A bearing fault diagnosis method and system based on hypergraph neural network

Abstract

The present invention discloses a bearing fault diagnosis method and system based on a hypergraph neural network, which relates to the technical field of hypergraph neural networks and includes the following steps: using a sensor to collect the vibration signal of the bearing under different operating conditions, and converting the vibration signal into a two-dimensional time-frequency graph; extracting features from the two-dimensional time-frequency graph to construct a feature training set; constructing a hypergraph structure according to the known associations between data in the feature training set; inputting the hypergraph structure into a preset hypergraph neural network for model training until the model converges to obtain a bearing fault diagnosis model; and using the bearing fault diagnosis model to perform fault diagnosis on a target bearing to obtain a diagnosis result. The present invention can improve the efficiency and accuracy of bearing fault diagnosis and solve the problem of insufficient bearing fault data.

Description

Technical Field

[0001] The present invention relates to the technical field of hypergraph neural network, and more specifically to a bearing fault diagnosis method and system based on hypergraph neural network. Background Art

[0002] Bearings are an important component in contemporary mechanical equipment. Their main function is to support mechanical rotating bodies, reduce the mechanical load friction coefficient of the equipment during the transmission process and ensure the rotation accuracy. Their accuracy, performance, life and reliability play a decisive role in the accuracy, performance and reliability of the main machine.

[0003] With the development of industrial automation and the widespread application of mechanical equipment, the detection of the health of bearings is a key link to ensure the normal operation of high-performance transmission systems. Once a transmission system fails, it may cause production stagnation and have a negative impact on the entire production process. Identifying and solving potential problems before failures occur is an important means to ensure the continued stable operation of equipment and avoid huge economic losses. However, the current bearing diagnosis method cannot achieve fault diagnosis efficiently and accurately, which affects normal maintenance; it is difficult to obtain a large amount of sample data, resulting in inaccurate models; it is highly complex and difficult to implement.

[0004] Therefore, how to improve the efficiency and accuracy of bearing fault diagnosis and solve the problem of insufficient bearing fault data are technical problems that technical personnel in this field urgently need to solve. Summary of the invention

[0005] In view of this, the present invention provides a bearing fault diagnosis method and system based on a hypergraph neural network, which solves the problems existing in the background technology.

[0006] In order to achieve the above object, the present invention provides the following technical solutions:

[0007] A bearing fault diagnosis method based on a hypergraph neural network comprises the following steps:

[0008] The sensor is used to collect the vibration signals of the bearing under different operating conditions and convert the vibration signals into a two-dimensional time-frequency diagram;

[0009] Extract features from the two-dimensional time-frequency graph and construct a feature training set;

[0010] Construct a hypergraph structure based on the known associations between data in the feature training set;

[0011] The hypergraph structure is input into the preset hypergraph neural network for model training until the model converges to obtain a bearing fault diagnosis model;

[0012] The bearing fault diagnosis model is used to diagnose the fault of the target bearing and obtain the diagnosis result.

[0013] Optionally, the sampling frequencies of vibration signals under different operating conditions are consistent, and the number of data sampled each time is consistent.

[0014] Optionally, the vibration signal is converted into a two-dimensional time-frequency diagram using a continuous wavelet transform algorithm, and the specific process is expressed as follows:

[0015]

[0016] Where f(x) represents the one-dimensional vibration signal, represents the wavelet basis function, * represents the complex conjugate symbol, a represents the scale factor, and b represents the translation parameter.

[0017] Optionally, feature extraction for the two-dimensional time-frequency graph is performed using a gated multi-layer convolution unit that combines a gating mechanism with a multi-layer convolutional network, taking the two-dimensional time-frequency graph as an input sequence and providing time-frequency domain features between adjacent time steps to an upper layer at each time step.

[0018] Optionally, construct a feature training set, specifically including the following steps:

[0019] Based on the vibration signal, extract the time domain features and frequency domain features;

[0020] The time domain features, frequency domain features and time-frequency domain features extracted through the two-dimensional time-frequency diagram are fused to obtain the fault feature vector and construct a feature training set.

[0021] Optionally, the method further includes performing sample expansion on the feature training set, specifically including the following steps:

[0022] The additive Gaussian noise model is used to introduce noise information into the time domain features at different time domain scales and perform weighted fusion to obtain time domain expansion samples.

[0023] The spectrum information is cross-fused between several samples with the same fault type by means of spectrum weighted superposition and the time domain feature inverse transformation is performed to obtain frequency domain expansion samples;

[0024] The data enhancement method is used to rotate, flip and translate the two-dimensional time-frequency diagram and then extract the time-frequency domain features to obtain the time-frequency domain expansion samples;

[0025] Based on the time domain expanded samples, the frequency domain expanded samples and the time-frequency domain expanded samples, an expanded sample set is generated.

