Bearing fault diagnosis method and device, computer equipment and storage medium

Abstract

The invention discloses a bearing fault diagnosis method and device, computer equipment and a storage medium, and belongs to the technical field of mechanical fault diagnosis. An improved empirical mode decomposition algorithm is adopted to decompose the bearing vibration signal data set to generate a time domain data set, and the time domain data set is processed to generate a frequency domain data set; improving a CNN-BiLSTM network model, wherein the model comprises a one-dimensional convolutional neural network, a bidirectional long short-term memory network, a time domain-frequency domain cross attention mechanism, a normalization layer and a full connection layer; the one-dimensional convolutional neural network performs feature extraction on the time domain data set to obtain time domain features, the bidirectional long-short-term memory network performs feature extraction on the frequency domain data set to obtain frequency domain features, and the time domain-frequency domain cross attention mechanism performs feature interaction on the time domain features and the frequency domain features to obtain fusion features; and inputting the fusion features into a normalization layer and a full connection layer and then outputting a fault category. The problem of motor bearing fault identification and diagnosis in a complex industrial environment can be solved.

Description

Technical Field

[0001] The present invention relates to the technical field of mechanical fault diagnosis, and more specifically, to a bearing fault diagnosis method, device, computer device, and storage medium. Background Art

[0002] In modern society, motors are widely used in multiple fields such as power systems, industrial production, and intelligent manufacturing. As an important component in motors, bearings affect the performance, efficiency, and lifespan of motors. Bearing faults can lead to further damage to the motor, thereby causing the shutdown of industrial equipment and the interruption of power supply, affecting people's production order and causing great harm to social life. However, by timely identifying faults through bearing fault diagnosis technology and taking preventive maintenance measures, economic losses can be greatly reduced. Therefore, achieving effective diagnosis and testing of motor bearing faults has become one of the important topics for high-quality development in the power and industrial fields.

[0003] Since motor bearings usually operate in a relatively complex industrial environment, the collected bearing vibration signals often contain a lot of non-linear and non-stationary interference noises. Traditional fault diagnosis methods mostly use stationary linear signals as references and cannot simultaneously take into account the time-domain and frequency-domain signals of bearing signals in a strong interference environment, thus showing certain limitations. How to effectively process the environmental noise in the vibration signal and accurately extract its signal features is a key step in the bearing fault diagnosis method.

[0004] Empirical Mode Decomposition (EMD) has significant advantages in processing bearing fault signals and can effectively process non-linear and non-stationary bearing vibration signals to extract useful fault features. However, the EMD method is prone to mode mixing, that is, an intrinsic mode component may contain multiple components of different scales, or components of the same scale may be distributed in different intrinsic mode components.

[0005] In recent years, with the development and progress of information technology and artificial intelligence technology, fault diagnosis methods based on deep learning have received increasing attention. Convolutional Neural Network (CNN) is a typical model in deep learning. It performs convolution and pooling operations on data by constructing filters with different features to fully extract the local and global features of the data. Research shows that the CNN model is also effective for one-dimensional time series data. Long Short-Term Memory Network (LSTM) is an enhanced recurrent neural network model, which has the advantages of capturing long-term dependencies, adapting to non-stationary signals, comprehensively using context information, and processing variable-length sequences in fault diagnosis tasks. Summary of the Invention

[0006] 1. Technical Problems to be Solved

[0007] Aiming at the problem of low accuracy in motor bearing fault diagnosis in the existing technology, the present invention provides a bearing fault diagnosis method, device, computer device and storage medium, which effectively solve the problem of motor bearing fault identification and diagnosis in complex industrial environments based on multi-modal information fusion and an improved CNN-BiLSTM model.

[0008] 2. Technical solutions

[0009] The object of the present invention is achieved through the following technical solutions.

