Bearing fault diagnosis method and system based on multi-scale wavelet thresholding network

Abstract

The present invention relates to a bearing fault diagnosis method and system based on a multi-scale wavelet thresholding network. The method is as follows: bearing fault vibration signals are collected at a preset sampling rate to form a bearing fault data set; the bearing fault data set is divided into a training set and a test set according to a preset ratio; the one-dimensional bearing fault vibration signal of the training set is input into a fault diagnosis model for training, the value of the overall loss function of the model is calculated, and the model parameters are updated using the Adam optimization algorithm; iterative training is performed until the model training converges to obtain a trained model; the data of the test set is input into the trained model for data classification, the loss and accuracy are calculated, and bearing fault diagnosis is achieved. During fault diagnosis, the present invention utilizes the soft thresholding structure in the deep residual shrinkage network in combination with the multi-scale convolutional wavelet decomposition module to achieve multi-scale wavelet decomposition and adaptive filtering, which is conducive to improving the accuracy of fault diagnosis.

Description

Technical Field

[0001] The present invention belongs to the technical field of bearing fault diagnosis, and in particular relates to a bearing fault diagnosis method and system based on a multi-scale wavelet thresholding network. Background Art

[0002] In recent years, intelligent manufacturing technology has continued to develop and improve. With the continuous integration of technologies such as artificial intelligence, big data, and the Internet of Things with traditional industries, the level of intelligence in the health monitoring of industrial machinery and equipment continues to improve. Bearings, as key components of most rotating machinery, often operate under complex operating conditions, with a high probability of failure. Bearing failures not only affect the safety and reliability of equipment but can also easily cause significant economic losses and casualties. Therefore, research on bearing fault diagnosis methods is crucial for timely and effective repairs.

[0003] Traditional fault diagnosis methods are based on fault feature extraction and expert models, requiring extensive expert knowledge. Human factors significantly interfere with the diagnostic results, making them unable to meet the requirements for intelligent health monitoring of rotating machinery. In recent years, with the rapid development of artificial intelligence technology, bearing fault diagnosis methods based on deep learning have been widely used in industrial scenarios, effectively addressing the aforementioned issues and gradually becoming a research hotspot. To effectively extract features from bearing vibration signals, existing signal processing methods often use time-frequency analysis methods for signal preprocessing before inputting them into a model for further processing. However, this approach prevents the time-frequency analysis process from being well integrated into the subsequent neural network, and the time-frequency analysis method lacks adaptability. Summary of the Invention

[0004] In view of the above-mentioned deficiencies in the prior art, the object of the present invention is to provide a bearing fault diagnosis method and system based on a multi-scale wavelet thresholding network.

[0005] In order to achieve the above-mentioned object of the invention, the present invention adopts the following technical solutions:

[0006] A bearing fault diagnosis method based on a multi-scale wavelet thresholding network comprises the following steps:

[0007] S1. Collect bearing fault vibration signals at a preset sampling rate to form a bearing fault data set;

[0008] S2. Divide the bearing fault dataset into a training set and a test set according to a preset ratio;

[0009] S3. Inputting the one-dimensional bearing fault vibration signal of the training set into the fault diagnosis model for training. The training process of the fault diagnosis model includes: performing frequency division analysis on the bearing fault vibration signal through a multi-scale convolutional wavelet decomposition network; wherein the multi-scale convolutional wavelet decomposition network includes a single-channel convolutional wavelet decomposition network with three convolution kernels of different scales; the convolutional wavelet decomposition networks of different channels are used to extract time-frequency features with different characteristics; using the parallel network DRSN-BiGRU of the deep residual shrinkage network and BiGRU to further extract the output features of the multi-scale convolutional wavelet decomposition network; and using the fully connected layer and the dropout layer to classify the output features of the DRSN-BiGRU.

