A bearing fault diagnosis method based on fusion of electric signals and vibration signals

Abstract

The application provides a bearing fault diagnosis method based on fusion of electric signals and vibration signals, comprising: collecting electric signal and vibration signal data generated by bearings under different operating states, and extracting MFCC feature matrices of the electric signals and the vibration signals through MFCC respectively; concatenating and splicing the MFCC feature matrices of the electric signals and the vibration signals under different operating states respectively, then processing the concatenated and combined feature matrices through normalization, generating a gray image according to the concatenated and combined feature matrices, taking the gray images under different labels as a training set, training a deep convolutional generative adversarial network model to reach Nash equilibrium, and obtaining the trained deep convolutional generative adversarial network model; taking the discriminator in the trained deep convolutional generative adversarial network model as a network convolutional neural network (CNN) model for fault diagnosis; extracting MFCC feature matrices from real-time acquired electric signals and vibration signal data to generate a gray image, inputting the gray image into the CNN model for fault diagnosis, and obtaining a diagnosis result.

Description

Technical Field

[0001] This invention relates to the technical field of fault diagnosis, and more particularly to a bearing fault diagnosis method based on the fusion of electrical signals and vibration signals. Background Technology

[0002] In recent years, with the rise of the concept of intelligent ships and the rapid development of the big data era, ship intelligence has become a global trend in shipping. In ship operation, the shafting system plays a crucial role in the normal operation of the ship, and bearings best reflect the state of the shafting system, making bearing fault diagnosis particularly important. Traditional diagnostic methods rely on manual feature extraction, requiring extensive signal processing and diagnostic expertise, resulting in extremely low accuracy for non-professionals. To address this issue, a convolutional neural network (CNN) fault diagnosis method is proposed. This method can automatically extract features from data, reducing reliance on manual intervention and achieving higher fault diagnosis accuracy. However, it places certain demands on the network, requiring balanced diagnostic category sample data. In practical fault diagnosis, bearings operate normally most of the time, with fault conditions being rare. To better train the network, balanced sample data is needed, with the number of training samples based on the class with the fewest samples. Limited training samples will lead to poor network training, and using the same type of signal as diagnostic training sample data will also result in unsatisfactory fault diagnosis results.

[0003] To address the above problems, this invention proposes a bearing fault diagnosis method based on the fusion of electrical and vibration signals. The method collects bearing electrical and vibration signal data. Both electrical and vibration signals reflect the operating state of the shaft system, and different types of signals contain different characteristic information. The fusion of these two signals fully utilizes their complementarity to enhance fault diagnosis features. Furthermore, compared with traditional neural network methods, this invention employs a convolutional generative adversarial network (GAN). The generator (G) generates artificial data similar to the original data to augment the samples, solving the problems of imbalanced diagnostic category sample datasets and limited training samples. Therefore, this method can perform better fault diagnosis. Summary of the Invention

[0004] In view of the technical problems mentioned in the background, this invention provides a bearing fault diagnosis method based on the fusion of electrical signals and vibration signals. The technical means adopted by this invention are as follows:

[0005] A bearing fault diagnosis method based on the fusion of electrical and vibration signals includes the following steps:

[0006] Step S1: Collect electrical and vibration signal data generated by the bearing under different operating conditions, and extract the MFCC feature matrix of the electrical and vibration signals respectively through MFCC;

[0007] Step S2: After concatenating and splicing the MFCC feature matrices of the electrical signal and vibration signal under different operating conditions, the concatenated feature matrix is ​​normalized and a grayscale image is generated based on the concatenated feature matrix.

[0008] Step S3: Build a deep convolutional generative adversarial network model using Python within the Torch framework. Use grayscale images under different label classifications as the training set to train the deep convolutional generative adversarial network model until Nash equilibrium is reached, and obtain the trained deep convolutional generative adversarial network model.

[0009] Step S4: Use the discriminator in the trained deep convolutional generative adversarial network model as the convolutional neural network (CNN) model for fault diagnosis;

[0010] Step S5: Extract the MFCC feature matrix from the real-time acquired electrical signal and vibration signal data to generate a grayscale image, input the grayscale image into the CNN model for fault diagnosis, and obtain the diagnosis result.