[0010] A bearing fault diagnosis method includes the following steps:

[0011] Collect a bearing vibration signal dataset;

[0012] Use an improved empirical mode decomposition algorithm to decompose the bearing vibration signal dataset to generate a time-domain dataset, and perform a fast Fourier transform operation on the time-domain dataset to generate a frequency-domain dataset;

[0013] Construct an improved CNN-BiLSTM network model, where the CNN-BiLSTM network model includes a one-dimensional convolutional neural network, a bidirectional long short-term memory network, a time-domain-frequency-domain cross-attention mechanism, a normalization layer and a fully connected layer;

[0014] The one-dimensional convolutional neural network extracts time-domain features from the time-domain dataset, the bidirectional long short-term memory network extracts frequency-domain features from the frequency-domain dataset, and the time-domain-frequency-domain cross-attention mechanism performs feature interaction on the time-domain features and frequency-domain features to obtain fused features;

[0015] Input the fused features into the normalization layer and the fully connected layer, and the normalization layer and the fully connected layer output the fault category.

[0016] As a further improvement of the present invention, using an improved empirical mode decomposition algorithm to decompose the bearing vibration signal dataset to generate a time-domain dataset, the specific steps include:

[0017] Add noise during the process of decomposing the bearing vibration signal dataset by the empirical mode decomposition algorithm to obtain an increased-noise bearing vibration signal dataset;

[0018] Calculate the average local mean of the increased-noise bearing vibration signal dataset to generate a time-domain dataset.

[0019] As a further improvement of the present invention, perform a fast Fourier transform operation on the time-domain dataset, output the corresponding amplitude spectrum, and then calculate the actual frequency value corresponding to each frequency component according to the length and sampling frequency set by the fast Fourier transform, and retain the positive frequency part in the actual frequency value and the amplitude spectrum to generate a frequency-domain dataset.

[0020] As a further improvement of the present invention, the one-dimensional convolutional neural network is composed of four convolutional modules connected in series in sequence. The convolutional module includes two 3×1 convolutional layers, one 2×1 max pooling layer, and a Relu activation function.

[0021] As a further improvement of the present invention, the bidirectional long short-term memory network includes a forward LSTM layer and a backward LSTM layer. The basic unit of the LSTM layer includes an input gate, a forget gate, an output gate, and a cell state.

[0022] As a further improvement of the present invention, after the time-domain features and frequency-domain features of the time-domain - frequency-domain cross-attention mechanism are linearly transformed respectively, the frequency-domain features generate query vectors, and the time-domain features generate key vectors and value vectors.

[0023] As a further improvement of the present invention, the attention weights are calculated through the query vectors of the frequency-domain features and the key vectors of the time-domain features, and then the value vectors of the time-domain features are weighted and summed with the attention weights to obtain the fused features.

[0024] A bearing fault diagnosis device includes:

[0025] A data acquisition module that acquires a bearing vibration signal dataset;

[0026] A data processing module that decomposes the bearing vibration signal dataset by using an improved empirical mode decomposition algorithm to generate a time-domain dataset, and performs a fast Fourier transform operation on the time-domain dataset to generate a frequency-domain dataset;

[0027] A model improvement module that constructs an improved CNN-BiLSTM network model. The CNN-BiLSTM network model includes a one-dimensional convolutional neural network, a bidirectional long short-term memory network, a time-domain - frequency-domain cross-attention mechanism, a normalization layer, and a fully connected layer;

[0028] A model processing module. The one-dimensional convolutional neural network extracts time-domain features from the time-domain dataset, the bidirectional long short-term memory network extracts frequency-domain features from the frequency-domain dataset, and the time-domain - frequency-domain cross-attention mechanism performs feature interaction on the time-domain features and the frequency-domain features to obtain fused features;

[0029] A fault diagnosis module that inputs the fused features into the normalization layer and the fully connected layer, and the normalization layer and the fully connected layer output the fault category.

[0030] A computer device includes a memory and a processor. A computer program that can run on the processor is stored on the memory. When the processor executes the computer program, the method described in any one of the above is implemented.

[0031] A computer-readable storage medium has a computer program stored thereon, and when the computer program is run by a processor, it executes the method described in any one of the above.

