Fault diagnosis method based on time-frequency angular domain three-channel parallel convolutional network
By using a three-channel parallel convolutional network in time-frequency angle domain in fault diagnosis, the problems of high complexity and long diagnosis time in the existing technology are solved, and efficient and accurate fault diagnosis is achieved.
Patent Information
- Application Number
- CN202411739681.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-29
- Publication Date
- 2025-03-07
AI Technical Summary
Existing machine learning methods are complex in the troubleshooting process, resulting in a long diagnosis time.
The fault diagnosis method based on the time-frequency angle domain three-channel parallel convolution network is adopted. By obtaining vibration signals and performing noise reduction processing, Fourier transform and angle resampling, the time-domain, frequency domain and angle domain signals are obtained, and then the multi-channel parallel convolutional neural network model is used to feature extraction and fusion of these signals, and finally fault diagnosis is achieved.
This method can improve the accuracy and efficiency of fault diagnosis, simplify the feature extraction process, reduce the complexity of diagnosis, and improve the efficiency of fault diagnosis in the context of big data.
Smart Images
Figure CN119577583A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of fault diagnosis. Background Art
[0002] With the rapid development of industrial production, equipment fault diagnosis has become the key to ensure the normal operation and safe operation of equipment. Due to the large amount of data for mechanical equipment status monitoring, deep learning technology has been widely used in the field of rotating machinery fault diagnosis. In the traditional vibration signal analysis process, there are high requirements for the basic theory of vibration signal analysis and corresponding work experience. Compared with traditional fault diagnosis methods, deep learning algorithms have the advantages of strong generalization ability, high recognition accuracy, and automatic extraction of mechanical equipment fault characteristics from experimental data without human intervention.
[0003] Gearboxes play a vital role in mechanical transmission systems, transmitting power between shafts and expected to work 24 hours a day in production systems. Any failure of the gearbox may lead to unnecessary downtime, expensive repair costs, and even casualties. Therefore, it is crucial to detect and diagnose faults at the initial stage. Fault diagnosis, as an effective component of condition-based maintenance, has received much attention in the safe operation of gearboxes.
[0004] In view of the problem that the fault diagnosis process of traditional machine learning methods is complicated and prone to omission of fault information, some people use deep learning methods based on convolutional neural networks to study the fault diagnosis technology of offshore wind turbine gearboxes. The time-frequency diagram of the vibration signal obtained by continuous wavelet transform is directly used as input, and the convolutional neural network is used to automatically extract features and perform fault diagnosis. The simulation verification results show that the method can accurately diagnose gear faults, and the accuracy of the diagnosis results of limited data is 100%. Compared with the machine learning method, the fault feature extraction process is simplified and the fault diagnosis accuracy is improved.
[0005] Some people have proposed an intelligent fault diagnosis method for bearings under different conditions based on angular domain resampling, symmetrical dot pattern (SDP) and deep convolutional neural network (DCNN). In order to further improve the proportion of fault information in the signal, CEEMDAN is applied to the denoising and reconstruction of angular domain vibration signals, and the denoised and reconstructed signals are visualized by SDP, where the SDP parameters are determined by the Pearson correlation coefficient. In addition, the SDP patterns are formed by mapping different reconstructed fault signals into the polar coordinate system, and the formed SDP patterns are classified by DCNN to achieve fault diagnosis.
[0006] Although the above two methods have high diagnostic accuracy, they do not take into account the complexity of feature extraction and its impact on diagnosis time. Summary of the invention
[0007] The present invention aims to solve the problems of high complexity and long diagnosis time in the fault diagnosis process of the existing machine learning method, and now provides a fault diagnosis method based on three-channel parallel convolutional network in time-frequency angle domain.