[0026] Optionally, construct a hypergraph structure, which includes the following steps:

[0027] Determine the similarity between each fault feature vector in the feature training set and obtain a similarity matrix;

[0028] The similarity matrix is ​​converted into a hypergraph, and a hypergraph structure G = {V, E, W} is constructed, where V represents the set of nodes of known fault type data and unknown fault type data, E represents the set of hyperedges obtained by transforming the pairwise associated edges abstracted from the known associations between the data, and W represents the weight corresponding to each hyperedge.

[0029] Optionally, the hypergraph neural network includes: an attention mechanism layer, a batch normalization layer, a dropout layer, and a fully connected layer;

[0030] Attention mechanism layer, used to introduce attention weights between channels;

[0031] Batch Normalization layers, which normalize the input of each layer to have zero mean and unit variance;

[0032] Dropout layer, used to randomly discard some neurons;

[0033] The fully connected layer is used to perform linear transformation and output to the next layer.

[0034] Optionally, select the binary cross entropy loss function as the loss function for model optimization, use the back-propagation gradient descent algorithm to update the weights, adjust the hyperparameters and verify the model stability through five-fold cross validation.

[0035] A bearing fault diagnosis system based on a hypergraph neural network, comprising:

[0036] A collection module, used to collect vibration signals of the bearing under different operating conditions through sensors;

[0037] A conversion module, used for converting the vibration signal into a two-dimensional time-frequency diagram;

[0038] The feature extraction module is used to extract features from the two-dimensional time-frequency graph and construct a feature training set;

[0039] The construction module is used to construct a hypergraph structure based on the known associations between data in the feature training set;

[0040] A training module is used to input the hypergraph structure into a preset hypergraph neural network for model training until the model converges to obtain a bearing fault diagnosis model;

[0041] The diagnosis module is used to perform fault diagnosis on the target bearing through the bearing fault diagnosis model to obtain the diagnosis result.

[0042] It can be seen from the above technical solutions that, compared with the prior art, the present invention provides a bearing fault diagnosis method and system based on a hypergraph neural network, which uses continuous wavelet transform to process several bearing vibration signals collected under various operating states and perform feature extraction and deep fusion, and then fuses with other sample data to construct a training set based on the more comprehensive features obtained; in addition, the training set can be further expanded with samples, and the sample set can be enriched with the original training set and new samples to solve the problem of insufficient bearing fault data. A hypergraph structure is constructed based on the known associations between the data in the training set, and the hypergraph structure is input into a preset hypergraph neural network and the model is trained using a loss function and a gradient descent algorithm, which can make the prediction accuracy of the obtained fault diagnosis model higher and improve the accuracy of the fault diagnosis results. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0044] Figure 1 A flow chart of a bearing fault diagnosis method based on a hypergraph neural network provided by the present invention;

[0045] Figure 2 A structural diagram of the bearing fault diagnosis system based on the hypergraph neural network provided by the present invention. DETAILED DESCRIPTION

[0046] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0047] The embodiment of the present invention discloses a bearing fault diagnosis method based on a hypergraph neural network, such as Figure 1 As shown, the following steps are included:

[0048] The sensor is used to collect the vibration signals of the bearing under different operating conditions and convert the vibration signals into a two-dimensional time-frequency diagram;

[0049] Extract features from the two-dimensional time-frequency graph and construct a feature training set;

[0050] Construct a hypergraph structure based on the known associations between data in the feature training set;

[0051] The hypergraph structure is input into the preset hypergraph neural network for model training until the model converges to obtain a bearing fault diagnosis model;

[0052] The bearing fault diagnosis model is used to diagnose the fault of the target bearing and obtain the diagnosis result.

[0053] Furthermore, the sampling frequencies of the vibration signals under different operating conditions are consistent, and the number of data sampled each time is consistent.

[0054] Furthermore, the vibration signal is converted into a two-dimensional time-frequency diagram using the continuous wavelet transform algorithm. CWT is a type of wavelet transform, which uses a time-frequency window that changes with frequency to overcome the short-time Fourier transform that does not change with frequency or time. It cannot confirm the relationship between the corresponding frequency and time, and is more suitable for processing signals with transient mutations. The specific process is expressed as follows:

[0055]

[0056] Where f(x) represents the one-dimensional vibration signal, represents the wavelet basis function, * represents the complex conjugate symbol, a represents the scale factor, and b represents the translation parameter.

[0057] Compared with the standard Fourier transform, the wavelet basis function used in wavelet analysis is not unique. Therefore, in engineering applications, a very important issue is the selection of the optimal wavelet basis function, which will affect the results of the same problem. In this embodiment, the quality of the wavelet basis function can be determined by the error between the processing result of the wavelet analysis method and the theoretical result. Commonly used wavelet basis functions include Haar wavelet, Daubechies wavelet, Mexican Hat wavelet, Morlet wavelet, Meyer wavelet, etc.

