Bearing Fault Diagnosis Method and System Based on WDCNN-LSTM

Abstract

The bearing fault diagnosis method and system based on WDCNN-LSTM disclosed by the present invention include: acquiring the vibration signal of a bearing; obtaining a bearing fault diagnosis result according to the vibration signal and a trained bearing fault diagnosis model; wherein, the bearing fault diagnosis model includes a spatial feature extraction path and a temporal feature extraction path; the temporal feature extraction path includes a convolutional layer and an LSTM connected in sequence; the vibration signal is respectively input into the temporal feature extraction path and the spatial feature extraction path to extract temporal features and spatial features; the temporal features and the spatial features are fused to obtain fused features; the fused features are recognized to obtain a bearing fault diagnosis result. Without preprocessing the vibration noise, it is also possible to accurately judge the bearing fault by identifying the vibration noise.

Description

Technical Field

[0001] The present invention relates to the technical field of bearing fault diagnosis, and particularly to a bearing fault diagnosis method and system based on WDCNN-LSTM. Background Art

[0002] The statements in this part merely provide background technical information related to the present invention and do not necessarily constitute prior art.

[0003] Bearings are one of the most important parts of motors. Bearing faults will directly affect the operation of the motors. Seriously, it may even cause damage to the motors and shorten their service life. Sudden faults are more likely to cause personal injuries and high economic losses. Therefore, it is very important to accurately and real-time diagnose the faults of motor bearings.

[0004] Vibration signal analysis is a commonly used method in motor bearing fault diagnosis. Traditional fault diagnosis methods rely on human professional knowledge to extract features and judge their fault states. The traditional feature extraction methods extract features from vibration signals in the time domain, frequency domain or time-frequency domain by using time domain statistical analysis, short-time Fourier transform, wavelet transform and empirical mode decomposition and other methods. The extracted features rely on rich human experience to judge the specific fault states, which is very time-consuming and laborious. Therefore, a more convenient method for feature extraction is needed.

[0005] In the past few years, methods based on deep learning have developed rapidly. Aiming at the problem of difficult identification of bearing fault types, pattern recognition methods mainly based on deep learning have become a research hotspot.

[0006] Similarly, certain achievements have also been made in bearing fault diagnosis and bearing life prediction using deep learning. Zhang et al. proposed a deep convolutional neural network with a wide first-layer kernel (WDCNN). The first layer of convolution uses a wide convolution kernel, increasing the receptive field and greatly improving the accuracy of fault diagnosis compared with traditional convolutional neural networks. Gong Wenfeng et al. used global average pooling technology to replace the fully connected part of the traditional CNN, effectively solving the problem of excessive parameters in the traditional CNN model and achieving bearing fault diagnosis while reducing parameters. Xiao Xiong et al. converted one-dimensional vibration signals into two-dimensional grayscale images and then used a convolutional neural network for feature extraction, obtaining good results. The above methods have achieved good results in the case of no noise. However, in actual situations, the collected vibration signals are often accompanied by various random noises, indicating that it is impossible to directly use the noisy vibration signals to make a clear diagnosis of bearing faults. Therefore, denoising has become a key issue in vibration signal processing. To solve the noise problem, Zhao et al. proposed a deep residual shrinkage network, which combines the attention mechanism with soft threshold filtering to achieve adaptive threshold filtering and obtained good results in the case of high noise. However, threshold filtering not only filters out noise but also filters out actual vibration signals, resulting in the loss of the original signal. Convolutional autoencoder networks have been widely used in image denoising. To better retain the original signal while filtering, Wan Qiyang et al. used a convolutional autoencoder network for denoising and then used a CNN network for fault diagnosis. However, this method is trained in a supervised manner, using the noisy time-frequency diagram of the original vibration signal as the input of the convolutional autoencoder network and the noise-free time-frequency diagram as the label. In actual situations, it is difficult for us to obtain the label of the noise signal. Ding Yunhao et al. used a one-dimensional multi-scale convolutional autoencoder network to diagnose bearing faults, which can better restore the original data while removing noise. This method first trains the autoencoder network well, adds a softmax classification layer on top of the network, and fine-tunes the network for classification training. However, as the classification training progresses, the pre-trained denoising knowledge is changed to learn classification, weakening the denoising ability of the network.

[0007] Therefore, the inventors believe that the existing bearing fault diagnosis methods cannot directly achieve accurate diagnosis of bearing faults based on noisy vibration signals. Summary of the Invention

[0008] To solve the above problems, the present invention proposes a bearing fault diagnosis method and system based on WDCNN-LSTM, constructing a bearing fault diagnosis model that can directly identify vibration signals without pre-denoising the vibration signals, improving the accuracy and efficiency of bearing fault diagnosis.