[0011] Compared with the prior art, the present invention has the following advantages:

[0012] This invention discloses a bearing fault diagnosis method based on the fusion of electrical and vibration signals. It collects both electrical and vibration signal data, leveraging the fact that both signals reflect the operating state of the shaft system. Furthermore, different signal types contain different characteristic information, and fusing these signals improves diagnostic accuracy. Simultaneously, a deep convolutional generative adversarial network model is employed to address the problem of imbalanced datasets, expanding the training set sample data and enabling better model training, thus improving diagnostic results. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 This is a schematic diagram of the overall process of the present invention.

[0015] Figure 2 This is a schematic diagram of the adversarial network of the present invention. Detailed Implementation

[0016] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0017] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0018] like Figure 1 As shown, this invention provides a bearing fault diagnosis method based on the fusion of electrical signals and vibration signals, comprising the following steps:

[0019] Step S1: Collect electrical and vibration signal data generated by the bearing under different operating conditions, and extract the MFCC feature matrix of the electrical and vibration signals respectively through MFCC; further, as a preferred embodiment, the different operating conditions in step S1 include: normal bearing operation state, misalignment state, imbalance state and combined fault state.

[0020] Step S2: After concatenating and splicing the MFCC feature matrices of the electrical signal and vibration signal under different operating conditions, the concatenated feature matrix is ​​normalized and a grayscale image is generated based on the concatenated feature matrix; it can be understood that the concatenation in this application is to splice two M*N matrices into an M*2N matrix.

[0021] Step S3: Build a deep convolutional generative adversarial network (DGAN) model using Python within the Torch framework. Use grayscale images from different label categories as the training set to train the DGAN model until Nash equilibrium is achieved, thus obtaining the trained DGAN model. For each state label category (normal, misaligned, imbalanced, compound fault), there is a certain number of grayscale images. These grayscale images are generated in step S2 and placed in the corresponding category file.

[0022] Step S4: Use the discriminator in the trained deep convolutional generative adversarial network model as the convolutional neural network (CNN) model for fault diagnosis;

[0023] Step S5: Extract the MFCC feature matrix from the real-time acquired electrical signal and vibration signal data to generate a grayscale image, input the grayscale image into the CNN model for fault diagnosis, and obtain the diagnosis result.

[0024] In step S2, extracting the MFCC feature matrix includes the following steps: Step 21: Signal pre-emphasis; First, amplify the high-frequency noise in the signal, then suppress the random noise in the signal through a pre-emphasis filter:

[0025] H(Z) = 1 - μZ -1 ;

[0026] Where Z represents the input signal, and μ takes values ​​in the range of [0.9-1].

[0027] Step 22: Signal Windowing; The signal is framed using a window function to obtain a stable frame signal. The commonly used window function is the Hamming window, and its expression is:

[0028]

[0029] Where α is 0.46, and K represents the number of data points in each frame;

[0030] Step 23: Discrete Fourier Transform of Signal Segment; Perform a Discrete Fourier Transform on the signal segment τ(n) to extract the spectral information T(k) from the signal segment, i.e.:

[0031]

[0032] Where N represents the number of samples in each frame;

[0033] Step 24: Power spectrum calculation; obtain the discrete power spectrum Pm through T(k), the process is as follows:

[0034] P m =T(k)·^T(k);

[0035] Step 25: Perform Mel-triangle filtering; describe the relationship between the Mel frequency and the signal frequency according to the Mel frequency scale. The expression for the relationship is:

[0036]

[0037] Next, the triangular filter HM(k) is used to filter the information of each frequency band, as follows:

[0038] 0,k <f(t-1);

[0039]

[0040]

[0041] Where t = 1, 2, ..., 24; f(t) is the center frequency, k = 0, 1, ..., N / 2–1; furthermore, HM(k) satisfies

[0042] Step 26: Calculate the logarithmic spectrum; perform a logarithmic operation on the obtained power spectrum to obtain the logarithmic spectrum, i.e.:

[0043]

[0044] Where S(m) represents the logarithmic spectrum, Pm represents the discrete power spectrum, and M represents the number of triangular filters;

[0045] Step 27: Discrete Cosine Transform; The discrete cosine transform is used to calculate the spectral components of different frequency bands, making the dimension vectors of each frequency band independent of each other, thus obtaining the information of each frequency band of the signal, i.e.:

[0046]

[0047] Assuming the MFCC feature of the data sample is c(n), the calculation process of its first-order and second-order differences is as follows:

[0048]

[0049]

[0050] Design a combination operation to combine the obtained c(n), d1(n), and d2(n). The operation process is as follows:

[0051] M data = [c(n)d1(n)d2(n)];

[0052] Here, [·] represents a concatenation operation, which connects different inputs end to end. Here, Mdata is defined as the constructed MFCC feature matrix.