[0032] 3. Beneficial effects

[0033] Compared with the prior art, the advantages of the present invention are as follows:

[0034] (1) For a bearing fault diagnosis method, device, computer device and storage medium of the present invention, by using an improved empirical mode decomposition algorithm for the bearing vibration signal dataset collected on a vibration sensor, the complex original bearing vibration signal is converted into multiple time-domain signals, and then the fast Fourier transform is performed on the decomposed time-domain signals to convert the time-domain signals into frequency-domain signals. After being processed, the original complex signal is decomposed and transformed into frequency characteristics, which can more meticulously analyze the frequency characteristics of the bearing vibration signal, help identify key frequency information in fault diagnosis, and at the same time facilitate the extraction of features from multiple angles by the subsequent network model, contributing to the construction of a more comprehensive fault diagnosis model.

[0035] (2) For a bearing fault diagnosis method, device, computer device and storage medium of the present invention, by improving the CNN-BiLSTM model, it can first use a one-dimensional convolutional neural network to extract local features, and then capture the global dependencies of these local features evolving over time through a bidirectional long short-term memory network, enabling the CNN-BiLSTM model to capture both subtle local changes and comprehensively consider the long-term dependence features of the sequence, achieving accurate identification of various fault states of the motor bearing, improving the reliability and accuracy of fault diagnosis, and optimizing the maintenance and operation management of the equipment. Description of the drawings

[0036] Figure 1 It is a time-series signal diagram of the vibration states of 10 types of motor bearings collected in the embodiment of the present invention;

[0037] Figure 2 It is a time-domain signal diagram of the fault signal of the bearing inner ring after being decomposed by the improved empirical mode decomposition algorithm in the embodiment of the present invention;

[0038] Figure 3 It is a frequency-domain signal diagram after the fast Fourier transform operation in the embodiment of the present invention;

[0039] Figure 4 It is a schematic structural diagram of the improved CNN-BiLSTM network model in the embodiment of the present invention;

[0040] Figure 5 It is a schematic structural diagram of the one-dimensional convolutional neural network in the embodiment of the present invention;

[0041] Figure 6Schematic diagram of the bidirectional long short-term memory network structure according to an embodiment of the present invention;

[0042] Figure 7 Schematic diagram of the time-domain to frequency-domain cross-attention mechanism structure according to an embodiment of the present invention. Detailed implementation manners

[0043] The present invention will be described in detail below in conjunction with the specification drawings and specific embodiments.

[0044] Embodiment

[0045] A bearing fault diagnosis method provided in this embodiment includes the following steps: collecting a bearing vibration signal dataset; using an improved empirical mode decomposition algorithm to decompose the bearing vibration signal dataset to generate a time-domain dataset, performing a fast Fourier transform operation on the time-domain dataset to generate a frequency-domain dataset; constructing an improved CNN-BiLSTM network model, where the CNN-BiLSTM network model includes a one-dimensional convolutional neural network, a bidirectional long short-term memory network, a time-domain to frequency-domain cross-attention mechanism, a normalization layer, and a fully connected layer; the one-dimensional convolutional neural network extracts time-domain features from the time-domain dataset, the bidirectional long short-term memory network extracts frequency-domain features from the frequency-domain dataset, and the time-domain to frequency-domain cross-attention mechanism performs feature interaction on the time-domain features and the frequency-domain features to obtain fused features; inputting the fused features into the normalization layer and the fully connected layer, and the normalization layer and the fully connected layer output the fault category.

[0046] Specifically in this embodiment, first, original bearing vibration signal data samples are collected on a bearing fault simulation test bench, that is, one-dimensional time-series vibration signals of a motor bearing, and then an original dataset is generated. As Figure 1 shown, it is the original bearing vibration signal data sample collected by the vibration sensor. This embodiment includes 10 bearing state types, namely 0.007-inch inner ring fault, 0.007-inch ball fault, 0.007-inch outer ring fault, 0.014-inch inner ring fault, 0.014-inch ball fault, 0.014-inch outer ring fault, 0.021-inch inner ring fault, 0.021-inch ball fault, 0.021-inch outer ring fault of the motor under different loads, and the bearing signal in the normal state.

