnormal states of the machine tool 30 while using the tool 38B may be acquired. A durability test is a test in which the machine tool 30 is operated until the tool 38B or the machine tool 30 deteriorates, thereby testing the durability of the tool 38B and the machine tool 30. In a durability test, the machine tool and tool are operated continuously under harsh conditions in order to cause abnormalities in a short period of time.
[0083] When the durability test begins, both tool 38B and machine tool 30 are in normal condition. 。 As the durability test progresses, the tool 38B and machine tool 30 deteriorate, and abnormalities occur. By acquiring datasets of normal and abnormal states along with the durability test, users can see which features change as the system deteriorates.
[0084] The following is an example of obtaining data sets in normal and abnormal conditions along with a durability test. In the equipment diagnostic system 200, the machine tool 30 using tool 38B is operated for 1000 hours under the conditions of a tool rotation speed of 3000 rpm and a feed rate of 600 mm / min. During this time, the sensor 31, which is a current sensor, measures the current value flowing through the machine tool 30.
[0085] The equipment diagnostic system 200 further changes the tool rotation speed from 3000 rpm to 6000 rpm as an accelerated endurance test. The equipment diagnostic system 200 continues the accelerated endurance test until a machining defect occurs in the product being machined, or until an abnormality occurs in the machine tool 30 or the tool 38B itself. When an abnormality occurs, the equipment diagnostic system 2 In step 00, it is determined whether or not an abnormality occurs when the tool rotation speed is returned from 6000 rpm to 3000 rpm.
[0086] If no abnormality occurs when the tool rotation speed is returned to 3000 rpm, the equipment diagnostic system 200 performs the accelerated durability test again. If an abnormality occurs even when the tool rotation speed is returned to 3000 rpm, it is determined that the equipment has deteriorated sufficiently, and the process of acquiring a dataset of abnormal conditions is started. Thus, in Embodiment 2, the acquisition of a dataset of abnormal conditions may be performed together with the durability test.
[0087] (summary) The following summarizes Embodiments 1 and 2.
[0088] As shown in Figures 1 to 9, the equipment diagnostic system 100 in this disclosure is an equipment diagnostic system that diagnoses a machine tool 10. The equipment diagnostic system 100 comprises a storage device 13 and a diagnostic device 14. The storage device 13 stores a first data set, a second data set, and a third data set. The first data set includes first data representing the state of the machine tool 10 when it is operating normally. The second data set includes second data representing the state of a machine tool 20 similar to the machine tool 10 when it is operating normally. The third data set includes third data representing the state of the machine tool 20 when it is not operating normally. The diagnostic device 14 diagnoses the state of the machine tool 10 based on the first data set, the second data set, and the third data set. The machine tool 20 is equipment similar to the machine tool 10. Each of the first data, the second data, and the third data includes at least a common first type of feature quantity. The variance of the first type of feature in the first and second datasets is within a specified range. The diagnostic device 14 compares the first and second datasets, selects the first type of feature from among the types of features contained in the first and second datasets, uses the first type of feature in the second dataset and the first type of feature in the third dataset as training data to generate a trained model for diagnosing the state of the machine tool 20, inputs the first type of feature in the first dataset into the generated trained model, and diagnoses the state of the machine tool 10.
[0089] The objective of the equipment diagnostic system 100 is to provide an equipment diagnostic system that can appropriately diagnose the condition of the machine tool 10 without intentionally causing any abnormalities in the machine tool 10 to be diagnosed.
[0090] Preferably, the use of the machine tool 10 is the same as the use of the machine tool 20. Preferably, the diagnostic device 14 generates a trained model using information indicating the state when the device is operating normally, linked to the second data, and information indicating the state when the device is not operating normally, linked to the third data, as training data.
[0091] Preferably, the diagnostic device 14 classifies the state of the machine tool 20 based on the second and third data sets.
[0092] Preferably, the state in which the machine tool 20 is not functioning properly is a state in which a machining defect has occurred.
[0093] As shown in Figures 10 to 12, the equipment diagnostic system 200 is an equipment diagnostic system that diagnoses a machine tool 30 that performs machining using tool 38A or tool 38B. It comprises a storage device 13 that stores a first data set containing first data representing the state of the machine tool 30 when tool 38A is used and the machine tool 30 is operating normally, a second data set containing second data representing the state of the machine tool 30 when tool 38B is used and the machine tool 30 is operating normally, and a third data set containing third data representing the state of the machine tool when tool 38B is used and the machine tool 30 is not operating normally, and a diagnostic device that diagnoses the state of the machine tool 30 based on the first data set, the second data set, and the third data set. Each of the first data, the second data set, and the third data set contains at least one common first type of feature. The variance of the first type of feature in the first data set and the second data set is within a specified range. The diagnostic device 14 compares the first dataset and the second dataset, selects the first type of feature from among the types of features contained in the first and second datasets, uses the first type of feature in the second dataset and the first type of feature in the third dataset as training data to generate a trained model for diagnosing the state of the machine tool 30 using tool 38B, inputs the first type of feature in the first dataset into the generated trained model, and diagnoses the state of the machine tool 30 using tool 38A.