[0053] In step S2, the signal data is denoised and the MFCC feature matrix is ​​extracted using MFCC. The MFCC feature matrices of the electrical signal and the vibration signal are cascaded, combined, and normalized to generate a grayscale image. The feature monitoring and fusion of different types of signals enhances the fault characteristics and improves the accuracy of fault diagnosis.

[0054] In step 3, a deep convolutional generative adversarial network model is built. Grayscale images under different label classifications are input into the deep convolutional generative adversarial network model. The model includes a generator G and a discriminator D. The generator G generates high-quality and realistic image data and the discriminator D classifies the dataset.

[0055]

[0056] Where G and D represent the generator and discriminator, respectively; z represents the input one-dimensional random noise vector, x is the input real data; and G(Z) represents the generated virtual sample.

[0057] In a preferred embodiment, the proposed deep convolutional generative adversarial network model in step S3 has the following generator network structure: transposed convolution, normalization and activation function, transposed convolution, normalization and activation function, transposed convolution, and activation function. In step S3, the LeakyReLU activation function is used, and the generator output layer uses Tanh, where all layers except the output layer use ReLU.

[0058]

[0059] Leakyrelu(x) = max(0.01x, x)

[0060]

[0061] Example 1

[0062] A bearing fault diagnosis method based on the fusion of electrical and vibration signals is presented in this embodiment, using a shaft test bench as an example. The detailed steps are as follows:

[0063] Step 1: Collect electrical and vibration signals of the shaft system under different operating conditions from the electrometer and accelerometer installed on the bearing housing. Each sample consists of 1024 sampling points, and there are 600 samples for each operating condition.

[0064] Step 2: Extract the MFCC feature matrix from the signal, normalize it, and generate a grayscale image. Classify the sample images of different bearing operating states: normal (0), misalignment (1), imbalance (2), and compound fault (3).

[0065] Step 3: Divide the bearing state images with different labels into two parts: training images (training deep convolutional generative adversarial network) and test images (testing deep convolutional generative adversarial network) in an 8:2 ratio.

[0066] Step 4: Set the parameters of the deep convolutional generative adversarial network model, input the training image into the deep convolutional generative adversarial network for adversarial training. The generator's role is to generate fake data, and the discriminator's role is to distinguish between real and fake data. During the adversarial training process, the generator (G) and the discriminator (D) are in a competitive process.

[0067] Step 5: During the training of G and D, iteratively optimize the generator (G) and discriminator (D) respectively. First, fix G and optimize the discriminator to maximize its accuracy. Then, fix D and optimize the generator to minimize its discrimination accuracy. Repeat this process until the generator network and discriminator network reach Nash equilibrium, at which point training stops.

[0068] Step 6: After the deep convolutional generative adversarial network reaches Nash equilibrium during training, fix and extract the parameters of the discriminator to serve as the CNN model for fault diagnosis. Input the grayscale image of the test sample into the CNN for fault diagnosis.

[0069] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments. In the above embodiments of the present invention, the descriptions of each embodiment have their own emphasis; parts not described in detail in a certain embodiment can be referred to in the relevant descriptions of other embodiments. It should be understood that the disclosed technical content in the several embodiments provided in this application can be implemented in other ways.

