Rotating machinery system fault diagnosis method, system, electronic equipment and storage medium

Abstract

The present invention discloses a method, system, electronic device, and storage medium for diagnosing faults in rotating machinery systems, relating to the field of fault diagnosis. The method comprises obtaining an original signal from the rotating machinery system; extracting features from the original signal using a short-time Fourier transform and a convolutional neural network to obtain feature information; pre-classifying the feature information using a multi-head attention mechanism and a feedforward neural network to obtain category perception information; classifying and correcting the category perception information using a graph neural network to obtain correction information; the graph neural network includes a static graph neural network and a dynamic graph neural network; weighting and activating the category perception information and the correction information to obtain a classification coding matrix; and using the classification coding matrix to determine the type of a composite fault. The present invention can improve fault diagnosis accuracy.

Description

Technical Field

[0001] The present invention relates to the field of fault diagnosis, and in particular to a method, system, electronic equipment and storage medium for diagnosing faults of a rotating machinery system. Background Art

[0002] With the rapid development of modern industry, modern industrial equipment is developing towards high complexity, high precision, high efficiency, large-scale and automation. Rotating machinery, such as gearboxes, are very basic and important equipment. Therefore, the failure life prediction and fault diagnosis of rotating machinery are very important. Under harsh environments and working conditions, rotating machinery systems, especially rotating components such as gears and bearings, are prone to failure. The failures may affect the normal operation of other components, resulting in catastrophic failures and huge economic losses. Most of these are complex failures, consisting of failures in different parts or different types of failures in the same part. Therefore, an accurate, intelligent and efficient diagnostic method is urgently needed.

[0003] Traditionally, common intelligent fault diagnosis solutions typically consist of two stages: feature extraction and fault mode classification. The feature extraction stage primarily utilizes signal processing techniques and neural network learning. In the fault mode classification stage, the data features extracted in the previous stage are used as input to an intelligent classifier. Commonly used methods include support vector machines (SVMs), K-nearest neighbor (KNN), naive Bayes, and softmax classifiers (SCs). Numerous intelligent fault diagnosis methods have been derived based on these approaches. While current solutions have achieved significant success in rotating machinery, several challenges remain. Rotating machinery typically operates in complex environments. Due to the contamination of vibration signatures with background noise, the performance of traditional methods often depends on the sensitivity of the selected features to different classes, requiring significant expert knowledge and experience. In the field of intelligent fault diagnosis, single and compound faults are often treated as independent fault modes, ignoring the potential relationships between them. Traditional classifiers can only output a single label for a compound fault test sample, not multiple labels. This results in compound faults being unable to be classified as two or more single faults. This means that compound faults cannot be decoupled from single faults during the fault diagnosis process. Based on the above-mentioned defects in intelligent fault diagnosis of rotating machinery in equipment production, it is necessary to study new intelligent fault diagnosis solutions for complex faults of rotating machinery.

[0004] An existing solution for complex faults in rotating machinery involves preprocessing experimental data. This processed one-dimensional data is then fed into a fault classifier for classification. The complex fault classifier primarily consists of a one-dimensional, three-layer convolutional neural network. The resulting data is then reshaped and passed through two fully connected layers to obtain the final result.

[0005] There is also an existing solution for compound fault classification that proposes two steps. The first step is to input the data into the feature extraction module. The feature extraction module here also consists of a one-dimensional neural network. The extracted features are then input into the compound fault decoupling classifier. The compound fault decoupling classifier is an application of the capsule network. Finally, the fault classification label is obtained through L2 normalization.

[0006] The first existing solution mentioned that the features extracted by a simple neural network are activated and classified through a nonlinear function, and the result is not accurate. As for the capsule neural network mentioned in the second existing solution, the network parallelism is not high and the training is difficult. At the same time, the characteristics of the complex faults of rotating machinery are more prominent in the frequency domain than in the time domain, and the above solutions all use one-dimensional signals and cannot extract frequency domain features, so the classification error is high. There is a relationship between the faults in rotating machinery. For example, the occurrence of fault A is often accompanied by fault B, or the occurrence of fault A generally does not cause fault C. These can be intervened according to the technical experience of experts, but there are many types of rotating machinery faults, which is very time-consuming and labor-intensive. The most important disadvantage is that the correlation between rotating machinery faults is not included in the fault classification, which is the second point mentioned, so the classification accuracy is not high. Summary of the Invention

[0007] The object of the present invention is to provide a method, system, electronic device and storage medium for diagnosing faults in a rotating machinery system, so as to improve the accuracy of fault diagnosis.

[0008] To achieve the above object, the present invention provides the following solutions:

[0009] A method for diagnosing faults in a rotating machinery system, comprising:

[0010] Obtain the original signal of the rotating mechanical system;

[0011] Performing feature extraction on the original signal using short-time Fourier transform and convolutional neural network to obtain feature information;

[0012] Pre-classifying the feature information using a multi-head attention mechanism and a feedforward connection neural network to obtain category perception information;

[0013] Classify and correct the category perception information using a graph neural network to obtain corrected information; the graph neural network includes a static graph neural network and a dynamic graph neural network;

[0014] The category perception information and the correction information are weighted and activated to obtain a classification coding matrix; the classification coding matrix is used to determine the type of composite fault.