[0008] The fault diagnosis method based on three-channel parallel convolutional network in time-frequency angle domain of the present invention comprises:
[0009] A vibration signal of a target rotating machine is obtained, and the vibration signal is used as an input of a convolutional diagnosis model. Fault diagnosis of the target rotating machine is performed through the convolutional diagnosis model, wherein the convolutional diagnosis model is obtained through pre-training, and the training method of the convolutional diagnosis model is:
[0010] Perform noise reduction, Fourier transform and angle resampling on the sample vibration signal respectively; obtain time domain signal, frequency domain signal and angle domain signal;
[0011] Performing standardization processing on the time domain signal, the frequency domain signal and the angle domain signal to obtain a standardized data set;
[0012] Dividing the standardized data set into a training set, a test set, and a validation set;
[0013] A multi-channel parallel convolutional neural network model is established, and the multi-channel parallel convolutional neural network model is trained using the training set, and then the test set and the validation set are used for testing and verification; the verification meets the accuracy requirements and the training of the convolutional diagnostic model is completed.
[0014] Further, in the present invention, the multi-channel parallel convolutional neural network model includes three single convolution channels, a fusion layer, a flattening layer, a fully connected layer W1, a fully connected layer W2 and an output layer;
[0015] The three single convolution channels are used to extract features of the time domain signal, the frequency domain signal and the angle domain signal, respectively, to obtain the time domain feature signal, the frequency domain feature signal and the angle domain feature signal;
[0016] The fusion layer is used to perform weighted summation on the signals of each channel to obtain a set of sum signals, and then transmit the sum signals to the flattening layer;
[0017] The flattening layer is used to reduce the dimension of the sum signal and then transmit it to the output layer after propagation through the fully connected layer W1 and the fully connected layer W2;
[0018] The output layer uses a Softmax activation function to judge the fault category, normalizes a received numerical vector into a fault category probability distribution vector, and takes the category with the highest fault probability as the final fault category.
[0019] Furthermore, in the present invention, before the time domain signal, the frequency domain signal and the angle domain signal are input into a single convolution channel, the two-dimensional data is decomposed into three-dimensional data through an input layer.
[0020] Furthermore, in the present invention, the single convolution channel includes, from the input end to the output end, in sequence: convolution layer 1, pooling layer 1, convolution layer 2, pooling layer 2, convolution layer 3, pooling layer 3, convolution layer 4, pooling layer 4, convolution layer 5, pooling layer 5, convolution layer 6, pooling layer 6 and fully connected layer 1.
[0021] Furthermore, in the present invention, the Softmax activation function is:
[0022]
[0023] Where x is the real input vector; x i and x j are the i-th and j-th elements in the input vector; n is the total number of elements in the input vector, each element of the input vector corresponds to a fault category; is the probability corresponding to the i-th fault category.
[0024] Firstly, the present invention can realize accurate identification of gearbox faults with high fault identification accuracy. Secondly, the use of time domain, frequency domain and angular domain ensures that the feature complexity is improved without affecting the diagnosis time. Finally, a convolutional neural network is used to automatically extract features from the original vibration time domain signal of the gearbox fault, replacing manual feature extraction and feature engineering, thereby improving the efficiency of gearbox fault diagnosis in the context of big data; through the divided validation set data verification, the network model can extract different signal features through different channels, and after the fusion of signal features, it can realize gearbox fault diagnosis with high accuracy, and effectively reduce the complexity of fault diagnosis and improve the diagnostic efficiency. BRIEF DESCRIPTION OF THE DRAWINGS
[0025] Figure 1 It is a time domain characteristic diagram involved in the present invention;
[0026] Figure 2 is a schematic diagram of Fourier transform involved in the present invention;
[0027] Figure 3 is a frequency amplitude diagram to which the present invention relates;
[0028] Figure 4 It is a schematic diagram of the angle domain transformation involved in the present invention;
[0029] Figure 5 It is a structural diagram of a three-channel convolution model involved in the present invention;
[0030] Figure 6 This is a confusion matrix result diagram involved in the present invention. DETAILED DESCRIPTION
[0031] 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 of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work belong to the scope of protection of the present invention. It should be noted that the embodiments of the present invention and the features in the embodiments can be combined with each other without conflict.