[0009] To achieve the above object, the present invention adopts the following technical solutions:

[0010] In the first aspect, a bearing fault diagnosis method based on WDCNN-LSTM is proposed, including:

[0011] Obtain the vibration signal of the bearing;

[0012] According to the vibration signal and the trained bearing fault diagnosis model, obtain the bearing fault diagnosis result;

[0013] Among them, the bearing fault diagnosis model includes a spatial feature extraction path and a temporal feature extraction path; the temporal feature extraction path includes a convolutional layer and an LSTM connected in sequence; the vibration signal is respectively input into the temporal feature extraction path and the spatial feature extraction path to extract temporal features and spatial features; the temporal features and spatial features are fused to obtain fused features; the fused features are identified to obtain the bearing fault diagnosis result.

[0014] In the second aspect, a bearing fault diagnosis system based on WDCNN-LSTM is proposed, including:

[0015] A signal acquisition module for acquiring the vibration signal of the bearing;

[0016] A fault diagnosis module for obtaining the bearing fault diagnosis result according to the vibration signal and the trained bearing fault diagnosis model;

[0017] Among them, the bearing fault diagnosis model includes a spatial feature extraction path and a temporal feature extraction path; the temporal feature extraction path includes a convolutional layer and an LSTM connected in sequence; the vibration signal is respectively input into the temporal feature extraction path and the spatial feature extraction path to extract temporal features and spatial features; the temporal features and spatial features are fused to obtain fused features; the fused features are identified to obtain the bearing fault diagnosis result.

[0018] In the third aspect, an electronic device is proposed, including a memory, a processor, and computer instructions stored on the memory and running on the processor. When the computer instructions are run by the processor, the steps of the bearing fault diagnosis method based on WDCNN-LSTM are completed.

[0019] In the fourth aspect, a computer-readable storage medium is proposed for storing computer instructions. When the computer instructions are executed by the processor, the steps of the bearing fault diagnosis method based on WDCNN-LSTM are completed.

[0020] Compared with the prior art, the beneficial effects of the present invention are:

[0021] 1. The bearing fault diagnosis model proposed by the present invention obtains spatial features and temporal features from vibration signals through a spatial feature extraction path and a temporal feature extraction path, and identifies the fused features of spatial features and temporal features, improving the accuracy of bearing fault diagnosis.

[0022] 2. The temporal feature extraction path of the present invention adds a convolutional layer with a wide convolutional kernel before the LSTM, which helps to suppress high-frequency noise in the original input signal and also helps to learn useful features to be fed into the LSTM block, making the temporal features extracted by the LSTM more comprehensive and effective.

[0023] 3. The present invention increases the depth of the long-term memory chain in the LSTM, discovers interesting temporal correlations in the data, further ensures the comprehensiveness and effectiveness of the extracted temporal features, and improves the accuracy of bearing fault diagnosis.

[0024] 4. The present invention combines WDCNN and LSTM. The proposed model does not require a large amount of preprocessing of data or feature engineering, can directly act on the original data, and does not require manual feature extraction or noise removal.

[0025] Advantages of additional aspects of the present invention will be partly given in the following description, partly will become apparent from the following description, or will be learned through the practice of the present invention. Description of the Drawings

[0026] The specification drawings constituting a part of this application are used to provide a further understanding of this application. The schematic embodiments of this application and their descriptions are used to explain this application and do not constitute an improper limitation of this application.

[0027] Figure 1 It is the overall architecture diagram of the bearing fault diagnosis model disclosed in Embodiment 1;

[0028] Figure 2 It is the structural diagram of WDCNN disclosed in Embodiment 1;

[0029] Figure 3 It is the effect verification diagram of the bearing fault diagnosis model disclosed in Embodiment 1. Detailed Embodiments

[0030] The present invention will be further described below in conjunction with the drawings and embodiments.

[0031] It should be noted that the following detailed description is illustrative and is intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which this application belongs.

[0032] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular forms are also intended to include the plural forms. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they specify the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0033] Embodiment 1

[0034] In this embodiment, a bearing fault diagnosis method based on WDCNN-LSTM is disclosed, including:

[0035] Obtain the vibration signal of the bearing;

[0036] According to the vibration signal and the trained bearing fault diagnosis model, obtain the bearing fault diagnosis result;

[0037] Among them, the bearing fault diagnosis model includes a spatial feature extraction path and a temporal feature extraction path; the temporal feature extraction path includes a convolutional layer and an LSTM connected in sequence; the vibration signal is respectively input into the temporal feature extraction path and the spatial feature extraction path to extract temporal features and spatial features; the temporal features and spatial features are fused to obtain fused features; the fused features are recognized to obtain the bearing fault diagnosis result.

[0038] The bearing fault diagnosis model is described in detail. The bearing fault diagnosis model is as Figure 1 shown, including a spatial feature extraction path, a temporal feature extraction path, a fully connected layer (concatenate), and a Softmax layer.

