Electric motor vibration model generation device, electric motor diagnosis system and electric motor diagnosis program

Abstract

To improve accuracy to detect an abnormality of an electric motor.SOLUTION: An electric motor vibration model generation device 100 includes a current data preprocessing unit 104, a vibration data preprocessing unit 109, a time-series adjustment unit 110, and a first arithmetic processing unit 111. The current data preprocessing unit 104 extracts a plurality of kinds of feature amounts from input current data acquired from an electric motor to output them as preprocessing current data in association with an acquisition time of the input current data. The vibration data preprocessing unit 109 extracts a plurality of kinds of feature amounts from vibration data acquired from the electric motor to output them as preprocessing vibration data in association with an acquisition time of the vibration data. The time-series adjustment unit 110 adjusts a time series so that the preprocessing current data and the preprocessing vibration data are regarded as data acquired at the same time to output both of the data. The first arithmetic processing unit 111 generates a learning data set from the output of the time-series adjustment unit 110 to use the learning data set to train an electric motor vibration model to store the trained electric motor vibration model.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present disclosure relates to a technique for detecting an abnormality in an electric motor. [Background technology]

[0002] If an electric motor installed in a plant suddenly stops, it can cause significant damage to the manufacturing process. Therefore, there is a demand for detecting abnormalities in the electric motor before the motor stops due to a failure, that is, while the motor is operating.

[0003] A known method for detecting motor abnormalities is to perform FFT analysis on motor current data and determine an abnormality when the spectrum intensity at a specific frequency exceeds a certain threshold. With this method, when a characteristic waveform appears in the current data, it is possible to detect the characteristic waveform as an abnormality. Another known method is to perform vector calculations on the current data and determine motor abnormalities based on feature quantities extracted from the current vector. For example, Patent Document 1 and Patent Document 2 can be cited as documents disclosing conventional techniques for detecting motor abnormalities.

[0004] As described above, various techniques for detecting motor abnormalities have been proposed in the past. However, because a motor failure can have a significant impact on the manufacturing process, there is a demand for further improvement in the accuracy of detecting motor abnormalities before a failure occurs. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] International Publication No. 2019 / 102545 [Patent Document 2] JP 2020-114084 A Summary of the Invention [Problem to be solved by the invention]

[0006] The present disclosure has been made in view of the above problems. It is an object of the present disclosure to improve the accuracy of predicting and detecting an abnormality in an electric motor. [Means for solving the problem]

[0007] In order to achieve the above object, the present disclosure provides an electric motor vibration model generating device for generating an electric motor vibration model for diagnosing an electric motor. a current data pre-processing unit configured to extract a plurality of types of feature amounts from the input current data acquired from the electric motor, and to output the pre-processed current data in association with an acquisition time of the input current data; a vibration data preprocessing unit configured to extract a plurality of types of feature amounts from the vibration data acquired from the electric motor, and output the preprocessed vibration data in association with an acquisition time of the vibration data; a time series adjustment unit configured to adjust a time series of the pre-processed current data and the pre-processed vibration data so that the pre-processed current data and the pre-processed vibration data are regarded as data acquired at the same time, and output the adjusted time series; A first arithmetic processing unit, The first calculation processing unit is generating a learning dataset from the output of the time series adjustment unit; training the motor vibration model using the training data set; storing the trained electric motor vibration model; and It is characterized by the following.

[0008] In order to achieve the above object, the present disclosure provides a motor diagnostic system for diagnosing a motor. a storage unit that stores the trained motor vibration model generated by the motor vibration model generation device; The current data preprocessing unit of the electric motor vibration model generating device according to the present disclosure; The vibration data preprocessing unit of the electric motor vibration model generating device according to the present disclosure; The time series adjustment unit of the electric motor vibration model generation device according to the present disclosure; a current sensor for measuring real-time input current of the motor; a current data input unit that links the input power measured by the current sensor to a real-time current measurement time so that the input power measured by the current sensor during diagnosis of the electric motor can be treated as real-time current data in the current data preprocessing unit; a vibration sensor for measuring real-time vibration of the electric motor; a vibration data input unit that links the vibration measured by the vibration sensor during diagnosis of the electric motor to a real-time vibration measurement time so that the vibration measured by the vibration sensor can be handled as real-time vibration data in the vibration data preprocessing unit; a second calculation processing unit that inputs to the time series adjustment unit real-time preprocessed current data processed in the current data preprocessing unit based on the real-time current data and linked to the real-time current measurement time, and real-time preprocessed vibration data processed in the vibration data preprocessing unit based on the real-time vibration data and linked to the real-time vibration measurement time, converts the real-time preprocessed current data and the real-time preprocessed vibration data into second time processed current data and second time processed vibration data whose time series have been adjusted so that they are regarded as data acquired at the same time, and inputs the second time processed current data and the second time processed vibration data to the learned motor vibration model, thereby predicting a vibration state of the motor after the time when real-time data is acquired; A determination unit that detects an abnormality in the electric motor based on the predicted vibration state of the electric motor. It is characterized by the following.

[0009] In order to achieve the above object, the present disclosure provides a motor diagnostic program for diagnosing a motor. a first step of generating preprocessed current data by extracting a plurality of types of feature amounts from input current data of an electric motor to be diagnosed and linking the extracted feature amounts to an acquisition time of the input current data, and generating preprocessed vibration data by extracting a plurality of types of feature amounts from vibration data acquired from the electric motor and linking the extracted feature amounts to an acquisition time of the vibration data; a second step of adjusting the time series so that the preprocessed current data and the preprocessed vibration data are regarded as data acquired at the same first time, and converting them into first time-processed current data and first time-processed vibration data; a third step of learning a motor vibration model using the first time, the first time-processed current data, and the first time-processed vibration data as a learning data set; A fourth step of verifying the motor vibration model; A fifth step of returning to the first step if the verification result in the fourth step does not satisfy a predetermined accuracy; a sixth step of generating real-time preprocessed current data by linking the measured input current to a real-time current measurement time so that the measured real-time input current of the motor can be treated as real-time current data, generating real-time preprocessed vibration data by linking the measured vibration to a real-time vibration measurement time so that the measured real-time vibration of the motor can be treated as real-time vibration data, and adjusting the time series so that the real-time preprocessed current data and the real-time preprocessed vibration data are regarded as data acquired at the same second time, thereby converting them into second-time processed current data and second-time processed vibration data; a seventh step of predicting a vibration state of the motor after the second time by inputting the second time-processed current data and the second time-processed vibration data into the learned motor vibration model; an eighth step of returning to the sixth step if the predicted vibration of the electric motor does not exceed a predetermined threshold; and a ninth step of executing a predetermined process corresponding to an abnormality when the predicted vibration exceeds the predetermined threshold. It is characterized by the following. Effect of the Invention