[0032] Specific implementation method 1: refer to Figures 1 to 6 Specifically describing this embodiment, the fault diagnosis method based on three-channel parallel convolutional network in time-frequency angle domain described in this embodiment includes:
[0033] A vibration signal of a target rotating machine is obtained, and the vibration signal is used as an input of a convolutional diagnosis model. Fault diagnosis of the target rotating machine is performed through the convolutional diagnosis model, wherein the convolutional diagnosis model is obtained through pre-training, and the training method of the convolutional diagnosis model is:
[0034] Perform noise reduction, Fourier transform and angle resampling on the sample vibration signal respectively; obtain time domain signal, frequency domain signal and angle domain signal;
[0035] Performing standardization processing on the time domain signal, the frequency domain signal and the angle domain signal to obtain a standardized data set;
[0036] Dividing the standardized data set into a training set, a test set, and a validation set;
[0037] A multi-channel parallel convolutional neural network model is established, and the multi-channel parallel convolutional neural network model is trained using the training set, and then the test set and the validation set are used for testing and verification; the verification meets the accuracy requirements and the training of the convolutional diagnostic model is completed.
[0038] Further, in the present invention, the multi-channel parallel convolutional neural network model includes three single convolution channels, a fusion layer, a flattening layer, a fully connected layer W1, a fully connected layer W2 and an output layer;
[0039] The three single convolution channels are used to extract features of the time domain signal, the frequency domain signal and the angle domain signal, respectively, to obtain the time domain feature signal, the frequency domain feature signal and the angle domain feature signal;
[0040] The fusion layer is used to perform weighted summation on the signals of each channel to obtain a set of sum signals, and then transmit the sum signals to the flattening layer;
[0041] The flattening layer is used to reduce the dimension of the sum signal and then transmit it to the output layer after propagation through the fully connected layer W1 and the fully connected layer W2;
[0042] The output layer uses a Softmax activation function to judge the fault category, normalizes a received numerical vector into a fault category probability distribution vector, and takes the category with the highest fault probability as the final fault category.
[0043] Furthermore, in the present invention, before the time domain signal, the frequency domain signal and the angle domain signal are input into a single convolution channel, the two-dimensional data is decomposed into three-dimensional data through an input layer.
[0044] Furthermore, in the present invention, the single convolution channel includes, from the input end to the output end, in sequence: convolution layer 1, pooling layer 1, convolution layer 2, pooling layer 2, convolution layer 3, pooling layer 3, convolution layer 4, pooling layer 4, convolution layer 5, pooling layer 5, convolution layer 6, pooling layer 6 and fully connected layer 1.
[0045] Furthermore, in the present invention, the Softmax activation function is:
[0046]
[0047] Where x is the real input vector; x i and x j are the i-th and j-th elements in the input vector; n is the total number of elements in the input vector, each element of the input vector corresponds to a fault category; is the probability corresponding to the i-th fault category.
[0048] The specific process of using the method of the present invention to diagnose gear faults is as follows:
[0049] S1. Build a fault simulation test bench, simulate different degrees of damage to key components, simulate local damage faults such as gear tooth breakage, peeling, wear, pitting, and various common operating signals, and use acceleration sensors to collect experimental data to form the required data set. Use 80% of the samples as the training set to train the neural network parameters, use 50% of the remaining 20% of the samples as the validation set to verify the feasibility of the training model, and use the remaining 10% of the samples as the test set for the estimated test of the final model.
[0050] S2. De-noise the divided data set to improve the signal-to-noise ratio of the signal. The de-noised data is used as a time domain signal, and then Fourier transform is performed on it to obtain a frequency domain signal, and angle resampling is performed to obtain an angle domain signal.
[0051] S3. Dataset standardization: One of the conditions that deep learning must meet is that the training set, validation set, and test set must be under the same distribution. Therefore, the mean and variance of the training set must be extracted to achieve training set standardization, and the mean and variance of the training set must be used to standardize the validation set and test set. In this way, the three data sets meet the same distribution.