[0039] Currently, the commonly used two-dimensional convolutional neural network has a stacked 3×3 convolutional kernel in this network structure. This can not only deepen the network depth but also achieve a larger receptive field with fewer parameters, thereby suppressing overfitting. However, for one-dimensional vibration signals, the structure of two layers of 3×1 convolution only obtains a 5×1 receptive field at the cost of 6 weights, turning the above advantages into disadvantages. Therefore, the network structure in the visual field is not applicable to the bearing fault diagnosis field.

[0040] However, in the field of bearing fault diagnosis, using this double-layer 3*1 convolutional structure cannot achieve good results and instead turns the advantages of the network into disadvantages.

[0041] The spatial feature extraction path of this embodiment adopts a convolutional neural network with a wide first-layer kernel (DeepConvolutional Neural Networks with Wide First-layer Kernel, WDCNN), asFigure 2 As shown in Figure 2 , WDCNN includes multiple sequentially connected convolutional blocks (CNN Block). Each convolutional block includes a convolutional layer, a RelU layer, and a pooling layer. The last convolutional block is sequentially connected to a fully connected layer and a Softmax layer. Among them, the first convolutional layer uses a wide convolutional kernel with a width of 64 and a specific size of 64x1. The convolutional kernel widths of the remaining convolutional layers are 3, and the sizes are 3x1.

[0042] The acquired vibration signal of the bearing is input into WDCNN, and a spatial feature vector of the vibration signal is output. The output of WDCNN is input into the Flatten layer for dimensionality reduction processing to output spatial features.

[0043] By using WDCNN, the high-frequency noise of the vibration signal can be suppressed, so that the acquired spatial features are more accurate.

[0044] The time feature extraction path includes a sequentially connected convolutional layer and LSTM. Among them, the convolutional layer uses a wide convolutional kernel with a size of 64x1. By using a 1D convolutional layer path with a wide filter kernel before LSTM, the high-frequency noise in the vibration signal can be suppressed, which helps to learn useful features to be fed into LSTM.

[0045] LSTM is a special type of recurrent neural network (RNN). RNN has a good effect in dealing with time series problems and can reduce the number of training times. It can suppress the high-frequency noise in the input signal in fault diagnosis. RNN is similar to other neural networks and also updates weights by using error backpropagation and gradient descent. Although RNN establishes the connection between hidden layers at different times and achieves the effect of memory, it is only based on the previous moment. Due to the "memory" of RNN being difficult to last and the temporal dependence not being able to extend infinitely, LSTM appears.

[0046] LSTM adds a new time chain long-term memory on the basis of RNN. f1 decides which records to modify according to S t-1 and the input X t as shown in formula (1):

[0047]

[0048] The function f2 decides which records to add according to S t-1 and the input X t as shown in formula (2); among them, the sigmoid function takes values between 0 and 1, and the tanh function takes values between -1 and 1.

[0049]

[0050] Combining the two steps gives

[0051] c t = f1 * c t-1 + f2

[0052] Where W1 represents the weight matrix during deletion, W2 represents the weight matrix during addition, b1 represents the bias during deletion, and b2 represents the bias during addition. is an approximate estimate of W2, is an approximate estimate of b2, S t is the data situation in the LSTM at time t, including the current input x, and also including the state information of S t-1 at the previous time and the information of C t of that time.

[0053] Retain the short-term memory chain S t and the long-term memory chain C t and update each other. This is the advantage of LSTM. LSTM deeply explores the interesting correlations of data in time series. By adding the long-term memory chain C t , it can help capture dynamic time features spanning a large number of time steps. The features learned by these different LSTM paths in time can complement the more local features learned by the convolutional path and improve the final classification result.

[0054] Input the obtained vibration signal of the bearing into the time feature extraction path. First, pass through the convolutional layer to suppress high-frequency noise in the vibration signal, and then obtain time features through LSTM.

[0055] Input the spatial features output by the spatial feature extraction path and the time features output by the time feature extraction path into the fully connected layer for fusion to obtain fused features.

[0056] Input the fused features into the Softmax layer, and obtain the bearing fault diagnosis result by identifying the fused features.

[0057] Use the dataset in the bearing data center to verify the bearing fault diagnosis model proposed in this embodiment. Through experiments, it is found that in the case of low noise, the performance of all algorithms is good. However, as the noise level increases, the performance of the bearing fault diagnosis model constructed in this embodiment is significantly better than other methods. In the case of a signal-to-noise ratio of -4 dB, an accuracy of 98% is even obtained, as Figure 3 shown.

[0058] The bearing fault diagnosis method disclosed in this embodiment can suppress high-frequency noise in vibration signals by adopting WDCNN in the spatial feature extraction path and using a convolutional layer with a wide convolutional kernel before LSTM, thereby achieving accurate identification of bearing faults based on vibration noise without preprocessing the vibration noise.