[0010] The vibration of a motor, which reflects an abnormality of the motor, has a close relationship with the current. This relationship differs for each motor, and also changes over time even for the same motor. The motor vibration model generation device of the present disclosure can generate a motor vibration model in which the relationship between the feature amount of the current data and the vibration data has been learned. The motor diagnosis system of the present disclosure can estimate the vibration state from real-time current data, etc., by using the learned motor vibration model, and can accurately predict and detect abnormalities in the motor based on the estimated vibration state. [Brief description of the drawings]

[0011] [Figure 1] 1 is a diagram illustrating a configuration of a motor vibration model generation device and a motor diagnosis system according to an embodiment of the present disclosure. FIG. [Diagram 2] FIG. 13 is a diagram illustrating the configuration of a motor vibration model generation device and a motor diagnosis system according to a modified example of an embodiment of the present disclosure. [Diagram 3] FIG. 13 is a diagram illustrating pre-processing of current data. [Figure 4] FIG. 11 is a diagram illustrating pre-processing of vibration data. [Diagram 5] FIG. 2 is a diagram for explaining items of process data. [Figure 6] FIG. 13 is a diagram illustrating time series adjustment of data. [Figure 7] FIG. 11 is a diagram illustrating a data smoothing process. [Figure 8] FIG. 11 is a diagram illustrating a data clipping process. [Figure 9] FIG. 13 is a diagram illustrating data interpolation processing. [Figure 10] FIG. 2 is a diagram illustrating a calculation unit of the electric motor vibration model generating device. [Figure 11] FIG. 4 is a diagram illustrating abnormality determination by the motor diagnostic system. [Figure 12] 3A to 3C are diagrams illustrating the operation of the motor vibration model generating device and the motor diagnostic system. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0012] Hereinafter, a motor vibration model generation device according to an embodiment of the present disclosure and a motor diagnosis system using the same will be described with reference to the drawings.

[0013] 1 shows the configuration of a motor vibration model generating device 100 and a motor diagnosis system 120. The motor vibration model generating device 100 includes an input data processing unit 500, a calculation unit 111, a vibration model storage unit 112, and a model display unit 113. The input data processing unit 500 includes a data input unit 530, a data storage unit 520, a data preprocessing unit 510, and a time series adjustment unit 110. The data input unit 530 includes a current data input unit 102 that acquires a current value input to a motor to be diagnosed measured by a current sensor 101, a vibration data input unit 107 that acquires a vibration value of the motor to be diagnosed from a vibration sensor 106, and a process data input unit 603 that acquires a process value. The data storage unit 520 includes a current data accumulation unit 103, a vibration data accumulation unit 108, and a process data accumulation unit 105. The data pre-processing unit 510 includes a current data pre-processing unit 104, a vibration data pre-processing unit 109, and a process data pre-processing unit 605. The motor diagnosis system 120 includes, in addition to the motor vibration model generating device 100, a calculation unit 118, an abnormality determination unit 115, and an abnormality detection display unit 116. Both the current sensor 101 and the vibration sensor 106 are provided in the same motor. The motor in which the current sensor 101 and the vibration sensor 106 are provided is the motor to be diagnosed by the motor diagnosis system 120. The process data input unit 603 receives as input a process sensor value from a process data sensor 601 or a process control command value from a process control device 602 installed in a plant in which the motor to be diagnosed by the motor diagnosis system 120 is installed.

[0014] First, the generation of the learning model will be described. The current data input unit 102 generates current data by linking the input current value to the time when it was measured (also called the acquisition time), and outputs it to the current data storage unit 103. The current data storage unit 103 stores the current data. The vibration data input unit 107 generates vibration data by linking the input vibration value to the time when it was measured (also called the acquisition time), and outputs it to the vibration data storage unit 108. The vibration data storage unit 108 stores the vibration data. The process data input unit 603 generates process data by linking the input process sensor value and the time when the process control command value was measured or acquired (both are simply called the acquisition time), and outputs it to the process data storage unit 105. The process data storage unit 105 stores the process data. The process sensor value and the process control command value may be collectively referred to simply as the process value. The current data storage unit 103, the vibration data storage unit 108, and the process data storage unit 105 are, for example, storage devices such as a magnetic disk, a flash memory, or an optical disk.

[0015] The data pre-processing unit 510 includes a current data pre-processing unit 104 that performs pre-processing on the current data stored in the current data storage unit 103. Fig. 3 shows the details of the current data pre-processing unit 104. The current data pre-processing unit 104 includes a current data acquisition unit 201, a current data calculation unit 202, and a feature extraction unit 203. The current data acquisition unit 201 acquires necessary current data from the current data storage unit 103. The necessary current data is, for example, current data for a predetermined time period in the past from a specified time.

[0016] The current data calculation unit 202 performs AC coupling processing on the current data acquired by the current data acquisition unit 201 to remove DC components contained in the current data, and further performs FFT processing such as overlap processing, FFT averaging processing, and window function processing. Then, the feature extraction unit 203 performs processing on the current data after FFT processing processed by the current data calculation unit 202 to extract features of the current data. The processing by the feature extraction unit 203 is, for example, calculation of an overall value, or calculation of a feature frequency and its peak value after various FFT processing performed by the current data calculation unit 202. The feature extraction unit 203 may perform both calculation of the overall value and calculation of a feature frequency and its peak value after various FFT processing.

[0017] The current data preprocessing unit 104 outputs the feature values ​​calculated by the feature value extraction unit 203, i.e., the overall value, the feature frequency, and the peak value. Furthermore, the current data preprocessing unit 104 may perform differentiation and integration on the feature values ​​after various FFT processes, and output these in addition to the feature values. The output of the current data preprocessing unit 104 is called preprocessed current data. The preprocessed current data refers to data that includes the feature values ​​of the input current of the electric motor to be diagnosed, its differential value (amount of change), and its integral value, and is linked to the time when the current value was acquired.

[0018] Returning to FIG. 1 again, the configuration of the motor vibration model generating device 100 will be described. The data pre-processing unit 510 further includes a vibration data pre-processing unit 109 that performs pre-processing on the vibration data stored in the vibration data storage unit 108. FIG. 4 shows the vibration data pre-processing unit 109 in detail. The vibration data pre-processing unit 109 includes a speed data calculation unit 301. The speed data calculation unit 301 acquires necessary vibration data from the vibration data storage unit 108. The necessary vibration data is, for example, vibration data for a predetermined time period in the past from a specified time. When the vibration data acquired from the vibration sensor 106 is represented by acceleration, the speed data calculation unit 301 converts the vibration data into speed data by integrating the vibration data. When the vibration data acquired from the vibration sensor 106 is represented by displacement, the speed data calculation unit 301 converts the vibration data into speed data by differentiating the vibration data.