[0047] Furthermore, using an improved empirical mode decomposition algorithm to decompose the bearing vibration signal dataset to generate a time-domain dataset includes adding noise during the process of decomposing the bearing vibration signal dataset by the empirical mode decomposition algorithm to obtain an increased-noise bearing vibration signal dataset, and calculating the average local mean of the increased-noise bearing vibration signal dataset to generate a time-domain dataset.

[0048] Specifically, special noise is added during the decomposition of the bearing vibration signal dataset by the existing empirical mode decomposition algorithm. That is, when the k-th IMF component is obtained by the empirical mode decomposition algorithm, noise is added. Gaussian white noise w with a mean of 0 and a unit covariance is added to the bearing vibration signal data x (i) (i = 1, 2, …, I), and we get:

[0049] x (i) = x + β0E1(w (i) )

[0050] where x represents the bearing vibration signal, and x (i) represents the bearing vibration signal dataset after adding noise. β0 represents the initial noise intensity coefficient, and w (i) represents the Gaussian white noise sequence added for the i-th time. E1(·) represents the standardization operation on the bearing vibration signal, and E n (w (i) ) represents the noise. In this embodiment, according to the added noise and the noise residue, the required signal-to-noise ratio is obtained and the constant is set to β k = ε k std(r k ). When k = 1, β0 = ε0std(x) / std(E1(w (i) )). Where k represents a natural number, ε k represents the required signal-to-noise ratio, which is used to quantify the relative intensity of the signal and the noise. r k represents the noise residue, and std represents the standard deviation of the noise residue r k . Thus, by introducing noise, the mode mixing effect of the traditional empirical mode decomposition algorithm can be effectively suppressed, and the recognizability of fault features can be improved.

[0051] After obtaining the bearing vibration signal dataset x (i) with added noise, the existing empirical mode decomposition algorithm is used to iteratively calculate its local mean for i times for the bearing vibration signal dataset x (i) . Thus, the time-domain signal component obtained is x - r1 = x - [M(x (i) )]. Through r k = r k-1 + β k-1 E k (w (i) ) to calculate the average local mean, and further the time-domain dataset obtained is r k-1 - r k = [M(r k-2 + β k-2 E k-1 (w (i)))]-[M(r k-1 +β k-1 E k (w (i) ))]. As Figure 2 shown, it is the time-domain signal diagram of the bearing inner ring fault signal after decomposition by the improved empirical mode decomposition algorithm. In this embodiment, by suppressing the random error of single decomposition, it is ensured that the time-domain signal components can retain the true impact characteristics, and the residual signal can be dynamically corrected through the recursive formula to suppress invalid decomposition.

[0052] In this embodiment, a fast Fourier transform operation is performed on the time-domain data set to generate a frequency-domain data set. Specifically, a fast Fourier transform operation is performed on the time-domain data set to output the corresponding amplitude spectrum, that is, the amplitude sizes corresponding to each frequency component of the bearing vibration signal. Then, according to the length and sampling frequency set by the fast Fourier transform, the actual frequency value corresponding to each frequency component is calculated, and the positive frequency part in the actual frequency value and the amplitude spectrum is retained to generate a frequency-domain data set. As Figure 3 shown, it is the frequency-domain signal diagram after the fast Fourier transform operation.

[0053] Thus, in this embodiment, the improved empirical mode decomposition algorithm decomposes the complex non-stationary bearing signal into multiple intrinsic mode components, thereby capturing some detailed features that may be related to faults. On the one hand, the improved empirical mode decomposition algorithm introduces local mean optimization, and the adaptive sliding window can avoid the low-frequency trend of high-frequency oscillation interference. On the other hand, the improved empirical mode decomposition algorithm separates the similar frequency components through noise combinations of different scales. For example, it distinguishes the fault characteristic frequency of the bearing inner ring from the outer ring interference. At the same time, in the data preprocessing stage, the fast Fourier transform is also used to convert the time-domain information of the intrinsic mode components into frequency-domain information, providing various types of fault characteristics for the subsequent diagnostic model.