[0052] S4. Use the processed data to train the model. The three channels of the multi-channel parallel convolutional neural network model use time domain data, frequency domain data, and angular domain data as input respectively, and then flatten and connect the outputs of each channel after convolution for fault diagnosis.
[0053] S5. Finally, the model is evaluated using the validation set and test set, and the loss graph and accuracy graph of the training set and validation set are output. The degree of fault classification of the test set is observed through the confusion matrix.
[0054] In summary, the actual operation process of this method is shown in the figure.
[0055] Although the present invention is described herein with reference to specific embodiments, it should be understood that these embodiments are merely examples of the principles and applications of the present invention. It should therefore be understood that many modifications may be made to the exemplary embodiments and that other arrangements may be devised without departing from the spirit and scope of the present invention as defined by the appended claims. It should be understood that the various dependent claims and features described herein may be combined in a manner different from that described in the original claims. It should also be understood that the features described in conjunction with a single embodiment may be used in other described embodiments.
Claims
1. A fault diagnosis method based on three-channel parallel convolutional network in time-frequency angle domain, characterized in that: include: A vibration signal of a target rotating machine is obtained, and the vibration signal is used as an input of a convolutional diagnosis model. Fault diagnosis of the target rotating machine is performed through the convolutional diagnosis model. The convolutional diagnosis model is obtained through pre-training. The training method of the convolutional diagnosis model is: Perform noise reduction, Fourier transform and angle resampling on the sample vibration signal respectively; obtain time domain signal, frequency domain signal and angle domain signal; Performing standardization processing on the time domain signal, the frequency domain signal and the angle domain signal to obtain a standardized data set; Dividing the standardized data set into a training set, a test set, and a validation set; A multi-channel parallel convolutional neural network model is established, the multi-channel parallel convolutional neural network model is trained using the training set, and then the test set and the validation set are used for testing and verification; the verification meets the accuracy requirements and the training of the convolutional diagnostic model is completed.
2. The fault diagnosis method based on three-channel parallel convolutional network in time-frequency angle domain according to claim 1 is characterized in that: The multi-channel parallel convolutional neural network model includes three single convolution channels, a fusion layer, a flattening layer, a fully connected layer W1, a fully connected layer W2, and an output layer; The three single convolution channels are used to extract features of the time domain signal, the frequency domain signal and the angle domain signal, respectively, to obtain the time domain feature signal, the frequency domain feature signal and the angle domain feature signal; The fusion layer is used to perform weighted summation on the signals of each channel to obtain a set of sum signals, and then transmit the sum signals to the flattening layer; The flattening layer is used to reduce the dimension of the sum signal and then transmit it to the output layer after propagation through the fully connected layer W1 and the fully connected layer W2; The output layer uses the Softmax activation function to judge the fault category, normalizes a received numerical vector into a fault category probability distribution vector, and takes the category with the highest fault probability as the final fault category.
3. The fault diagnosis method based on three-channel parallel convolutional network in time-frequency angle domain according to claim 1 is characterized in that: Before the time domain signal, frequency domain signal and angle domain signal are input into a single convolution channel, the two-dimensional data is decomposed into three-dimensional data through the input layer.
4. The fault diagnosis method based on three-channel parallel convolutional network in time-frequency angle domain according to claim 2 is characterized in that: The single convolution channel includes, from the input end to the output end, convolution layer 1, pooling layer 1, convolution layer 2, pooling layer 2, convolution layer 3, pooling layer 3, convolution layer 4, pooling layer 4, convolution layer 5, pooling layer 5, convolution layer 6, pooling layer 6 and fully connected layer 1.
5. The fault diagnosis method based on three-channel parallel convolutional network in time-frequency angle domain according to claim 2 is characterized in that: The Softmax activation function is: Where x is the real input vector; x i and x j are the i-th and j-th elements in the input vector; n is the total number of elements in the input vector, each element of the input vector corresponds to a fault category; is the probability corresponding to the i-th fault category.