[0015] Optionally, feature extraction is performed on the original signal using short-time Fourier transform and convolutional neural network to obtain feature information, specifically including:

[0016] Performing short-time Fourier transform on the original signal to obtain a transformed signal;

[0017] The transformed signal is subjected to feature extraction using a convolutional neural network to obtain feature information; the convolutional neural network includes a two-dimensional convolutional layer, a batch normalization layer, a rule activation function, and a maximum pooling layer connected in sequence.

[0018] Optionally, the category perception information is classified and corrected using a graph neural network to obtain corrected information; the graph neural network includes a static graph neural network and a dynamic graph neural network, specifically including:

[0019] Inputting the category perception information into the static graph neural network to obtain a static graph neural network output result;

[0020] The output result of the static graph neural network is subjected to global maximum pooling using a maximum pooling layer to obtain global feature information;

[0021] The dynamic graph neural network is used to perform dynamic category perception and activation according to the global feature information to obtain correction information.

[0022] Optionally, the category perception information and the correction information are weighted and activated to obtain a classification coding matrix, specifically including:

[0023] The category perception information and the correction information are weighted and activated using a fully connected layer and a Sigmoid function to obtain a classification coding matrix.

[0024] The present invention also provides a rotating machinery system fault diagnosis system, comprising:

[0025] An acquisition module, used to acquire the original signal of the rotating mechanical system;

[0026] A feature extraction module is used to extract features from the original signal using short-time Fourier transform and convolutional neural network to obtain feature information;

[0027] A pre-classification module is used to pre-classify the feature information using a multi-head attention mechanism and a feedforward connection neural network to obtain category perception information;

[0028] A classification correction module is used to classify and correct the category perception information using a graph neural network to obtain correction information; the graph neural network includes a static graph neural network and a dynamic graph neural network;

[0029] The weighting and activation module is used to weight and activate the category perception information and the correction information to obtain a classification coding matrix; the classification coding matrix is used to determine the type of composite fault.

[0030] Optionally, the feature extraction module specifically includes:

[0031] A short-time Fourier transform unit, configured to perform a short-time Fourier transform on the original signal to obtain a transformed signal;

[0032] A feature extraction unit is used to extract features from the transformed signal using a convolutional neural network to obtain feature information; the convolutional neural network includes a two-dimensional convolution layer, a batch normalization layer, a rule activation function and a maximum pooling layer connected in sequence.

[0033] Optionally, the classification correction module specifically includes:

[0034] a static graph neural network processing unit, configured to input the category perception information into the static graph neural network to obtain a static graph neural network output result;

[0035] A global maximum pooling unit is used to perform global maximum pooling on the output result of the static graph neural network using a maximum pooling layer to obtain global feature information;

[0036] A dynamic category perception and activation unit is used to perform dynamic category perception and activation using the dynamic graph neural network according to the global feature information to obtain correction information.

[0037] Optionally, the weighted and activated module specifically includes:

[0038] The weighted and activated unit is used to weight and activate the category perception information and the correction information using a fully connected layer and a Sigmoid function to obtain a classification coding matrix.

[0039] The present invention further provides an electronic device, comprising:

[0040] one or more processors;

[0041] a storage device having one or more programs stored thereon;

[0042] When the one or more programs are executed by the one or more processors, the one or more processors are caused to implement the above method.

[0043] The present invention also provides a storage medium having a computer program stored thereon, wherein the computer program implements the above method when executed by a processor.

[0044] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0045] The present invention extracts features from the original signal using a short-time Fourier transform (SFT) and a convolutional neural network to obtain feature information. This feature information is pre-classified using a multi-head attention mechanism and a feedforward neural network to obtain category-aware information. This category-aware information is then classified and corrected using a graph neural network to obtain corrected information. The graph neural network includes a static graph neural network and a dynamic graph neural network. The category-aware information and the corrected information are weighted and activated to obtain a classification encoding matrix. This classification encoding matrix is used to determine the type of complex fault. Vibration signals collected from rotating machinery are typically one-dimensional, but fault features in the frequency domain are more prominent than those in the time domain. Therefore, a short-time Fourier transform is performed on the signal to obtain a two-dimensional signal containing both time and frequency domain information. This allows for more efficient extraction of distinguishing features in the next step of feature extraction. The multi-head attention mechanism employed has the advantage of adaptively focusing on different regions of the embedded label, resulting in high final classification accuracy. By correcting the pre-classified signal through a graph neural network, the correlation between faults plays a role in the diagnosis of complex faults in rotating machinery, allowing for more accurate identification of complex fault types. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0047] Figure 1 A flow chart of the rotating machinery system fault diagnosis method provided by the present invention;

[0048] Figure 2 Schematic diagram of the fault diagnosis method for rotating machinery systems. DETAILED DESCRIPTION

[0049] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. 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 making creative efforts are within the scope of protection of the present invention.