[0019] The vibration data preprocessing unit 109 outputs the vibration data converted into speed data. However, when the vibration data acquired from the vibration sensor 106 is speed data, the vibration data preprocessing unit 109 outputs the input vibration data as it is. In the subsequent processing by the electric motor vibration model generating device 100, the vibration data means the vibration data represented by speed. The vibration data preprocessing unit 109 may perform various FFT processing on the vibration data converted by the speed data calculation unit 301 in the same manner as the feature amount extraction unit 203 of the current data preprocessing unit 104, and may calculate both the overall value and the feature frequency and its peak value after various FFT processing. The vibration data preprocessing unit 109 may output only a value not subjected to FFT processing, or the overall value plus both the feature frequency and its peak value after various FFT processing. The vibration data preprocessed by the vibration data preprocessing unit 109 is called preprocessed vibration data. The preprocessed vibration data means data linked to the acquisition time.

[0020] The vibration data preprocessing unit 109 may perform differentiation and integration on the feature quantities after various FFT processes, and output these in addition to the feature quantities. The output of the vibration data preprocessing unit 109 is called preprocessed current vibration data. The preprocessed current vibration data includes the feature quantities of the vibration (speed) of the electric motor to be diagnosed, as well as their differential values ​​(amount of change) and integral values, and refers to data linked to the time when the vibration was acquired.

[0021] Returning to FIG. 1 again, the configuration of the motor vibration model generating device 100 will be described. In the process data input unit 603, when the acquisition time differs depending on the type of the input process sensor value and process control command value, the process data and the process data acquisition time are linked according to the acquisition time of each type. FIG. 5 shows an example of items of process data stored in the process data storage unit 105. The process data includes, for example, motor speed sensor data, pressure sensor data, flow rate sensor data, liquid level sensor data, valve opening data, temperature sensor data, PH sensor data, film thickness sensor data, motor speed command data, pressure command data, flow rate command data, liquid level command data, valve opening command data, temperature command data, tension sensor data, etc., or start commands and stop commands for the motor and other devices. These data are directly or indirectly related to the operation of the motor. However, it is not necessary that all of these data are stored in the process data storage unit 105, and data other than these may be stored in the process data storage unit 105. If the plant is normal, the output values ​​of various sensors in the plant will follow the command values, so the command values ​​may be used as process data instead of the sensor values. For example, in the case of a sensorless induction motor drive, it is difficult to obtain the actual rotation speed of the induction motor, so the rotation speed command of a drive device such as a sensorless induction motor drive that drives the induction motor corresponds to the command value.

[0022] The data preprocessing unit 510 further includes a process data preprocessing unit 605 that performs preprocessing on the process data stored in the process data storage unit 105. The process data preprocessing unit 605 may output the input process data as is to the time series adjustment unit 110, or may perform a numerical calculation such as addition, subtraction, multiplication, division, differentiation, integration, or transfer function calculation such as first-order lag on the process data that can be expressed numerically, and then output the process data to the time series adjustment unit 110. For logical values ​​such as a start command or a stop command, an integrated value of the number of times the operation is performed may be output. The process data preprocessed by the process data preprocessing unit 605 is called preprocessed process data. The preprocessed process data means data linked to the acquisition time. The processing performed by the process data preprocessing unit 605 (output of the value as is, addition, subtraction, multiplication, division, differentiation, integration, transfer function calculation such as first-order lag, and integration of the number of operations) is an example of a numerical calculation.

[0023] Returning to FIG. 1 again, the configuration of the electric motor vibration model generating device 100 will be described. The input data processing unit 500 includes a time series adjustment unit 110. The preprocessed current data obtained by the current data preprocessing unit 104 is linked to the current data acquisition time, the preprocessed vibration data processed by the vibration data preprocessing unit 109 is linked to the vibration data acquisition time, and the preprocessed process data processed by the process data preprocessing unit is linked to the process data acquisition time, and these are input to the time series adjustment unit 110. In general, the measurement intervals of the current data, vibration data, and process data acquired by different sensors are different because the data are acquired by sampling. For this reason, the acquisition times of the sampling data for a certain period of time are often different. The difference in the acquisition times of the sampling data between the data makes it difficult to perform machine learning to associate the data with each other, which will be described later. Therefore, the time series adjustment unit 110 performs the process described below in order to match the sampling times between the data and associate the data in a time series.

[0024] FIG. 6 shows details of the process by the time series adjustment unit 110. When data with a short measurement interval is to be matched to data with a long measurement interval, smoothing or clipping is performed on the data with a short measurement interval. The smoothing process is a method of calculating a moving average value for b minutes while shifting by a minutes as shown in FIG. 7. By adjusting the time of a minutes, data with a short measurement interval whose time is aligned with data with a long measurement interval can be obtained. The clipping process is a method of calculating a maximum value, a minimum value, a median value, and a change amount for d minutes while shifting by c minutes as shown in FIG. 8. By adjusting the time of c minutes, data with a short measurement interval whose time is aligned with data with a long measurement interval can be obtained. The averaging process and the clipping process may be freely selected by the operator. Conversely, when data with a long measurement interval is to be matched to data with a short measurement interval, data interpolation is performed on the data with a long measurement interval. The data interpolation process is a process of interpolating data at intervals of e minutes between measurement data as shown in FIG. 9. By adjusting the time of e minutes, data with a long measurement interval whose time is aligned with data with a short measurement interval can be obtained. The above method allows the current data, vibration data, and process data to be regarded as data obtained by sampling at the same time. In this way, the time series adjustment unit 110 converts the preprocessed current data, preprocessed vibration data, and preprocessed process data into data that can be regarded as obtained at the same time. The sampling time of the learning data set that can be regarded as obtained at the same time is called the first time. The preprocessed current data that can be regarded as obtained at the first time is an example of the first time processed current data, the preprocessed vibration data that can be regarded as obtained at the first time is an example of the first time processed vibration data, and the preprocessed process data that can be regarded as obtained at the first time is an example of the first time processed process data. The first time processed current data, the first time processed vibration data, and the first time processed process data that can be regarded as sampled at a plurality of the same times are collectively called processed data for model generation, and constitute a data set for learning the electric motor vibration model.

