Motor abnormity identification method fusing low-frequency characteristics and auto-encoder

Abstract

The invention relates to the field of degradation trend evaluation of rotary machinery, in particular to a motor abnormity identification method fusing low-frequency characteristics and an auto-encoder, and the method comprises the following steps: S1, collecting a vibration signal of target motor equipment; s2, data preprocessing; s3, building an auto-encoder network model; s4, training an auto-encoder network model; s5, calculating an equipment anomaly detection threshold value index; and S6, deploying applications. According to the method, vibration signal low-frequency feature mapping of the healthy operation state of the motor is self-learned through the self-encoder, the mean square error of the reconstructed feature and the original low-frequency feature is tracked, and the statistical abnormality trigger threshold value is calculated, so that feature information loss caused by direct extraction of a vibration signal monitoring index is avoided; extraction of original vibration low-frequency signal features is enhanced, the abnormal state of the motor is estimated in advance, and early warning and operation and maintenance of early degradation of the motor are achieved.

Description

Technical Field

[0001] The present invention relates to the field of rotating machinery degradation trend assessment, and in particular to a motor abnormality recognition method integrating low-frequency features and autoencoders. Background Art

[0002] Electric motors play a vital role in industrial production, but their abnormal failures may lead to unplanned downtime, power waste, and increased maintenance costs. Therefore, accurate and timely identification of abnormal motor conditions is crucial to ensure production safety and operational efficiency.

[0003] The core of motor abnormality identification lies in how to learn and extract abnormal monitoring indicators. The actual operating conditions are often more complicated. If the characteristic indicators are not extracted accurately, the trend tracking will be biased, and then the equipment operating status may be falsely reported or missed, which will have an adverse impact on industrial production. Therefore, the present invention focuses on the problem of abnormal state identification of motors in actual industrial production to meet the actual engineering requirements of abnormal operation status detection of equipment.

[0004] Existing anomaly identification methods are often based on a single data source, such as temperature or current, or extract characteristic values ​​that reflect the vibration state of the equipment, such as vibration intensity RMS or peak-to-peak values, track their trends, and perform anomaly identification and early warning based on set operation deterioration thresholds. The shortcoming is that the extracted single characteristic value focuses on one aspect of the equipment's operating conditions and cannot fully reflect and evaluate whether the overall performance of the equipment has deteriorated. When the equipment is in an abnormal state, there is an operational risk of deviations in trend tracking, frequent triggering of false alarms, and missed alarms. Summary of the invention

[0005] The purpose of the present invention is to overcome the deficiencies of the prior art and provide a motor abnormality identification method that integrates low-frequency features and autoencoders. The autoencoder self-learns the low-frequency feature mapping of the vibration signal of the healthy operation state of the motor, tracks the mean square error between the reconstructed feature and the original low-frequency feature, and calculates its statistical abnormal trigger threshold, thereby avoiding the loss of feature information caused by directly extracting the vibration signal monitoring index, enhancing the extraction of the original vibration low-frequency signal features, and then predicting the abnormal state of the motor in advance, realizing early warning and operation and maintenance of the early deterioration of the motor.

[0006] To achieve the above-mentioned purpose of the invention, the present invention provides a technical solution: a method for identifying motor abnormality by integrating low-frequency features and autoencoders, which includes the following steps:

[0007] S1. Data acquisition: Collect vibration signals of target motor equipment.

[0008] S2. Data preprocessing: The collected vibration acceleration signal is converted into spectrum data through FFT, and the spectrum amplitude sequence below the set frequency value F is normalized to the interval [0, 1] and used as the input feature quantity for autoencoder training. The healthy status data of the stable operation section of the equipment is taken as the training set, and the status data of the performance degradation section is taken as the test set. At the same time, considering that the model is an unsupervised learning model, no label data is required. Training set ,in is the input feature set of the autoencoder training set, is the output feature set of the autoencoder training set. Test set ,in is the output feature set of the autoencoder test set, is the output feature set of the autoencoder test set.