[0058] Furthermore, the feature extraction of the two-dimensional time-frequency graph uses a gated multi-layer convolution unit that combines a gating mechanism and a multi-layer convolutional network. The two-dimensional time-frequency graph is used as the input sequence, and the time-frequency domain features between adjacent time steps are provided to the upper layer at each time step, taking into account the strong dependency that may exist between adjacent time steps in the time series. Compared with a single-layer convolutional network, a multi-layer convolutional network can extract complex features in the two-dimensional time-frequency graph at a deeper level and improve the prediction accuracy of the model; in addition, the gated multi-layer convolution unit can also provide the upper layer with local feature information between adjacent time steps, enhancing the feature extraction capability.

[0059] Furthermore, constructing a feature training set specifically includes the following steps:

[0060] Based on the vibration signal, extract the time domain features and frequency domain features;

[0061] The time domain features, frequency domain features and time-frequency domain features extracted through the two-dimensional time-frequency diagram are fused to obtain the fault feature vector and construct a feature training set.

[0062] In this embodiment, by extracting time domain features and frequency domain features and then fusing them with the time-frequency domain features extracted based on the two-dimensional time-frequency diagram, more comprehensive features related to the bearing fault can be obtained. Constructing a feature training set on this basis can improve the prediction accuracy of the bearing fault diagnosis model and avoid other serious consequences caused by prediction errors.

[0063] Furthermore, the method also includes performing sample expansion on the feature training set, specifically including the following steps:

[0064] The additive Gaussian noise model is used to introduce noise information into the time domain features at different time domain scales and perform weighted fusion to obtain time domain expansion samples.

[0065] The spectrum information is cross-fused between several samples with the same fault type by means of spectrum weighted superposition and the time domain feature inverse transformation is performed to obtain frequency domain expansion samples;

[0066] The data enhancement method is used to rotate, flip and translate the two-dimensional time-frequency diagram and then extract the time-frequency domain features to obtain the time-frequency domain expansion samples;

[0067] Based on the time domain expanded samples, the frequency domain expanded samples and the time-frequency domain expanded samples, an expanded sample set is generated.

[0068] In response to the situation where fault data is insufficient, this embodiment uses the above different methods to expand the samples of time domain features, frequency domain features and time-frequency domain features, which can solve the problems of fault diagnosis being prone to overfitting on small sample data sets and poor model generalization ability; model training based on the sample expansion set can improve the accuracy of the model in bearing fault diagnosis.

[0069] Further, constructing a hypergraph structure specifically includes the following steps:

[0070] Determine the similarity between each fault feature vector in the feature training set and obtain a similarity matrix;

[0071] The similarity matrix is ​​converted into a hypergraph, and a hypergraph structure G = {V, E, W} is constructed, where V represents the set of nodes of known fault type data and unknown fault type data, E represents the set of hyperedges obtained by transforming the pairwise associated edges abstracted from the known associations between the data, and W represents the weight corresponding to each hyperedge.

[0072] Furthermore, the hypergraph neural network includes: attention mechanism layer, batch normalization layer, dropout layer and fully connected layer;

[0073] The attention mechanism layer is used to introduce attention weights between channels, which can enhance the model's attention to the features of different channels, help better capture the most useful features, and improve performance;

[0074] Batch Normalization layers, which normalize the input of each layer to have zero mean and unit variance, can speed up training and improve the stability of the model;

[0075] The Dropout layer is used to randomly discard a number of neurons to reduce overfitting and improve the generalization performance of the model;

[0076] The fully connected layer is used to perform linear transformation and output to the next layer.

[0077] Furthermore, during model training, the binary cross entropy loss function is selected as the loss function for model optimization, the back-propagation gradient descent algorithm is used to update the weights, and the hyperparameters are adjusted and the model stability is verified through a five-fold cross-validation method.

[0078] and Figure 1 Corresponding to the method described above, an embodiment of the present invention further provides a bearing fault diagnosis system based on a hypergraph neural network for Figure 1 In the specific implementation of the method, a bearing fault diagnosis system based on a hypergraph neural network provided by an embodiment of the present invention can be applied to a computer terminal or various mobile devices, such as Figure 2 As shown, specifically including:

[0079] A collection module, used to collect vibration signals of the bearing under different operating conditions through sensors;

[0080] A conversion module, used for converting the vibration signal into a two-dimensional time-frequency diagram;

[0081] The feature extraction module is used to extract features from the two-dimensional time-frequency graph and construct a feature training set;

[0082] The construction module is used to construct a hypergraph structure based on the known associations between data in the feature training set;

[0083] A training module is used to input the hypergraph structure into a preset hypergraph neural network for model training until the model converges to obtain a bearing fault diagnosis model;

[0084] The diagnosis module is used to perform fault diagnosis on the target bearing through the bearing fault diagnosis model to obtain the diagnosis result.