[0025] Returning to FIG. 1 again, the configuration of the electric motor vibration model generating device 100 will be described. The electric motor vibration model generating device 100 includes a calculation unit 111. The calculation unit 111 generates a vibration model of an electric motor using first time processed current data, first time processed vibration data, and first time processed process data that can be regarded as data acquired at the same time (first time) that have been time-series adjusted by the time series adjustment unit 110 of the input data processing unit 500. FIG. 10 shows an overview of processing by the calculation unit 111. The calculation unit 111 learns and generates a vibration model using a learning data set in which a pre-specified feature amount of the first time processed vibration data consisting of data at multiple times is set as target data (output variable data), and the first time processed current data, first time processed vibration data, and first time processed process data consisting of data at multiple times are set as input variable data. For example, a sparse model is used for the vibration model. The sparse model is one of the machine learning techniques, and is a technique for calculating a model formula such as Equation 1 using output variable data and input variable data that is closely related to the output variable data. In Equation 1, Y is the objective variable (output variable), xi (i = 1, 2, 3, 4, …) is the input variable, and βj (j = 0, 1, 2, 3, 4, …) is the parameter obtained by learning. Y = β0 + β1*x1 + β2*x2 + β3*x3 + β4*x4 + … Formula 1

[0026] As a specific example of the learning data of the input variable, x1 may be the effective value of the current, which is one of the features of the first time-processed current data, x2 may be the overall value of the current, which is one of the features of the first time-processed current data, x3 may be the first process data (pressure), which is one of the first time-processed process data, x4 may be the second process data (flow rate), which is one of the first time-processed process data, and x5 may be a certain frequency component of the vibration data, which is one of the features of the first time-processed vibration data. As a specific example of the learning data of the output variable, Y may be, for example, a peak value (maximum value of simple velocity), which is one of the features of the first time-processed vibration data after a predetermined time has elapsed from the first time. However, if no process data available for learning has been accumulated, only the first time-processed current data may be used as the learning data of the input variable, or the first time-processed vibration data may be excluded from the input variable and the first time-processed vibration data may be used only as the target data for learning. The first time-processed current data, the first time-processed vibration data, and the first time-processed vibration data at time T-ΔT, which is a predetermined time ΔT back from a certain time T that has been time-series adjusted, may be used as an input variable set of learning data, and one of the feature quantities of the time-series adjusted first time-processed vibration data at time T may be used as an output variable (target data). Then, a plurality of different input variable sets at the first time and a target data set may be used as a learning data set. Alternatively, in the above-mentioned learning data set, the time-series adjusted and preprocessed first time-processed vibration data at time T-ΔT and the current data and vibration data of a specific frequency component of the first time processing may be used as an input variable set, and another frequency component of the time-series adjusted vibration data at time T or a peak value of the time-series adjusted simple vibration data may be used as the target data set.

[0027] According to the sparse model, it is possible to obtain a parameter βi by learning such that the target data (output variable) is a function of time t as shown in the following formula 2 for the input variables as described above. For example, t is the current time. In formula 2, Y(t) is one of the features indicated by the first time processing vibration data that can be considered to have been measured at time t, and xi(t-t1) is a feature indicated by the first time processing current data, the first time processing process data, or the first time processing vibration data, which are input variables considered to have been measured at time t-t1, which is t1 back from the present (time t). It is also possible to create a learning model and a learning data set that do not use part or all of the first time processing vibration data as an input variable, or a learning model and a learning data set that do not use part or all of the first time processing process data as an input variable. Y(t) = β0 + β1*x1(t-t1) + β2*x2(t-t1) + β3*x3(t-t1) + β4*x4(t-t1) + … Formula 2

[0028] If the parameter βi that satisfies Equation 2 can be obtained by learning, it becomes possible to predict the vibration Y(t+t1) as a target variable (output variable) for a time t1 in the future from the first time processing current data, the first time processing vibration data, and the first time processing process data at the current time t, as shown in Equation 3. Note that the time t1 does not have to be one type of time, and the parameter βi may be learned for a number of different times. In that case, the parameter βi is learned and set for each different time of the time t1. Y(t+t1) = β0 + β1*x1(t) + β2*x2(t) + β3*x3(t) + β4*x4(t) + … Equation 3

[0029] The motor vibration model generating device 100 includes a model storage unit 112 that stores and calculates a vibration model used for diagnosis by the motor diagnostic system 120. The model storage unit 112 is, for example, a storage device such as a magnetic disk, a flash memory, or an optical disk. A trained vibration model obtained by learning by the calculation unit 111 is stored in the model storage unit 112. When a vibration model is re-trained, the vibration model stored in the model storage unit 112 is updated to the re-trained vibration model.

[0030] The motor vibration model generating device 100 also includes a model display unit 113 that displays the trained vibration model stored in the model storage unit 112 to the operator. Specifically, the model display unit 113 displays a model formula of the vibration model as expressed by Equation 2 or Equation 3. Furthermore, the model display unit 113 displays an accuracy index of the vibration model together with the model formula. For example, the accuracy index may be an R2 score, which is a coefficient of determination in machine learning. The R2 score can numerically express how well the vibration model fits the measured vibration data. That is, the input variables at time T-t1 of the learning data set are applied to the vibration model, and the vibration Y(T), which is the output variable at time T obtained from Equation 2 as a result, is compared with the vibration (the same feature component as the vibration Y(T) in the first time processing vibration data) included in the learning data set as the output variable at time T.

[0031] By checking the R2 score displayed on the model display unit 113, the operator can determine whether or not to use the vibration model in diagnosis by the motor diagnostic system 120. For example, if the R2 score is equal to or greater than a threshold, the generated vibration model can be used for diagnosis, and if the R2 score is less than the threshold, the vibration model can be re-learned and a new vibration model can be generated. If the R2 score is less than the threshold, a determination to re-learn the vibration model can be automatically made, and the re-learning can be performed automatically.

[0032] Moreover, the operator can determine the degree of contribution of the input variable data corresponding to the parameter to the output variable by checking the value of the parameter of the model formula displayed on the model display unit 113. For example, if the absolute value of β3 in formula 1 is extremely small compared to other parameters, it can be determined that the input variable data represented by x3 has an extremely low effect on the vibration, which is the output variable. In other words, when x3 indicates a certain first time processing process data, for example, when it is the first time processing process data representing pressure, it can be determined that the first time processing process data of pressure represented by x3 is the first time processing process data with low importance in the vibration model. In that case, from the next vibration model learning, the vibration model may be learned using a learning data set in which the input variable data represented by x3 is excluded. Since the product of the parameter βi and the input variable xi(t-t1) affects the formula or the output variable Y(t), the product of the parameter βi and the input variable xi(t-t1) may be displayed on the model display unit 113. For example, the horizontal axis may be the time axis, and the vertical axis may be a graph of the product xi of each parameter βi and each input variable, making it easy for the operator to check. Alternatively, the product xi of each parameter βi and each input variable may be displayed in ascending or descending order. Data of input variables with small absolute values ​​of the products of the parameter βi and the input variable xi may be excluded from the learning data set for the next vibration model. In this way, the efficiency of learning the vibration model can be improved by limiting the learning data set to useful ones.