[0094] The embodiments disclosed herein should be considered in all respects to be illustrative and not restrictive. The scope of the invention is indicated by the claims rather than by the foregoing description, and all modifications within the meaning and scope of the claims are intended to be included. [Explanation of symbols]
[0095] 10, 20, 30 Machine tools, 11, 21, 31 Sensors, 12, 22, 32 Signal processing devices, 13 Storage devices, 14 Diagnostic devices, 14A Data acquisition unit, 14B Model generation unit, 15 Control devices, 16 Status indicators, 36 Workpieces, 37 Holders, 38A~38F Tools, 100, 200 Equipment diagnostic systems.
Claims
1. A system for diagnosing equipment, the first equipment is the target of the diagnosis. A storage device that stores a first data set including first data representing the state of the first equipment when the first equipment is operating normally, a second data set including second data representing the state of a second equipment similar to the first equipment when it is operating normally, and a third data set including third data representing the state of the second equipment when it is not operating normally. The system includes a diagnostic device that diagnoses the state of the first equipment based on the first data set, the second data set, and the third data set. Each of the first data, the second data, and the third data includes at least multiple types of features, The diagnostic device is The first dataset and the second dataset are compared, and from among the multiple types of features contained in each of the first and second datasets, a first type of feature is selected in which the variance of the type of feature common to both the first and second datasets is within a specified range. Using the first type of feature in the second data and the first type of feature in the third data as training data, a trained model for diagnosing the state of the second equipment is generated. The generated trained model is input to the first type of feature in the first data, and the state of the first equipment is diagnosed. The variance is calculated using the population obtained by combining the number of common features included in the first data and the number of common features included in the second data, in an equipment diagnostic system.
2. The equipment diagnostic system according to claim 1, wherein the use of the first equipment is the same as the use of the second equipment.
3. The diagnostic device includes information indicating the state when the device is operating normally, associated with the second data, and information indicating the state when the device is not operating normally, associated with the third data. The equipment diagnostic system according to claim 1 or claim 2, which generates the trained model using the above as training data.
4. The equipment diagnostic system according to claim 1 or 2, wherein the diagnostic device classifies the state of the second equipment based on the second data set and the third data set.
5. The equipment diagnostic system according to claim 1 or claim 2, wherein the state in which the equipment is not operating normally is a state in which a processing defect has occurred in the second equipment.
6. A machine diagnostic system that diagnoses machining equipment that uses a first tool or a second tool, A storage device that stores a first data set including first data representing the state of the processing equipment when the first tool is used and the processing equipment is operating normally, a second data set including second data representing the state of the processing equipment when the second tool is used and the processing equipment is operating normally, and a third data set including third data representing the state of the processing equipment when the second tool is used and the processing equipment is not operating normally. The system includes a diagnostic device that diagnoses the state of the processing equipment based on the first dataset, the second dataset, and the third dataset, Each of the first data, the second data, and the third data includes at least multiple types of features, The diagnostic device is The first dataset and the second dataset are compared, and from among the multiple types of features contained in each of the first and second datasets, a first type of feature is selected in which the variance of the type of feature common to both the first and second datasets is within a specified range. Using the first type of feature in the second data and the first type of feature in the third data as training data, a trained model is generated for diagnosing the state of the processing equipment using the second tool. The generated trained model is input to the first type of feature in the first data, and the state of the machining equipment using the first tool is diagnosed. The variance is calculated using the population obtained by combining the number of common features included in the first data and the number of common features included in the second data, in an equipment diagnostic system.
7. A learning device that generates a trained model used to diagnose the state of the first equipment, A data acquisition unit that acquires a first data set including first data representing the state of the first equipment when the first equipment is operating normally, a second data set including second data representing the state of a second equipment similar to the first equipment when it is operating normally, and a third data set including third data representing the state of the second equipment when it is not operating normally. The system comprises a model generation unit that generates a trained model for inferring the state of the first equipment from features representing the state of the first equipment using the first dataset, the second dataset, and the third dataset, The trained model is input to a first type of feature, which is selected by comparing the first dataset and the second dataset, and whose variance in features common to the first and second datasets is within a specified range. The variance is calculated using the population as the sum of the number of common features included in the first data and the number of common features included in the second data. Learning device.
8. A method for generating a trained model used to diagnose the state of a first piece of equipment, Steps to obtain: a first dataset containing first data representing the state of the first equipment when it is operating normally; a second dataset containing second data representing the state of a second equipment similar to the first equipment when it is operating normally; and a third dataset containing third data representing the state of the second equipment when it is not operating normally. The process includes the step of generating a trained model for inferring the state of the first equipment from features representing the state of the first equipment, using the first dataset, the second dataset, and the third dataset. The trained model is input to a first type of feature, which is selected by comparing the first dataset and the second dataset, and whose variance in features common to the first and second datasets is within a specified range. A method for generating a trained model, wherein the variance is calculated using the population as the sum of the number of common features included in the first data and the number of common features included in the second data.
Citation Information
Patent Citations
Deterioration detection system
JP2020057144A
Abnormality detection device and abnormality detection method of cutting tool
JP2020104257A
Failure predictor detection device, failure predictor detection method, failure predictor detection program, learning device, method for generating trained learning model, and trained learning model generation program
WO2021245898A1