[0010] S4. Calculate the value of the overall loss function of the model and use the Adam optimization algorithm to update the model parameters;

[0011] S5. Iterate steps S3 to S4 until the model training converges and a trained model is obtained;

[0012] S6. Input the test set data into the trained model to classify the data, calculate the loss and accuracy, and realize bearing fault diagnosis.

[0013] As a preferred solution, in step S1, the preset sampling rate is 12KHz.

[0014] As a preferred solution, in step S2, the ratio of the training set to the test set is 7:3.

[0015] As a preferred solution, in step S3, the scales of the convolution kernels set in the multi-scale convolutional wavelet decomposition network are 4×1, 6×1 and 8×1 respectively, and each convolution kernel is initialized using the DB2, DB3 and DB4 wavelet basis functions respectively.

[0016] As a preferred solution, in step S3, the BiGRU is composed of a forward GRU and a reverse GRU, and the number of neural units of the forward GRU and the reverse GRU is set to 32.

[0017] As a preferred solution, in step S3, the parameter of the Dropout layer is set to 0.5.

[0018] As a preferred solution, in step S3, the learning rate of the Adam optimization algorithm used is set to 0.01.

[0019] As a preferred solution, in step S4, the number of iterations is 100.

[0020] As a preferred solution, in step S6, accuracy is used as the evaluation indicator of the model.

[0021] The present invention further provides a bearing fault diagnosis system based on a multi-scale wavelet thresholding network, applying the bearing fault diagnosis method described in any of the above solutions, the bearing fault diagnosis system comprising:

[0022] An acquisition module is used to collect bearing fault vibration signals at a preset sampling rate to form a bearing fault data set;

[0023] A data set partitioning module is used to partition the bearing fault data set into a training set and a test set according to a preset ratio;

[0024] The training module is used to input the one-dimensional bearing fault vibration signal of the training set into the fault diagnosis model for training. The training process of the fault diagnosis model includes: performing frequency analysis on the bearing fault vibration signal through a multi-scale convolutional wavelet decomposition network; wherein the multi-scale convolutional wavelet decomposition network includes three single-channel convolutional wavelet decomposition networks with convolution kernels of different scales; the convolutional wavelet decomposition networks of different channels are used to extract time-frequency features with different characteristics; using the parallel network DRSN-BiGRU of the deep residual shrinkage network and BiGRU to further extract the output features of the multi-scale convolutional wavelet decomposition network; and using the fully connected layer and the dropout layer to classify the output features of the DRSN-BiGRU.

[0025] The update module is used to calculate the value of the overall loss function of the model and update the model parameters using the Adam optimization algorithm;

[0026] The iteration module is used to iteratively train the model until the model training converges and a trained model is obtained;

[0027] The test module is used to input the test set data into the trained model to classify the data, calculate the loss and accuracy, and realize bearing fault diagnosis.

[0028] Compared with the prior art, the present invention has the following beneficial effects:

[0029] (1) The present invention proposes a new multi-scale convolutional wavelet decomposition module, which uses a multi-scale convolutional neural network to implement the traditional multi-level discrete wavelet decomposition process. On the one hand, the multi-scale convolution structure can more comprehensively extract the characteristics of bearing fault data. On the other hand, the use of a multi-scale convolutional neural network to implement multi-level discrete wavelet decomposition can make the time-frequency analysis process and the subsequent neural network achieve a good integration, and the time-frequency analysis has a higher degree of adaptability;

[0030] (2) The present invention uses a deep residual shrinkage network for feature extraction, wherein the deep residual shrinkage network contains a soft thresholding structure, which can be combined with a multi-scale convolutional wavelet decomposition module to filter out irrelevant components in the input data, thereby improving the effect of fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 1 is a flow chart of a bearing fault diagnosis method based on a multi-scale wavelet thresholding network according to an embodiment of the present invention;

[0032] Figure 2 4 is a network structure diagram of a fault diagnosis model according to an embodiment of the present invention. DETAILED DESCRIPTION

[0033] To more clearly illustrate the embodiments of the present invention, specific embodiments of the present invention will be described below with reference to the accompanying drawings. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings and other embodiments can be obtained based on these drawings without inventive efforts.