[0033] The motor vibration model generation device 100 may be programmed to evaluate the parameters of the learned vibration model or the product of the parameters and the input variables, and to exclude input variable data with low importance from the target of the next learning, instead of the operator. For example, thresholds (parameter threshold 1 or parameter-input variable product threshold 1) may be set in advance for the absolute value of the parameters or the product of the parameters and the input variables, and when the absolute value of the parameters is smaller than the parameter threshold 1 or the absolute value of the product of the parameters and the input variables is smaller than the parameter-input variable product threshold 1, the motor vibration model generation device 100 may be programmed to automatically exclude the target input variables from the target of the learning data set for the next learning.

[0034] Also, for example, if the value of β4 or the product of β4 and x4 in formula 1 is extremely large compared to other parameters, it can be determined that the input variable represented by x4 has an extremely large effect on vibration. In other words, it can be determined that the input variable represented by x4 is an input variable of high importance in the trained vibration model (an input variable requiring attention). By monitoring such input variables, the operator can detect abnormalities in the motor before they occur.

[0035] The motor vibration model generating device 100 may be programmed to evaluate the parameters of the trained vibration model and monitor the input variables of high importance instead of the operator. For example, a threshold (parameter threshold 2 or parameter-input variable product threshold 2) may be set in advance for the absolute value of the parameter or the absolute value of the product of the parameter and the input variable, and the motor vibration model generating device 100 may be programmed to alert the operator to data indicating an input variable whose parameter value is greater than the parameter threshold 2 or the parameter-input variable product threshold 2. For example, when the parameter value exceeds the threshold, a means for alerting the operator may be configured to highlight the corresponding parameter or input variable when displaying the trained vibration model on the model display unit 113.

[0036] The data pre-processing unit 510 includes a temporary storage unit required for performing necessary pre-processing, and the time series adjustment unit 110 also includes a temporary storage unit required for performing necessary time series adjustment. In the motor vibration model generating device 100 configured as above, the data pre-processing unit 510, the time series adjustment unit 110, and the calculation unit 111 correspond to an example of a first calculation processing unit. These may each be realized by independent hardware, or may be realized by a program (software) that operates on a single common computer or multiple distributed computers.

[0037] FIG. 2 is a diagram for explaining the configuration of a motor vibration model generating device 100A and a motor diagnostic system 120A according to a modified example of the embodiment of the present disclosure. The same reference numerals are used for components representing the same functions as those in FIG. 1, and the description thereof will be omitted. The motor vibration model generating device 100A includes an input data processing unit 500A, a calculation unit 111, a model storage unit 112, and a model display unit 113. The motor diagnostic system 120A includes a calculation unit 118, an abnormality determination unit 115, and an abnormality detection display unit 116 in addition to the motor vibration model generating device 100A. The input data processing unit 500A includes a data input unit 530, a data storage unit 520A, a data preprocessing unit 510A, and a time series adjustment unit 110. The configuration and function of the data input unit 530 are the same as those in FIG. 1, and therefore the description thereof will be omitted. The data preprocessing unit 510A includes a current data preprocessing unit 104A, a vibration data preprocessing unit 109A, and a process data preprocessing unit 605A. The data storage unit 520A includes a current data storage unit 103A, a vibration data storage unit 108A, and a process data storage unit 105A. The difference between the embodiment in Fig. 1 and the modified embodiment in Fig. 1A is that in Fig. 1, data flows from the data input unit 530 to the data storage unit 520, then to the data pre-processing unit 510A, and then to the time series adjustment unit 110, whereas in Fig. 1A, data flows from the data input unit 530 to the data pre-processing unit 510A, then to the data storage unit 520A, and then to the time series adjustment unit 110.

[0038] The current data linked to the acquisition time output from the current data input unit 102 is output to the current data preprocessing unit 104A, and the current data preprocessing unit 104A performs preprocessing similar to that of the current data preprocessing unit 104. The preprocessed current data linked to the acquisition time is accumulated in the current data storage unit 103A, and the accumulated preprocessed current data linked to the acquisition time is output to the time series adjustment unit 110. The vibration data linked to the acquisition time output from the vibration data input unit 107 is output to the vibration data preprocessing unit 109A, and the vibration data preprocessing unit 109A performs preprocessing similar to that of the vibration data preprocessing unit 109. The preprocessed vibration data linked to the acquisition time is accumulated in the vibration data storage unit 108A, and the accumulated preprocessed vibration data linked to the acquisition time is output to the time series adjustment unit 110. The process data linked to the acquisition time output from the process data input unit 603 is output to the process data preprocessing unit 605A, and the process data preprocessing unit 605A performs preprocessing similar to that of the process data preprocessing unit 605. The pre-processing process data linked to the acquisition time is accumulated in the process data accumulation unit 105A, and the accumulated pre-processing process data linked to the acquisition time is output to the time series adjustment unit 110. The data processing after the time series adjustment unit 110 is the same as in the embodiment shown in FIG. 1, so a description thereof will be omitted.

[0039] The data pre-processing unit 510A includes a temporary storage unit required for performing necessary pre-processing, and the time series adjustment unit 110 also includes a temporary storage unit required for performing necessary time series adjustment. In the motor vibration model generation device 100A configured as above, the data pre-processing unit 510A, the time series adjustment unit 110, and the calculation unit 111 correspond to a first calculation processing unit. These may each be realized by independent hardware, or may be realized by a program (software) that operates on a single common computer or multiple distributed computers.