[0059] Embodiment 2

[0060] In this embodiment, a bearing fault diagnosis system based on WDCNN-LSTM is disclosed, including:

[0061] A signal acquisition module for acquiring the vibration signal of the bearing;

[0062] A fault diagnosis module for obtaining a bearing fault diagnosis result according to the vibration signal and the trained bearing fault diagnosis model;

[0063] Among them, the bearing fault diagnosis model includes a spatial feature extraction path and a temporal feature extraction path; the temporal feature extraction path includes a convolutional layer and an LSTM connected in sequence; the vibration signal is respectively input into the temporal feature extraction path and the spatial feature extraction path to extract temporal features and spatial features; the temporal features and spatial features are fused to obtain fused features; the fused features are identified to obtain a bearing fault diagnosis result.

[0064] Embodiment 3

[0065] In this embodiment, an electronic device is disclosed, including a memory, a processor, and computer instructions stored on the memory and running on the processor. When the computer instructions are run by the processor, the steps of the bearing fault diagnosis method based on WDCNN-LSTM disclosed in Embodiment 1 are completed.

[0066] Embodiment 4

[0067] In this embodiment, a computer-readable storage medium is disclosed for storing computer instructions. When the computer instructions are executed by the processor, the steps of the bearing fault diagnosis method based on WDCNN-LSTM disclosed in Embodiment 1 are completed.

[0068] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: modifications or equivalent replacements can still be made to the specific implementation manners of the present invention, and any modification or equivalent replacement that does not depart from the spirit and scope of the present invention shall be covered by the protection scope of the claims of the present invention.

Claims

1. A bearing fault diagnosis method based on WDCNN-LSTM, characterized in that, Including: Obtain the vibration signal of the bearing; According to the vibration signal and the trained bearing fault diagnosis model, obtain the bearing fault diagnosis result; Among them, the bearing fault diagnosis model includes a spatial feature extraction path and a temporal feature extraction path; the spatial feature extraction path uses WDCNN; the temporal feature extraction path includes a convolutional layer and an LSTM connected in sequence; the vibration signal is input into the temporal feature extraction path and the spatial feature extraction path respectively to extract temporal features and spatial features; the temporal features and spatial features are fused to obtain fused features; the fused features are identified to obtain the bearing fault diagnosis result.

2. The bearing fault diagnosis method based on WDCNN-LSTM according to claim 1, wherein WDCNN includes multiple convolutional blocks connected in sequence, each convolutional block includes a convolutional layer, a RuLe layer and a pooling layer, and the last convolutional block is connected to a fully connected layer and a Softmax layer in sequence, where the first convolutional layer uses a wide convolutional kernel.

3. The bearing fault diagnosis method based on WDCNN-LSTM according to claim 2, wherein The width of the wide convolutional kernel is 64, and the width of the convolutional kernels of the remaining convolutional layers is 3.

4. The bearing fault diagnosis method based on WDCNN-LSTM according to claim 1, characterized in that, The spatial feature extraction path reduces the dimension of the output of WDCNN and then outputs spatial features.

5. The bearing fault diagnosis method based on WDCNN-LSTM according to claim 1, characterized in that, The convolutional layer in the spatial feature extraction path uses a wide convolutional kernel.

6. The bearing fault diagnosis method based on WDCNN-LSTM according to claim 1, characterized in that, The temporal features and spatial features are fused through a fully connected layer, and the fused features are identified through Softmax to obtain the bearing fault diagnosis result.

7. A bearing fault diagnosis system based on WDCNN-LSTM, characterized in that, Including: A signal acquisition module for acquiring the vibration signal of the bearing; A fault diagnosis module for obtaining the bearing fault diagnosis result according to the vibration signal and the trained bearing fault diagnosis model; Among them, the bearing fault diagnosis model includes a spatial feature extraction path and a temporal feature extraction path, the spatial feature extraction path uses WDCNN; the temporal feature extraction path includes a convolutional layer and an LSTM connected in sequence; the vibration signal is input into the temporal feature extraction path and the spatial feature extraction path respectively to extract temporal features and spatial features; the temporal features and spatial features are fused to obtain fused features; the fused features are identified to obtain the bearing fault diagnosis result.

8. An electronic device, characterized in that, Including a memory, a processor, and computer instructions stored on the memory and running on the processor. When the computer instructions are run by the processor, the steps of the bearing fault diagnosis method based on WDCNN-LSTM according to any one of claims 1-6 are completed.

9. A computer-readable storage medium, characterized in that, For storing computer instructions, when the computer instructions are executed by the processor, the steps of the bearing fault diagnosis method based on WDCNN-LSTM according to any one of claims 1-6 are completed.

Images