[0054] Furthermore, a network model based on the improved CNN-BiLSTM is constructed. In this embodiment, as Figure 4 shown, the CNN-BiLSTM network model includes a one-dimensional convolutional neural network (1DCNN), a bidirectional long short-term memory network (BiLSTM), a time-domain-frequency-domain cross-attention mechanism, a normalization layer, and a fully connected layer.

[0055] As Figure 5 shown, the one-dimensional convolutional neural network is composed of four convolutional modules connected in series in sequence. In this embodiment, the convolutional module includes two 3×1 convolutional layers, a 2×1 max pooling layer, and a Relu activation function. In this embodiment, the number of channels of the convolutional layer and the pooling layer in the four convolutional modules are 32, 64, 128, and 256 respectively. The one-dimensional convolutional neural network extracts features from the time-domain data set to obtain time-domain features.

[0056] As shown Figure 6 in the figure, the bidirectional long short-term memory network includes a forward LSTM layer and a backward LSTM layer. The length of the input data for each memory network is 128. When solving, the forward LSTM layer state is calculated separately to generate a new forward state and the backward LSTM layer state to generate a new backward state The forward state and the backward state are combined to obtain the result Y t , and the calculation formula is:

[0057]

[0058] where x t represents the input vector at time step t, tanh represents the hyperbolic tangent activation function, which is used to capture complex patterns in sequence data, and represent the weight matrices connecting each layer to the previous state, and b y represents the bias term.

[0059] In this embodiment, the basic unit of the LSTM layer includes an input gate, a forget gate, an output gate, and a cell state. The bidirectional long short-term memory network extracts frequency domain features from the frequency domain data set. The input gate determines how much information at the current moment is written into the cell state, the forget gate determines how much information from the previous moment is discarded, and the output gate determines how much information is output to the next moment.

[0060] As shown Figure 7 in the figure, after the time-domain - frequency-domain cross-attention mechanism linearly transforms the time-domain features and the frequency-domain features respectively, the frequency-domain features generate a query vector Q, and the time-domain features generate a key vector K and a value vector V. The time-domain - frequency-domain cross-attention mechanism performs feature interaction on the time-domain features and the frequency-domain features to obtain fused features. Specifically, the attention weights are calculated through the query vector Q of the frequency-domain features and the key vector K of the time-domain features, and then the value vector V of the time-domain features is weighted and summed with the attention weights to obtain the fused features. Finally, the fused features are input into a normalization layer and a fully connected layer, and the normalization layer and the fully connected layer output the fault category.

[0061] Therefore, in this embodiment, the CNN-BiLSTM network model is improved by adopting the collaborative effect of the dual channels in the time domain and the frequency domain, and introducing the time-domain to frequency-domain cross-attention mechanism. In the improved CNN-BiLSTM network model, the time-domain channel can capture the transient impact characteristics of the bearing vibration signal and retain the temporal correlation of the signal; the frequency-domain channel can effectively extract the resonance frequency band characteristics of the bearing vibration signal and effectively suppress noise interference; the time-domain to frequency-domain cross-attention mechanism generates attention weights by calculating the similarity matrix between the time-domain and frequency-domain feature maps, thereby mining the correlation and sharing information of the two different modalities of the time domain and the frequency domain, and enhancing the diagnostic ability of the CNN-BiLSTM network model. Therefore, the improved CNN-BiLSTM network model can exhibit better adaptability and generalization ability under different working conditions and various types of complex environments through multi-level feature extraction and sequence modeling.

[0062] As shown in Table 1, it is a comparison of the accuracy of a bearing fault diagnosis method provided in this embodiment and other traditional bearing fault diagnosis methods tested on the Case Western Reserve University bearing dataset.