[0040] Next, in the embodiment of the present disclosure in FIG. 1, diagnosis of an electric motor using the learned vibration model stored in the above-mentioned model storage unit 112 of the electric motor diagnostic system 120 will be described. When diagnosing an electric motor, real-time data is acquired by the following means, and vibration of the electric motor after time t1 is predicted from the real-time data. As in the generation of the learning model, the data input unit 530 outputs real-time current data linked to the acquisition time from the current sensor 101, outputs real-time vibration data linked to the acquisition time from the vibration sensor 106, and outputs real-time process data linked to the acquisition time such as a process sensor value from the process data sensor 601 and a process control command value from the process control device 602. Then, as in the case of generating the learning model, the input data processing unit 500 performs data storage by the data storage unit 520 and preprocessing by the data preprocessing unit 510 on the current data, vibration data, and process data, and converts the preprocessed current data, preprocessed vibration data, and preprocessed process data acquired at different times into second time processed current data, second time processed vibration data, and second time processed process data that can be regarded as acquired at the same second time by the time series adjustment unit 110. Furthermore, in order to predict the vibration of the electric motor, the sampling time of the input variable data as real time data is converted into a value at a second time that can be regarded as acquired at the same time (adjustment of the time series), and inputs it to the calculation unit 118 as input variable data for prediction (real time processed data). The electric motor diagnosis system 120 reads out the learned vibration model stored in the model storage unit 112 and inputs it to the calculation unit 118. The learning model called from the model storage unit and input to the calculation unit 118 is a parameter defined in a sparse model. The real-time input variable data considered to have been acquired at the second time and input from the time series adjustment unit 110 to the calculation unit 118 is an input variable defined in the sparse model. Each piece of real-time input variable data is associated with a time series.The feature amount of the real-time current data, the feature amount of the real-time vibration data, and the feature amount of the real-time process data are processed by the data preprocessing unit 510, and the time series adjustment is performed by the time series adjustment unit 110. The second time processed current data, the second time processed vibration data, and the second time processed process data obtained in this way are input to the learned vibration model as real-time input variable data, thereby obtaining vibration data corresponding to the predicted value of the vibration. That is, the parameter βi of Equation 3 is called from the model storage unit 112 as the learned vibration model, and input to the calculation unit 118. The output of the time series adjustment unit 110, which can be regarded as having been obtained at the second time at the current time t, is set as the input variable xi(t) of Equation 3, and input to the calculation unit 118 as real-time input variable data to perform the calculation of Equation 3, thereby obtaining the vibration predicted value Y(t+t1) after the time t1 from the current time t (time t+t1). In Fig. 1, the input data processing unit 500 is the same when generating the learning model and when diagnosing the electric motor, but part or all of the input data processing unit 500 may be configured differently when generating the learning model and when diagnosing the electric motor. In order to obtain real-time input variable data, the current data, vibration data, and process data may be input from the data input unit 530 to the data preprocessing unit 510 without passing through the data storage unit 520, as indicated by the dashed lines in Fig. 1. The current data, vibration data, and process data stored in the data storage unit 520 when diagnosing the electric motor may be used when generating the next learning dataset.

[0041] The vibration data calculated using the trained vibration model is input to the abnormality determination unit 115. A threshold is set in advance in the abnormality determination unit 115. When the magnitude of vibration expressed by speed exceeds the threshold, a detection message for notifying an operator of an abnormality is displayed on the abnormality detection display unit 116. Specifically, as shown in FIG. 11, a threshold for determining that a mechanical abnormality has occurred in the motor may be set in two stages, and a detection message may be displayed on the abnormality detection display unit 116 every time the magnitude of vibration exceeds each threshold. For example, the detection message may be "Caution" when the magnitude of vibration exceeds threshold 1, and the detection message may be "Alarm" when the magnitude of vibration exceeds threshold 2 that is larger than threshold 1. The abnormality detection display unit 116 may be a display device different from the model display unit 113, or may be the same display device. In addition, the detection message may be configured to display a predicted time to reach threshold 1 or threshold 2.

[0042] Diagnosis of an electric motor using a trained vibration model stored in the model storage unit 112 of the electric motor diagnostic system 120A in the modified example of the embodiment of the present disclosure shown in Fig. 2 will be described. Components having the same functions as those in Fig. 1 are given the same reference numerals and description thereof will be omitted. When diagnosing an electric motor, similarly to when generating the trained model, the data input unit 530 outputs current data linked to the acquisition time from the current sensor 101, outputs vibration data linked to the acquisition time from the vibration sensor 106, and outputs process data linked to the acquisition time such as a process sensor value from the process data sensor 601 and a process control command value from the process control device 602. Then, in the same manner as in the generation of the learning model, the input data processing unit 500A performs preprocessing on the current data, vibration data, and process data in the data preprocessing unit 510A, stores the data in the data storage unit 520A, and converts the preprocessed current data, preprocessed vibration data, and preprocessed process data acquired at different times into second time processed current data, second time processed vibration data, and second time processed process data that can be regarded as having been acquired at the same second time in the time series adjustment unit 110, and inputs them to the calculation unit 118 as real-time input variable data. The following description is the same as that in FIG. 1 and will be omitted. Note that, as indicated by the dashed lines in FIG. 1A, the preprocessed current data, preprocessed vibration data, and preprocessed process data may be input from the data preprocessing unit 510A to the time series adjustment unit 110 without passing through the data storage unit 520A. Also, the preprocessed current data, preprocessed vibration data, and preprocessed process data stored in the data storage unit 520A during diagnosis of the electric motor may be used when generating the next learning data set.

[0043] In the motor diagnostic system 120 and the motor diagnostic system 120A configured as above, the calculation unit 118 and the abnormality determination unit 115 correspond to an example of a second calculation processing unit. These may be realized by independent hardware, or may be realized by a program (software) that runs on one common computer or multiple distributed computers.

[0044] 12 is a flow chart for explaining the operation of the motor vibration model generating device 100 and the motor diagnostic system 120. That is, the explanation will be given assuming the configuration of FIG.

[0045] First, a vibration model of a motor is generated by the motor vibration model generating device 100. Steps S100 to S105 are processes performed by the motor vibration model generating device 100.

[0046] In step S100, in the input data processing unit 500, the current value acquired by the current sensor 101 is linked to the acquisition time and accumulated in the current data storage unit 103 via the current data input unit 102. Furthermore, the process value and the process command value acquired by the process data sensor 601 and the process control device 602 are linked to the acquisition time and accumulated in the process data storage unit 105 via the process data input unit 603. Furthermore, the vibration value acquired by the vibration sensor 106 is linked to the acquisition time and accumulated in the vibration data storage unit 108 via the vibration data input unit 107. When the number of accumulated data becomes equal to or greater than a predetermined value, the process proceeds to step 101. The predetermined number is a number at which a significant learning result (for example, an R score equal to or greater than a predetermined value) is expected. The current data, process data, and vibration data stored in the data storage unit 520 including the current data storage unit 103, the process data storage unit 105, and the vibration data storage unit 108 are a database for generating a learning dataset.