[0034] The bearing fault diagnosis method based on the multi-scale wavelet thresholding network of the embodiment of the present invention is as follows: Figure 1 As shown in FIG, which is a flow chart of the method, the effect of the method of the present invention on improving the accuracy of bearing fault diagnosis is described in detail according to the flow chart. Specifically, it includes a model training stage and a fault detection stage.

[0035] Specifically, the model training phase is carried out as follows:

[0036] In step S1, a bearing under different fault conditions is fabricated using an electrospark method. An accelerometer is used to collect bearing vibration signals under different fault conditions at a sampling rate of 12 kHz. The collected bearing vibration signals are divided into samples with 1200 sampling points per sample to obtain the final data set.

[0037] Step S2: Divide the data set into a training set and a test set according to a preset ratio. Specifically, the preset ratio is 7:3. Then, the training set is input into the fault diagnosis model for training. Figure 2 As shown in Figure 1, the fault diagnosis model includes a multi-scale convolutional wavelet decomposition module (composed of a multi-scale convolutional wavelet decomposition network), a feature extraction module (composed of DRSN-BiGRU) and a classification module (composed of a fully connected layer and a Dropout layer, and finally output through Softmax).

[0038] In step S3, the one-dimensional bearing fault vibration signal is input into a multi-scale convolutional wavelet decomposition network for time-frequency analysis, wherein the multi-scale convolutional wavelet decomposition network includes a single-channel convolutional wavelet decomposition network with three convolution kernels of different scales, and the scales of the convolution kernels are 4×1, 6×1, and 8×1, respectively.

[0039] The single-channel convolutional wavelet decomposition network is initialized with three different wavelet basis functions, namely DB2, DB3, and DB4. The convolutional wavelet decomposition networks of different channels will extract time-frequency features with different characteristics.

[0040] In step S4, the output of the multi-scale convolutional wavelet decomposition network contains rich time-frequency features, and the output features of the multi-scale convolutional wavelet decomposition network are further extracted using the parallel network DRSN-BiGRU of the deep residual shrinkage network and BiGRU.

[0041] Specifically, BiGRU consists of a forward GRU and a reverse GRU, and the number of neural units in the forward GRU and the reverse GRU is set to 32. The calculation formula for a single GRU is:

[0042] r t =σ(W rh h t-1 +W rx x t )

[0043] z t =σ(W zh h t-1 +W zx x t )

[0044]

[0045]

[0046] Among them, x t and h t-1 Respectively represent the input and output at the current moment; h t-1 Represents the output of the previous moment; r t With z t Represent the reset gate and update gate in the GRU module respectively; represents the output candidate value after the reset gate processing; W rh With W rx To reset the parameters in the gate; W zh With W zx is the parameter in the update gate; W hh With W hx To obtain the output candidate value Parameters in the process; tanh represents the activation function.

[0047] In step S5, the output features of the DRSN-BiGRU are classified using the fully connected layer and the Dropout layer to achieve fault diagnosis. The parameter of the Dropout layer is set to 0.5.

[0048] Step S6, model parameter update: Calculate the overall model loss function and update the model parameters using the Adam optimization algorithm with a learning rate of 0.01. Backpropagate the model weights and biases for fine-tuning.

[0049] Specifically, the value of the loss function is mainly the cross entropy loss function. Its formula is:

[0050]

[0051] Where C is the number of categories; y n is the one-hot encoding of the sample; p n Represents the predicted value.

[0052] Step S7, repeat steps S3 to S6, the number of iterations is 100 times; and determine whether the training has converged based on the loss function curve; if it has converged, save the current model network parameters.

[0053] The fault detection stage specifically includes the following steps:

[0054] Step Z1, obtaining fault diagnosis results: inputting the data of the bearing fault test set into the model network saved in step S7 to classify the data.