[0009] S3. Build an autoencoder network model: including input layer, encoding layer, intermediate layer, decoding layer and output layer. The encoding layer maps the input data to a low-dimensional feature representation space, the intermediate layer acts as a bridge, and the decoding layer decodes the low-dimensional feature vector into reconstructed data. The model is based on minimizing the error between the reconstructed data and the input data so that the model can learn the most comprehensive input data features.

[0010] S4. Train the autoencoder network model: Use the training set in step S3 to train the autoencoder model that integrates low-frequency features, so that the loss function between the reconstructed data output by it and the original input data vector is minimized. The formula is:

[0011]

[0012] in is the characteristic component of the decoding layer output, is the component of the input data feature.

[0013] S5. Calculate the threshold index for abnormal detection of equipment: According to the autoencoder model trained in step S4, the root mean square error sequence of the input feature quantity of the training set and the reconstructed output quantity Based on the mean and standard deviation of the sequence, the anomaly detection threshold is calculated. The threshold formula is:

[0014]

[0015] in is the root mean square error sequence The mean of is the root mean square error sequence The standard deviation of .

[0016] S6. Deployment and application: Apply the autoencoder model trained in step S4 to the low-frequency data of the spectrum of the newly collected vibration data to perform abnormal detection of the motor operation status.

[0017] Preferably, the spectrum amplitude components in S2 mainly include frequency components below the frequency value F.

[0018] Preferably, the autoencoder model in S3 includes at least an input layer-encoding layer structure and a decoding layer-output layer structure.

[0019] Preferably, the autoencoder model in S3 adopts the small batch normalization principle during the input sample training process.

[0020] The present invention has the following beneficial effects:

[0021] 1. The present invention proposes a motor abnormality identification method that integrates low-frequency features and autoencoders, collects vibration data from the bearing end of the target motor equipment, uses the low-frequency amplitude information of the healthy vibration of the equipment as a training set, and uses it to train the constructed low-frequency autoencoder model. After the training is completed, it is used to calculate the abnormal detection threshold and deploy the application to identify the abnormal state of the target vibration signal data. Among them, the autoencoder model proposed by this method can perform unsupervised learning on the equipment operation characteristics, adaptively extract the inherent characteristic information of the vibration signal, avoid the loss of characteristic information caused by artificial calculation of a single indicator, and improve the ability to identify and warn of abnormal states during equipment operation.

[0022] 2. The present invention proposes a motor abnormality identification method that integrates low-frequency features and autoencoders. The low-frequency processing of vibration signals is simple and easy to operate, reducing the workload of preprocessing. In addition, the parameters of the autoencoder model are streamlined, which improves training and computing efficiency, and is convenient for later platform deployment and batch processing of industrial equipment big data. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 is a flow chart of the present invention;

[0024] Figure 2 is a schematic diagram of the autoencoder model;

[0025] Figure 3 are the bearing end vibration signals in the healthy state and the degraded state of the data set shown in the embodiment;

[0026] Figure 4 is a series of bearing end vibration acceleration spectra in the healthy state and the degraded state of the data set shown in the embodiment;

[0027] Figure 5 is the amount of data contained in the data set shown in the embodiment;

[0028] Figure 6 is the loss function corresponding to the autoencoder model training process shown in the embodiment;

[0029] Figure 7 is the vibration acceleration RMS trend change and the corresponding abnormality identification threshold line shown in the embodiment;

[0030] Figure 8 The embodiment shows that the original time domain vibration signal is input into the autoencoder to construct the abnormality recognition index trend change and the corresponding abnormality recognition threshold line;

[0031] Fig. 9 The embodiment shows that the fusion low frequency and autoencoder are used to construct the abnormal identification indicator trend change and the abnormal identification threshold line; DETAILED DESCRIPTION

[0032] The following will be combined with the accompanying drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments in the present invention, they all belong to the scope of protection of the present invention.