[0047] In step S101, current data is called from the current data storage unit 103, and converted into preprocessed current data through preprocessing by the current data preprocessing unit 104. Also, process data is called from the process data storage unit 105, and converted into preprocessed process data through preprocessing by the process data preprocessing unit 605. Furthermore, vibration data is called from the vibration data storage unit 108, and converted into preprocessed vibration data through preprocessing by the vibration data preprocessing unit 109. A time series association is performed between these preprocessed data in the time series adjustment unit 110. Then, a learning data set in which the time series association (which can be considered to have the same acquisition time) has been performed is generated.

[0048] In step S102, the calculation unit 111 learns a vibration model (a vibration model of the electric motor) using the learning data set for which time series association has been performed.

[0049] In step S103, the vibration model being trained is verified. For example, the parameter value of the model formula, the product value of the parameter and the input variable, and an accuracy index such as an R2 score are referenced for the verification of the vibration model. At this time, the importance of each input variable is also verified based on the model formula of the vibration model.

[0050] In step S104, the accuracy of the vibration model (R2 score, etc.), parameter values, and product values ​​of parameters and input variables are displayed to the operator as verification results.

[0051] Then, in step S105, it is determined whether the verification result of the vibration model satisfies the desired accuracy. If the verification result does not satisfy the desired accuracy, the process returns to step S100. When step S100 is repeatedly performed, the accuracy may be improved by adding stored data, or when a predetermined number of data or more is stored, new data may be stored and the oldest data may be deleted in order to prevent overflow of the storage area. The processes of steps S100 to S104 are repeated until the verification result satisfies the desired accuracy, and the learning of the vibration model is continued. If the verification result satisfies the desired accuracy, the process proceeds to step S106.

[0052] If the verification result of the vibration model satisfies the desired accuracy, the vibration model is determined to have been trained, and is stored in the vibration model storage unit 112 as a trained vibration model in step S106 so that it can be used in diagnosing the motor by the motor diagnosis system 120. Then, the process proceeds to step S206. Steps S206 to S209 are processes performed by the motor diagnosis system 120. The trained model is, for example, a vibration model in which the parameter βi for the input variable in a sparse model is determined to satisfy a predetermined accuracy in step S105 as a result of the learning in step S102.

[0053] In step S206, real-time data of current data, process data, and vibration velocity data are collected by the data input unit 530 in association with the acquisition time, and are stored in the data storage unit 520 and preprocessed by the data preprocessing unit 510. The preprocessed data is associated in time series with the acquired data by the time series adjustment unit 110 to obtain real-time input variable data. Then, the process proceeds to step S207.

[0054] In step S207, the trained vibration model is called from the storage unit 112 to the calculation unit 118, and further, data corresponding to the input variables corresponding to the parameters of the trained vibration model among the real-time input variable data is input to the calculation unit 118 to calculate a predicted value of vibration. A predicted value of vibration of the electric motor is output from the vibration model. As described above, the types of feature quantities such as current data, process data, and vibration data need only be the input variables necessary for the vibration model. Then, the process proceeds to step S208.

[0055] In step S208, it is determined whether the predicted vibration value exceeds the threshold value. If the predicted vibration value exceeds the threshold value, an abnormality handling process is performed in step S209. As the abnormality handling process, for example, a message notifying the operator that an abnormality has occurred is displayed on a display device. If the predicted vibration value does not exceed the threshold value, the process proceeds to step S210.

[0056] Steps S210 to S212 are processes performed by the motor vibration model generating device 100. If the predicted value of vibration calculated by the vibration model does not exceed the threshold value, the data collected in step S206 is stored as additional data for re-learning the vibration model. In step S210, it is determined whether a predetermined number of additional data has been collected. The predetermined number is a number that is determined to be enough to obtain a significant learning result. If the predetermined number is not reached, the process returns to step S206.

[0057] When a predetermined amount of additional data has been collected, the process of step S211 is performed first. In step S211, input variables with high importance are displayed to the operator as data indicating input variables requiring attention. In addition, input variable data with low importance is determined as learning data to be excluded. The input variables to be excluded are also displayed to the operator.

[0058] In step S212, preprocessing is performed on the data added to the data storage unit 520 in step 106. Then, the preprocessed added data is added to the database for generating the learning dataset, and the database is updated. Then, the process returns to step S101. The motor vibration model generating device 100 performs the process of steps S101 to S105 again, and re-learns the vibration model based on the updated database. Note that, during re-learning, if the accuracy of the re-learned vibration model is not higher than that of the vibration model before re-learning, the trained model before re-learning may be used as is, and if the accuracy of the re-learned vibration model is higher, a determination may be added to proceed to step S206 with the re-learned vibration model as a new trained model.

[0059] When FIG. 12 is interpreted as a flow chart for explaining the operation of the motor vibration model generating device 100A and the motor diagnostic system 120A configured in FIG. 1A, steps S100, S101, and S212 should be interpreted as follows.

[0060] In step S100, in the input data processing unit 500A, the current value from the current sensor 101 is linked to the acquisition time, passes through the current data input unit 102, and is preprocessed by the current data preprocessing unit 104A, and is stored in the current data storage unit 103A as preprocessed current data. Furthermore, the process value and the process command value from the process data sensor 601 and the process control device 602 are linked to the acquisition time, passes through the process data input unit 603, and is preprocessed by the process data preprocessing unit 605A, and is stored in the process data storage unit 105A as preprocessed process data. Furthermore, the vibration value from the vibration sensor 106 is linked to the acquisition time, passes through the vibration data input unit 107, and is preprocessed by the vibration data preprocessing unit 109A, and is stored in the vibration data storage unit 108A as preprocessed vibration data. When the number of stored data becomes equal to or greater than a predetermined value, the process proceeds to step 101. The predetermined number is a number at which a significant learning result (for example, an R score equal to or greater than a predetermined value) is expected. The preprocessed current data, preprocessed process data, and preprocessed vibration data stored in data storage unit 520A including current data storage unit 103A, process data storage unit 105A, and vibration data storage unit 108A are a database for generating a learning data set.

[0061] In step S101, preprocessed current data, preprocessed process data, and preprocessed vibration velocity data (preprocessed vibration data expressed in velocity) are read out from the data storage unit 520A, and these data are associated in time series by the time series adjustment unit 110. Then, a learning data set in which the data are associated in time series (the data can be considered to have the same acquisition time) is generated.

[0062] In step S206, real-time data of current data, process data, and vibration velocity data are collected by the data input unit 530 in association with the acquisition time, and are preprocessed by the data preprocessing unit 510A. The preprocessed data is stored in the data storage unit 520A, and the acquired data is associated in time series by the time series adjustment unit 110 to obtain real-time input variable data. In step S212, preprocessing is performed on the additional data. Then, the preprocessed additional data is added to the database, and the database is updated.