[0070] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A bearing fault diagnosis method based on the fusion of electrical signals and vibration signals, characterized in that, Includes the following steps: S1: Collect electrical and vibration signal data generated by the bearing under different operating conditions, and extract the MFCC feature matrix of the electrical and vibration signals respectively through MFCC; S2: After concatenating and splicing the MFCC feature matrices of the electrical signal and vibration signal under different operating conditions, the concatenated feature matrix is ​​normalized and a grayscale image is generated based on the concatenated feature matrix. S3: Build a deep convolutional generative adversarial network model using Python within the Torch framework. Use grayscale images under different label classifications as the training set to train the deep convolutional generative adversarial network model until Nash equilibrium is reached, and obtain a trained deep convolutional generative adversarial network model. S4: Use the discriminator in the trained deep convolutional generative adversarial network model as a convolutional neural network (CNN) model for fault diagnosis; S5: Extract the MFCC feature matrix from the real-time acquired electrical and vibration signal data to generate a grayscale image, input the grayscale image into the CNN model for fault diagnosis, and obtain the diagnosis result.

2. The bearing fault diagnosis method based on the fusion of electrical signals and vibration signals according to claim 1, characterized in that: In step S1, the different operating states include: normal bearing operation state, misalignment state, imbalance state, and combined fault state.

3. The bearing fault diagnosis method based on the fusion of electrical signals and vibration signals according to claim 1, characterized in that: In S2, the extraction of the MFCC feature matrix includes the following steps: Step 21: Signal pre-emphasis; First, the high-frequency noise in the signal is amplified, and then the random noise in the signal is suppressed by the pre-emphasis filter: H(Z)=1-μZ -1 ; Where Z represents the input signal, and μ takes values ​​in the range of [0.9-1]. Step 22: Signal Windowing; The signal is framed using a window function to obtain a stable frame signal. The commonly used window function is the Hamming window, and its expression is: Where α is 0.46, and K represents the number of data points in each frame; Step 23: Discrete Fourier Transform of Signal Segment; Perform a Discrete Fourier Transform on the signal segment τ(n) to extract the spectral information T(k) from the signal segment, i.e.: Where N represents the number of samples in each frame; Step 24: Power spectrum calculation; obtain the discrete power spectrum Pm through T(k), the process is as follows: P m =T(k)· ∧ T(k); Step 25: Perform Mel-triangle filtering; describe the relationship between the Mel frequency and the signal frequency according to the Mel frequency scale. The expression for the relationship is: Next, the triangular filter HM(k) is used to filter the information of each frequency band, as follows: 0,k <f(t-1); Where t = 1, 2, ..., 24; f(t) is the center frequency, k = 0, 1, ..., N / 2–1; furthermore, HM(k) satisfies Step 26: Calculate the logarithmic spectrum; perform a logarithmic operation on the obtained power spectrum to obtain the logarithmic spectrum, i.e.: Where S(m) represents the logarithmic spectrum, Pm represents the discrete power spectrum, and M represents the number of triangular filters; Step 27: Discrete Cosine Transform; The discrete cosine transform is used to calculate the spectral components of different frequency bands, making the dimension vectors of each frequency band independent of each other, thus obtaining the information of each frequency band of the signal, i.e.: Assuming the MFCC feature of the data sample is c(n), the calculation process of its first-order and second-order differences is as follows: Design a combination operation to combine the obtained c(n), d1(n), and d2(n). The operation process is as follows: M data =[c(n)d1(n)d2(n)]; Here, [·] represents a concatenation operation, which connects different inputs end to end. Here, Mdata is defined as the constructed MFCC feature matrix.

4. The bearing fault diagnosis method based on the fusion of electrical signals and vibration signals according to claim 1, characterized in that: In step 3, a deep convolutional generative adversarial network model is built. Grayscale images under different label classifications are input into the deep convolutional generative adversarial network model. The model includes a generator G and a discriminator D. The generator G generates high-quality and realistic image data and the discriminator D classifies the dataset. Where G and D represent the generator and discriminator, respectively; z represents the input one-dimensional random noise vector, x is the input real data; and G(Z) represents the generated virtual sample.

5. The bearing fault diagnosis method based on the fusion of electrical signals and vibration signals according to claim 1, characterized in that: In step S3, the proposed deep convolutional generative adversarial network model has a generator network structure of: transposed convolution, normalization and activation function, transposed convolution, normalization and activation function, transposed convolution, and activation function.

6. The bearing fault diagnosis method based on the fusion of electrical signals and vibration signals according to claim 1, characterized in that: In step S3, the Leakyrelu activation function is used, and the generator output layer uses Tanh, where all layers except the output layer use Relu; Leakyrelu(x) = max(0.01x, x)

Images