[0033] The specific implementation of the present invention is described below in conjunction with the accompanying drawings so that those skilled in the art can better understand the present invention. It should be noted that in the following description, when the detailed description of known functions and designs may dilute the main content of the present invention, these descriptions will be omitted here.

[0034] Please refer to Figure 1-Figure 8 The present invention provides a technical solution: a method for identifying motor abnormality by integrating low-frequency features and autoencoders, comprising the following steps:

[0035] S1. Data acquisition: Collect vibration signals of target motor equipment.

[0036] S2. Data preprocessing: The collected vibration acceleration signal is converted into spectrum data through FFT, and the spectrum amplitude sequence below 1000Hz is normalized to the interval [0, 1] and used as the input feature quantity for autoencoder training. The healthy status data of the stable operation section of the equipment is taken as the training set, and the status data of the performance degradation section is taken as the test set. At the same time, considering that the model is an unsupervised learning model, no label data is required. Training set ,in is the input feature set of the autoencoder training set, is the output feature set of the autoencoder training set. Test set ,in is the output feature set of the autoencoder test set, is the output feature set of the autoencoder test set.

[0037] S3. Build an autoencoder network model: including input layer, encoding layer, intermediate layer, decoding layer and output layer. The encoding layer maps the input data to a low-dimensional feature representation space, the intermediate layer acts as a bridge, and the decoding layer decodes the low-dimensional feature vector into reconstructed data. The model is based on minimizing the error between the reconstructed data and the input data so that the model can learn the most comprehensive input data features.

[0038] Figure 2 The autoencoder model architecture used in this embodiment is shown, which includes 1 input layer, 1 encoding layer, 1 intermediate layer, 1 decoding layer and 1 output layer.

[0039] S4. Train the autoencoder network model: Use the training set in step S3 to train the autoencoder model that integrates low-frequency features, so that the loss function between the reconstructed data output by it and the original input data vector is minimized. The formula is:

[0040]

[0041] in is the characteristic component of the decoding layer output, is the component of the input data feature.

[0042] S5. Calculate the threshold index for abnormal detection of equipment: According to the autoencoder model trained in step S4, the root mean square error sequence of the input feature quantity of the training set and the reconstructed output quantity Based on the mean and standard deviation of the sequence, the anomaly detection threshold is calculated. The threshold formula is:

[0043]

[0044] in is the root mean square error sequence The mean of is the root mean square error sequence The standard deviation of .

[0045] S6. Deployment and application: Apply the autoencoder model trained in step S4 to the low-frequency data of the spectrum of the newly collected vibration data to perform abnormal detection of the motor operation status.

[0046] In order to better illustrate the technical effect of the present invention, a specific example is used to test and verify the present invention. The data used in the embodiment is a vibration operation data set of an asynchronous motor driven by a coal crusher in a coal-fired power plant. The data set contains acceleration vibration data in healthy and degraded states, such as Figure 3 The corresponding spectrum data is shown as Figure 4 The number of samples in the data set is shown in Figure 5 shown.

[0047] In this embodiment, the input layer size is 1*2250, and the corresponding number of neurons is 2250. The encoding layer size is 2250*100, the intermediate layer size is 100*50, the decoding layer size is 50*100, and the output layer size is 100*2250, and the corresponding number of neurons is 2250.