[0063] Table 1

[0064] model accuracy LSTM 81.51 GRU 87.76 Transformer 89.96 1DCNN 94.97 CNN-BiLSTM 95.54 Improved CNN-BiLSTM 97.82

[0065] As can be seen from Table 1, a bearing fault diagnosis method provided in this embodiment can decompose complex non-stationary bearing signals into multiple intrinsic mode components by using the improved empirical mode decomposition algorithm and fast Fourier transform, and convert the time-domain information into frequency-domain information, so that it can not only capture the detailed features that may be related to faults, but also provide various types of fault features for the subsequent diagnostic model. In addition, by introducing the time-domain to frequency-domain cross-attention mechanism to mine the correlation and share information of the feature information of the two different modalities of the time domain and the frequency domain, the diagnostic ability of the CNN-BiLSTM model is effectively improved. Therefore, compared with the existing bearing fault diagnosis methods, the bearing fault diagnosis method provided in this embodiment has higher diagnostic accuracy and can exhibit better adaptability and generalization ability under different working conditions and various types of complex environments.

[0066] This embodiment also provides a bearing fault diagnosis device, which includes a data acquisition module, a data processing module, a model improvement module, a model processing module, and a fault diagnosis module. The data acquisition module acquires a bearing vibration signal dataset. The data processing module decomposes the bearing vibration signal dataset by using an improved empirical mode decomposition algorithm to generate a time-domain dataset; performs a fast Fourier transform operation on the time-domain dataset to generate a frequency-domain dataset. The model construction module constructs an improved CNN-BiLSTM network model, and the CNN-BiLSTM network model includes a one-dimensional convolutional neural network, a bidirectional long short-term memory network, a time-domain to frequency-domain cross-attention mechanism, a normalization layer, and a fully connected layer. The model processing module: the one-dimensional convolutional neural network extracts features from the time-domain dataset to obtain time-domain features, the bidirectional long short-term memory network extracts features from the frequency-domain dataset to obtain frequency-domain features, and the time-domain to frequency-domain cross-attention mechanism performs feature interaction on the time-domain features and the frequency-domain features to obtain fused features. The fault diagnosis module inputs the fused features into the normalization layer and the fully connected layer, and the normalization layer and the fully connected layer output the fault category. The bearing fault diagnosis device provided in this embodiment can implement any of the bearing fault diagnosis methods, and the specific working process of the bearing fault diagnosis device can refer to the corresponding process in the embodiment of the bearing fault diagnosis method. The methods and devices provided in this embodiment can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for example, the division of a certain module is only a logical function division, and there may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the connections or communication connections shown or discussed with each other can be indirect coupling or communication connections through some interfaces, devices or units, and can also be electrical, mechanical or other forms of connections.

[0067] This embodiment also provides a computer device. A computer device includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it implements the bearing fault diagnosis method described above.

[0068] This embodiment also provides a computer-readable storage medium. A computer-readable storage medium stores a computer program, and when the computer program is run by a processor, it executes the bearing fault diagnosis method described in this embodiment. Among them, the computer-readable storage medium can be any tangible medium that contains or stores a program, and this program can be used by or combined with an instruction execution system, device, or device; the program code contained on the computer-readable medium can be transmitted by any suitable medium, including but not limited to wireless, wire, optical fiber, RF, etc., or any suitable combination of the above.

[0069] The above has schematically described the present invention and its implementation manners. This description is not restrictive. Without departing from the spirit or basic characteristics of the present invention, the present invention can be implemented in other specific forms. What is shown in the drawings is only one of the implementation manners of the present invention, and the actual structure is not limited thereto. Any reference signs in the claims should not limit the claims involved. Therefore, if those of ordinary skill in the art are inspired by it and, without departing from the purpose of this creation, design similar structural manners and embodiments to this technical solution without creative efforts, they shall fall within the protection scope of the present invention. In addition, the term "comprising" does not exclude other elements or steps, and the term "a" before an element does not exclude including "a plurality of" such elements. The plurality of elements stated in the product claims can also be implemented by one element through software or hardware. The terms such as "first" and "second" are used to indicate names and do not represent any specific order.