[0063] By operating the motor vibration model generation device 100 or the motor vibration model generation device 100A as described above, it is possible to generate a vibration model in which the relationships between the current data, the process data, and the vibration data have been learned. Then, by operating the motor diagnosis system 120 or the motor diagnosis system 120A as described above, it is possible to calculate a predicted value of vibration using the vibration model generated by the motor vibration model generation device 100 or the motor vibration model generation device 100A. This makes it possible to predict the vibration state of the motor from real-time data of the current data, the process data, and the vibration data, and to accurately detect an abnormality in the motor based on the predicted vibration state. [Explanation of symbols]

[0064] 100 Electric motor vibration model generator 101 Current Sensor 102 Current data input section 103 Current data storage unit 104 Current data preprocessing unit 105 Process Data Storage Unit 106 Vibration Sensor 107 Vibration data input section 108 Vibration Data Storage Unit 109 Vibration data preprocessing section 110 Time Series Adjustment Unit 111 Arithmetic section 112 Model Memory Section 113 Model display section 118 Arithmetic section 115 Abnormality determination section 116 Abnormality detection display unit 120 Motor Diagnostic System

Claims

1. An electric motor vibration model generating device for generating an electric motor vibration model for diagnosis of an electric motor, comprising: a current data pre-processing unit configured to extract a plurality of types of feature amounts from the input current data acquired from the electric motor, and to output the pre-processed current data in association with an acquisition time of the input current data; a vibration data preprocessing unit configured to extract a plurality of types of feature amounts from the vibration data acquired from the electric motor, and output the preprocessed vibration data in association with an acquisition time of the vibration data; a time series adjustment unit configured to adjust a time series of the pre-processed current data and the pre-processed vibration data so that the pre-processed current data and the pre-processed vibration data are regarded as data acquired at the same time, and output the adjusted time series; A first arithmetic processing unit, The first arithmetic processing unit is generating a learning dataset from the output of the time series adjustment unit; training the motor vibration model using the training data set; storing the trained electric motor vibration model; and 2. A motor vibration model generating device comprising:

2. 2. The motor vibration model generating device according to claim 1, a process data pre-processing unit configured to perform a numerical calculation on the process data acquired from the equipment in which the electric motor is installed, and to output the process data as pre-processed process data by linking the process data to an acquisition time of the process data; The time series adjustment unit is configured to adjust the time series of the preprocessing process data so that the preprocessing process data is regarded as data acquired at the same time as the preprocessing current data and the preprocessing vibration data, and output the adjusted time series.

2. A motor vibration model generating device comprising:

3. 3. The motor vibration model generating device according to claim 2, The motor vibration model is a sparse model, The first arithmetic processing unit is Evaluating a model equation of the learned electric motor vibration model; and excluding input variables with low importance determined based on the evaluation from the next learning.

2. A motor vibration model generating device comprising:

4. 3. The motor vibration model generating device according to claim 2, The motor vibration model is a sparse model, The first arithmetic processing unit is Evaluating a model equation of the learned electric motor vibration model; and alerting an operator to input variables of high importance determined based on the evaluation.

2. A motor vibration model generating device comprising:

5. A motor diagnostic system for diagnosing an electric motor, comprising: A storage unit that stores the trained motor vibration model generated by the motor vibration model generation device according to claim 1; The current data preprocessing unit of the electric motor vibration model generating device according to claim 1 ; The vibration data preprocessing unit of the electric motor vibration model generating device according to claim 1 ; The time series adjustment unit of the electric motor vibration model generating device according to claim 1 ; a current sensor for measuring real-time input current of the motor; a current data input unit that links the input power measured by the current sensor during diagnosis of the electric motor to a real-time current measurement time so that the input power can be handled as real-time current data by the current data preprocessing unit; a vibration sensor for measuring real-time vibration of the electric motor; a vibration data input unit that links the vibration measured by the vibration sensor during diagnosis of the electric motor to a real-time vibration measurement time so that the vibration measured by the vibration sensor can be handled as real-time vibration data in the vibration data preprocessing unit; a second calculation processing unit that inputs to the time series adjustment unit real-time preprocessed current data processed by the current data preprocessing unit based on the real-time current data and linked to the real-time current measurement time, and real-time preprocessed vibration data processed by the vibration data preprocessing unit based on the real-time vibration data and linked to the real-time vibration measurement time, converts the real-time preprocessed current data and the real-time preprocessed vibration data into second time processed current data and second time processed vibration data whose time series have been adjusted so that they are regarded as data acquired at the same time, and inputs the second time processed current data and the second time processed vibration data to the learned motor vibration model, thereby predicting a vibration state of the motor after the time when real-time data is acquired; A determination unit that detects an abnormality in the electric motor based on the predicted vibration state of the electric motor. A motor diagnostic system comprising:

6. a first step of generating preprocessed current data by extracting a plurality of types of feature amounts from input current data of an electric motor to be diagnosed and linking the input current data to an acquisition time of the input current data, and generating preprocessed vibration data by extracting a plurality of types of feature amounts from vibration data acquired from the electric motor and linking the feature amounts to an acquisition time of the vibration data; a second step of adjusting a time series of the preprocessed current data and the preprocessed vibration data so that the preprocessed current data and the preprocessed vibration data are regarded as data acquired at the same first time, and converting the data into first time processed current data and first time processed vibration data; a third step of learning a motor vibration model using the first time, the first time-processed current data, and the first time-processed vibration data as a learning data set; a fourth step of verifying the motor vibration model; a fifth step of returning to the first step if the verification result in the fourth step does not satisfy a predetermined accuracy; a sixth step of generating real-time preprocessed current data by linking the measured input current value to a real-time current measurement time so that the measured real-time input current value of the motor can be treated as real-time current data, generating real-time preprocessed vibration data by linking the measured vibration value to a real-time vibration measurement time so that the measured real-time vibration value of the motor can be treated as real-time vibration data, and adjusting the time series so that the real-time preprocessed current data and the real-time preprocessed vibration data are regarded as data acquired at the same second time, thereby converting them into second-time processed current data and second-time processed vibration data; a seventh step of predicting a vibration state of the motor after the second time by inputting the second time-processed current data and the second time-processed vibration data into a learned motor vibration model; an eighth step of returning to the sixth step if the predicted vibration of the electric motor does not exceed a predetermined threshold; and a ninth step of executing a predetermined process corresponding to an abnormality when the predicted vibration exceeds the predetermined threshold. A motor diagnostic program comprising:

Images