[0055] Step Z2, calculate the loss and accuracy, and finally realize the bearing fault diagnosis. The accuracy formula is:

[0056]

[0057] Among them, N True Represents the number of correctly classified samples; N Total Represents the total number of classification samples.

[0058] To validate the performance of the proposed method, we used the proposed method and four bearing fault diagnosis methods to classify a test data set of bearing faults. Accuracy was used as the evaluation metric. The final results are shown in Table 1. All experimental results were repeated five times and the average was taken to improve the validity of the experimental results. It can be seen that the proposed method achieved a high accuracy, demonstrating its superior performance.

[0059] Table 1 Accuracy of different methods

[0060] Methods Accuracy(%) CNN 95.88 CNN-BiLSTM 96.82 ResNet18 95.59 AlexNet 94.05 MSWTNET 99.52

[0061] Figure 2The network structure diagram of the method of the present invention is shown below. CWD(DB2), CWD(DB3), and CWD(DB4) represent single-channel convolutional wavelet decomposition networks initialized using the DB2, DB3, and DB4 wavelet basis functions, respectively. h represents a low-pass filter, and g represents a high-pass filter. ↓2 represents downsampling, DRSN represents a deep residual shrinkage network, and BiGRU represents a bidirectional gated recurrent unit.

[0062] Based on the above-mentioned bearing fault diagnosis method, an embodiment of the present invention also provides a bearing fault diagnosis system based on a multi-scale wavelet thresholding network, including an acquisition module, a data set partitioning module, a training module, an update module, an iteration module and a testing module.

[0063] Specifically, the acquisition module is used to collect bearing fault vibration signals at a preset sampling rate to form a bearing fault data set.

[0064] The data set partitioning module is used to divide the bearing fault data set into a training set and a test set according to a preset ratio.

[0065] The training module is used to input the one-dimensional bearing fault vibration signal of the training set into the fault diagnosis model for training. The training process of the fault diagnosis model includes: performing frequency division analysis on the bearing fault vibration signal through a multi-scale convolutional wavelet decomposition network; wherein, the multi-scale convolutional wavelet decomposition network includes a single-channel convolutional wavelet decomposition network with three convolution kernels of different scales; the convolutional wavelet decomposition networks of different channels are used to extract time-frequency features with different characteristics; the output features of the multi-scale convolutional wavelet decomposition network are further extracted using the parallel network DRSN-BiGRU of the deep residual shrinkage network and BiGRU; and the output features of the DRSN-BiGRU are classified using the fully connected layer and the Dropout layer.

[0066] The update module is used to calculate the value of the overall loss function of the model and use the Adam optimization algorithm to update the model parameters.

[0067] The iteration module is used to iteratively train the model until the model training converges and a trained model is obtained.

[0068] The test module is used to input the test set data into the trained model to classify the data, calculate the loss and accuracy, and realize bearing fault diagnosis.

[0069] The specific processing procedures of the above-mentioned functional modules can be referred to the detailed description in the above-mentioned bearing fault diagnosis method, which will not be repeated here.

[0070] This paper uses a multiscale convolutional wavelet decomposition network for time-frequency analysis and a deep residual shrinkage network and BiGRU for further feature extraction. The deep residual shrinkage network includes a soft thresholding structure. Combining soft thresholding with the multiscale convolutional wavelet decomposition network effectively filters out irrelevant components of the input features while retaining valid features. Finally, fully connected layers and dropout layers are used for classification. Experiments using the proposed fault diagnosis method ultimately achieved good fault diagnosis results.

[0071] The above description is only a detailed description of the preferred embodiments and principles of the present invention. For ordinary technicians in this field, based on the ideas provided by the present invention, there may be changes in the specific implementation methods, and these changes should also be considered as the scope of protection of the present invention.