[0085] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0086] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A bearing fault diagnosis method based on hypergraph neural network, characterized in that: The following steps are involved: The sensor is used to collect the vibration signals of the bearing under different operating conditions and convert the vibration signals into a two-dimensional time-frequency diagram; Extract features from the two-dimensional time-frequency graph and construct a feature training set; Construct a hypergraph structure based on the known associations between data in the feature training set; The hypergraph structure is input into the preset hypergraph neural network for model training until the model converges to obtain a bearing fault diagnosis model; Use the bearing fault diagnosis model to perform fault diagnosis on the target bearing and obtain the diagnosis result; The method further includes performing sample expansion on the feature training set, specifically comprising the following steps: The additive Gaussian noise model is used to introduce noise information into the time domain features at different time domain scales and perform weighted fusion to obtain time domain expansion samples. The spectrum information is cross-fused between several samples with the same fault type by means of spectrum weighted superposition and the time domain feature inverse transformation is performed to obtain frequency domain expansion samples; The data enhancement method is used to rotate, flip and translate the two-dimensional time-frequency diagram and then extract the time-frequency domain features to obtain the time-frequency domain expansion samples; Generate a sample expansion set based on the time domain expansion samples, the frequency domain expansion samples, and the time-frequency domain expansion samples; Constructing a hypergraph structure includes the following steps: Determine the similarity between each fault feature vector in the feature training set and obtain a similarity matrix; The similarity matrix is ​​converted into a hypergraph, and a hypergraph structure G = {V, E, W} is constructed, where V represents the set of nodes of known fault type data and unknown fault type data, E represents the set of hyperedges obtained by transforming the pairwise associated edges abstracted from the known associations between the data, and W represents the weight corresponding to each hyperedge; The hypergraph neural network includes: attention mechanism layer, batch normalization layer, dropout layer and fully connected layer; Attention mechanism layer, used to introduce attention weights between channels; Batch Normalization layers, which normalize the input of each layer to have zero mean and unit variance; Dropout layer, used to randomly discard some neurons; The fully connected layer is used to perform linear transformation and output to the next layer.

2. A bearing fault diagnosis method based on a hypergraph neural network according to claim 1, characterized in that: The sampling frequency of the vibration signals under different operating conditions is consistent, and the number of data sampled each time is consistent.

3. A bearing fault diagnosis method based on hypergraph neural network according to claim 1, characterized in that: The continuous wavelet transform algorithm is used to convert the vibration signal into a two-dimensional time-frequency diagram. The specific process is expressed as follows: Where f(x) represents the one-dimensional vibration signal, represents the wavelet basis function, * represents the complex conjugate symbol, a represents the scale factor, and b represents the translation parameter.

4. The method for bearing fault diagnosis based on hypergraph neural network according to claim 1 is characterized in that: The feature extraction of the two-dimensional time-frequency graph adopts a gated multi-layer convolution unit that combines a gating mechanism and a multi-layer convolutional network. The two-dimensional time-frequency graph is used as the input sequence, and the time-frequency domain features between adjacent time steps are provided to the upper layer at each time step.

5. A bearing fault diagnosis method based on hypergraph neural network according to claim 4, characterized in that: Constructing a feature training set includes the following steps: Based on the vibration signal, extract the time domain features and frequency domain features; The time domain features, frequency domain features and time-frequency domain features extracted through the two-dimensional time-frequency diagram are fused to obtain the fault feature vector and construct a feature training set.

6. A bearing fault diagnosis method based on hypergraph neural network according to claim 1, characterized in that: During model training, the binary cross entropy loss function is selected as the loss function for model optimization, the back-propagation gradient descent algorithm is used to update the weights, and the hyperparameters are adjusted and the model stability is verified through a five-fold cross-validation method.

7. A bearing fault diagnosis system based on a hypergraph neural network for implementing the method according to any one of claims 1 to 6, characterized in that: include: A collection module, used to collect vibration signals of the bearing under different operating conditions through sensors; A conversion module, used for converting the vibration signal into a two-dimensional time-frequency diagram; The feature extraction module is used to extract features from the two-dimensional time-frequency graph and construct a feature training set; The construction module is used to construct a hypergraph structure based on the known associations between data in the feature training set; A training module is used to input the hypergraph structure into a preset hypergraph neural network for model training until the model converges to obtain a bearing fault diagnosis model; The diagnosis module is used to perform fault diagnosis on the target bearing through the bearing fault diagnosis model to obtain the diagnosis result.

Images