Claims

1. A bearing fault diagnosis method, comprising the following steps: Collect bearing vibration signal data set; The improved empirical mode decomposition algorithm is used to decompose the bearing vibration signal data set into a time domain data set, and the time domain data set is subjected to a fast Fourier transform operation to generate a frequency domain data set. Constructing an improved CNN-BiLSTM network model, wherein the CNN-BiLSTM network model includes a one-dimensional convolutional neural network, a bidirectional long short-term memory network, a time-domain-frequency-domain cross attention mechanism, a normalization layer, and a fully connected layer; The one-dimensional convolutional neural network extracts features from the time domain data set to obtain time domain features, the bidirectional long short-term memory network extracts features from the frequency domain data set to obtain frequency domain features, and the time domain-frequency domain cross attention mechanism performs feature interaction on the time domain features and the frequency domain features to obtain fusion features; The fused features are input into the normalization layer and the fully connected layer, which output the fault category.

2. A bearing fault diagnosis method according to claim 1, characterized in that: The improved empirical mode decomposition algorithm is used to decompose the bearing vibration signal data set into a time domain data set. The specific steps include: Noise is added during the process of decomposing the bearing vibration signal data set using the empirical mode decomposition algorithm to obtain the bearing vibration signal data set after the noise is added; The average local mean of the bearing vibration signal data set after adding noise is calculated to generate a time domain data set.

3. A bearing fault diagnosis method according to claim 2, characterized in that: Perform a fast Fourier transform operation on the time domain data set, output the corresponding amplitude spectrum, and then calculate the actual frequency value corresponding to each frequency component according to the length and sampling frequency set by the fast Fourier transform, retain the actual frequency value and the positive frequency part in the amplitude spectrum, and generate a frequency domain data set.

4. A bearing fault diagnosis method according to claim 1, characterized in that: The one-dimensional convolutional neural network is composed of four convolutional modules connected in series in sequence, and the convolutional module includes two 3×1 convolutional layers, a 2×1 maximum pooling layer and a Relu activation function.

5. A bearing fault diagnosis method according to claim 1, characterized in that: The bidirectional long short-term memory network includes a forward LSTM layer and a backward LSTM layer, and the basic unit of the LSTM layer includes an input gate, a forget gate, an output gate and a cell state.

6. A bearing fault diagnosis method according to claim 4-5, characterized in that: After the time domain features and frequency domain features of the time domain-frequency domain cross attention mechanism are linearly transformed respectively, the frequency domain features generate a query vector, and the time domain features generate a key vector and a value vector.

7. A bearing fault diagnosis method according to claim 6, characterized in that: The attention weight is calculated by the query vector of the frequency domain feature and the key vector of the time domain feature, and then the value vector of the time domain feature is weighted and summed with the attention weight to obtain the fused feature.

8. A bearing fault diagnosis device, characterized in that: include: Data acquisition module, collecting bearing vibration signal data set; The data processing module uses an improved empirical mode decomposition algorithm to decompose the bearing vibration signal data set to generate a time domain data set, and performs a fast Fourier transform operation on the time domain data set to generate a frequency domain data set; A model improvement module is constructed based on an improved CNN-BiLSTM network model, wherein the CNN-BiLSTM network model includes a one-dimensional convolutional neural network, a bidirectional long short-term memory network, a time-domain-frequency-domain cross attention mechanism, a normalization layer, and a fully connected layer; The model processing module comprises: the one-dimensional convolutional neural network extracts features from the time domain data set to obtain time domain features; the bidirectional long short-term memory network extracts features from the frequency domain data set to obtain frequency domain features; and the time domain-frequency domain cross attention mechanism performs feature interaction on the time domain features and the frequency domain features to obtain fusion features; The fault diagnosis module inputs the fused features into a normalization layer and a fully connected layer, and the normalization layer and the fully connected layer output a fault category.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program that can be run on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is executed.

Images