[0048] In this embodiment, Figure 5 In the sample data shown in , 400 sets of training samples running in healthy state are taken for model training. The loss function of the model training process is as follows Figure 6 Compared with two widely used methods: (1) vibration acceleration RMS; (2) using the original vibration signal as input to the autoencoder to construct an abnormality recognition index, the results are shown in Figure 2. Figure 7 and Figure 8 As shown. It can be seen that when the equipment is abnormal, although the characteristics of the two existing methods change, the growth is not obvious and the fluctuation is large; while the method proposed in the present invention has obvious growth and small fluctuation, which greatly reduces the impact of characteristic quantity changes caused by operating condition fluctuations. The characteristic trend becomes more obvious as the motor deteriorates, indicating that the method proposed in the present invention is more sensitive to faults and has good robustness. Moreover, compared with the two existing methods, the method proposed in the present invention can detect abnormalities earlier, about 58.5 hours earlier than method (1) and about 203 hours earlier than method (2), indicating that the method proposed in the present invention has excellent recognition ability for early abnormalities, such as Fig. 9 shown.

[0049] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A motor abnormality identification method integrating low-frequency features and autoencoder, characterized in that: The following steps are involved: S1. Data acquisition: Collect vibration signals of target motor equipment. S2. Data preprocessing: The collected vibration acceleration signal is converted into spectrum data through FFT, and the spectrum amplitude sequence below the set frequency value F is normalized to the interval [0, 1] and used as the input feature quantity for autoencoder training. The healthy status data of the stable operation section of the equipment is taken as the training set, and the status data of the performance degradation section is taken as the test set. At the same time, considering that the model is an unsupervised learning model, no label data is required. Training set ,in is the input feature set of the autoencoder training set, is the output feature set of the autoencoder training set. Test set ,in is the output feature set of the autoencoder test set, is the output feature set of the autoencoder test set. S3. Build an autoencoder network model: including input layer, encoding layer, intermediate layer, decoding layer and output layer. The encoding layer maps the input data to a low-dimensional feature representation space, the intermediate layer acts as a bridge, and the decoding layer decodes the low-dimensional feature vector into reconstructed data. The model is based on minimizing the error between the reconstructed data and the input data so that the model can learn the most comprehensive input data features. S4. Train the autoencoder network model: Use the training set in step S3 to train the autoencoder model that integrates low-frequency features, so that the loss function between the reconstructed data output by it and the original input data vector is minimized. The formula is: in is the characteristic component of the decoding layer output, is the component of the input data feature. S5. Calculate the threshold index for abnormal detection of equipment: According to the autoencoder model trained in step S4, the root mean square error sequence of the input feature quantity of the training set and the reconstructed output quantity Based on the mean and standard deviation of the sequence, the anomaly detection threshold is calculated. The threshold formula is: in is the root mean square error sequence The mean of is the root mean square error sequence The standard deviation of . S6. Deployment and application: Apply the autoencoder model trained in step S4 to the low-frequency data of the spectrum of the newly collected vibration data to perform abnormal detection of the motor operation status.

2. A motor abnormality identification method integrating low-frequency features and autoencoder according to claim 1, characterized in that: The vibration signal collected in S1 is not limited to motor equipment, but also applicable to other rotating and reciprocating mechanical equipment, such as gear boxes, fans, pumps and compressors. The vibration data preprocessing step 1 in S2 is: Normalization is performed. The autoencoder network model in S3 includes at least one encoding layer and one decoding layer. If the number of layers is increased, the structure between the input layer to the encoding layer and the decoding layer to the output layer should be kept symmetrical. The number of network neurons from the input layer to the encoding layer adopts a dimensionality reduction mode, and the number of network neurons from the decoding layer to the output layer adopts a dimensionality increase mode. The minimum mean square error between the reconstructed data of the output layer and the input data is used as the optimization objective function, but is not limited to the optimization function. The gradient update of the autoencoder network model adopts an adaptive low-order moment estimation algorithm, but is not limited to the gradient update algorithm. The input layer and the output layer sample linear activation functions, while the encoding layer and the decoding layer adopt the design idea of ​​the Relu activation function, but are not limited to the order and type of use of the activation function. The specific formula for the normalization process in S2 is: in is a low frequency amplitude sequence The maximum value of is a low frequency amplitude sequence The minimum amplitude value of is the normalized amplitude component.

Images