Claims

1. A bearing fault diagnosis method based on a multi-scale wavelet thresholding network, characterized in that: The following steps are involved: S1. Collect bearing fault vibration signals at a preset sampling rate to form a bearing fault data set; S2. Divide the bearing fault dataset into a training set and a test set according to a preset ratio; S3. Inputting the one-dimensional bearing fault vibration signal of the training set into the fault diagnosis model for training. The training process of the fault diagnosis model includes: performing frequency division analysis on the bearing fault vibration signal through a multi-scale convolutional wavelet decomposition network; wherein the multi-scale convolutional wavelet decomposition network includes a single-channel convolutional wavelet decomposition network with three convolution kernels of different scales; the convolutional wavelet decomposition networks of different channels are used to extract time-frequency features with different characteristics; using the parallel network DRSN-BiGRU of the deep residual shrinkage network and BiGRU to further extract the output features of the multi-scale convolutional wavelet decomposition network; and using the fully connected layer and the dropout layer to classify the output features of the DRSN-BiGRU. S4. Calculate the value of the overall loss function of the model and use the Adam optimization algorithm to update the model parameters; S5. Iterate steps S3 to S4 until the model training converges and a trained model is obtained; S6. Input the test set data into the trained model to classify the data, calculate the loss and accuracy, and realize bearing fault diagnosis.

2. The bearing fault diagnosis method according to claim 1, characterized in that: In step S1, the preset sampling rate is 12KHz.

3. The bearing fault diagnosis method according to claim 1, characterized in that: In step S2, the ratio of the training set to the test set is 7:

3.

4. The bearing fault diagnosis method according to claim 1, characterized in that: In step S3, the scales of the convolution kernels set in the multi-scale convolutional wavelet decomposition network are 4×1, 6×1 and 8×1 respectively, and each convolution kernel is initialized using the DB2, DB3 and DB4 wavelet basis functions respectively.

5. The bearing fault diagnosis method according to claim 1, characterized in that: In step S3, the BiGRU is composed of a forward GRU and a reverse GRU, and the number of neural units of the forward GRU and the reverse GRU is set to 32.

6. The bearing fault diagnosis method according to claim 1, characterized in that: In step S3, the parameter of the Dropout layer is set to 0.

5.

7. The bearing fault diagnosis method according to claim 1, characterized in that: In step S3, the learning rate of the Adam optimization algorithm used is set to 0.

01.

8. The bearing fault diagnosis method according to claim 1, characterized in that: In step S4, the number of iterations is 100.

9. The bearing fault diagnosis method according to any one of claims 1 to 8, characterized in that: In step S6, the accuracy is used as the evaluation index of the model.

10. A bearing fault diagnosis system based on a multi-scale wavelet thresholding network, applying the bearing fault diagnosis method according to any one of claims 1 to 9, characterized in that: The bearing fault diagnosis system comprises: An acquisition module is used to collect bearing fault vibration signals at a preset sampling rate to form a bearing fault data set; A data set partitioning module is used to partition the bearing fault data set into a training set and a test set according to a preset ratio; The training module is used to input the one-dimensional bearing fault vibration signal of the training set into the fault diagnosis model for training. The training process of the fault diagnosis model includes: performing frequency analysis on the bearing fault vibration signal through a multi-scale convolutional wavelet decomposition network; wherein the multi-scale convolutional wavelet decomposition network includes three single-channel convolutional wavelet decomposition networks with convolution kernels of different scales; the convolutional wavelet decomposition networks of different channels are used to extract time-frequency features with different characteristics; using the parallel network DRSN-BiGRU of the deep residual shrinkage network and BiGRU to further extract the output features of the multi-scale convolutional wavelet decomposition network; and using the fully connected layer and the dropout layer to classify the output features of the DRSN-BiGRU. The update module is used to calculate the value of the overall loss function of the model and update the model parameters using the Adam optimization algorithm; The iteration module is used to iteratively train the model until the model training converges and a trained model is obtained; The test module is used to input the test set data into the trained model to classify the data, calculate the loss and accuracy, and realize bearing fault diagnosis.

Images