[0050] The object of the present invention is to provide a method, system, electronic device and storage medium for diagnosing faults in a rotating machinery system, so as to improve the accuracy of fault diagnosis.

[0051] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0052] like Figure 1 and Figure 2 As shown, the present invention provides a method for diagnosing faults in a rotating machinery system, comprising:

[0053] Step 101: Acquire the original signal of the rotating mechanical system.

[0054] Step 102: Extract features from the original signal using a short-time Fourier transform and a convolutional neural network to obtain feature information. Step 102 specifically includes: performing a short-time Fourier transform on the original signal to obtain a transformed signal; extracting features from the transformed signal using a convolutional neural network to obtain feature information; the convolutional neural network includes a two-dimensional convolutional layer, a batch normalization layer, a rule activation function, and a maximum pooling layer connected in sequence.

[0055] Generally speaking, the characteristics of different faults are more prominent in the frequency domain than in the time domain. Therefore, a short-time Fourier transform is performed on the signal. This extracts the frequency-domain information of the vibration signal while also preserving some time-domain information. The processed time-frequency signal is then fed into the feature extraction module. Since the time-frequency signal is two-dimensional, a three-layer two-dimensional convolutional neural network is selected as the feature extraction module.

[0056] The original information is the input composite fault signal Singal, which provides the original signal As the input signal, x is a one-dimensional signal, where L represents the length of the signal. After the short-time Fourier transform STFT, the signal Where L1 and H1 are the length and width of the transformed signal respectively. After that, the obtained signal is used as the input of feature extraction and passes through a three-layer two-dimensional convolutional neural network CNN, which consists of two-dimensional convolution, batch normalization, relu activation function and maximum pooling layer. The feature signal is finally extracted as Where L2 is the length, H2 is the width, and M is the intermediate dimension related to the fault label query in the next stage.

[0057] Step 103: Pre-classify the feature information using a multi-head attention mechanism and a feedforward connection neural network to obtain category perception information.

[0058] The extracted feature information is then fed into the proposed composite fault decoupling classification module. This module consists of two main parts: 1. a classification module centered around a multi-head attention mechanism, and 2. a classification correction module centered around a graph neural network. The multi-head attention mechanism and the feedforward neural network form N×N pre-classification modules.

[0059] The multi-head attention mechanism Multihead has three inputs (Q, K, V,) where Q represents the input of the query label and generates a learnable query label matrix Where K1 represents the number of single faults included in the fault classification, and M is the intermediate dimension corresponding to the feature matrix obtained by feature extraction. K is the output matrix of the feature extraction module, that is, the extracted features, and V is the position encoding embedding matrix added on the basis of K. According to the input features and label space, it will pass through the multi-head attention module and the feedforward connection neural network FFN in turn to iteratively update the label space layer by layer. The iterative formula is as follows (1) The multi-head attention mechanism can adaptively focus on the local area related to the fault category in the feature to obtain category perception information This stage is to pre-classify the complex faults of rotating machinery and input the pre-classification results into the graph neural network to further correct the classification results.

[0060] Multi-head attention formula: Q i =Multihead(Q i-1 , K, V)#(1)

[0061] where Q i is the i-th query label.

[0062] Step 104: Classify and correct the category perception information using a graph neural network to obtain corrected information; the graph neural network includes a static graph neural network and a dynamic graph neural network. Step 104 specifically includes: inputting the category perception information into the static graph neural network to obtain a static graph neural network output result; performing global maximum pooling on the static graph neural network output result using the dynamic graph neural network to obtain global feature information; and performing dynamic category perception and activation using the dynamic graph neural network based on the global feature information to obtain corrected information.

[0063] Graph neural network mainly consists of static graph neural network (SGCN) and dynamic graph neural network (DGCN). The formula of SGCN is the output result of static graph neural network S = δ(C s AW s ), where δ is the activation function Relu, S is the output of the static graph neural network, which represents the coarse dependency between single fault labels, and A is the category perception information obtained by the previous pre-classification step, C s is the category-aware correlation matrix, is the state weight update matrix, these two parameters are initialized according to the neural network training, and C sAfter determination, it is fixed and works for all samples, so that the coarse dependency relationship of a single fault in a composite fault is obtained. DGCN further processes the obtained S, first performing global maximum pooling on S, extracting global features and merging them with S to obtain the global feature signal A'. The key in DGCN is the dynamic category perception matrix C d =Sigmoid(WA'), where W is the weight matrix. From this we can see that C d It changes with the input sample, which gives it the ability to target different data. The formula of DGCN is D=δ(C d SW d ),in δ is the activation function Relu.

[0064] Step 105: Weighting and activating the category perception information and the correction information to obtain a classification encoding matrix; the classification encoding matrix is used to determine the type of composite fault. Step 105 specifically includes: weighting and activating the category perception information and the correction information using a fully connected layer and a Sigmoid function to obtain a classification encoding matrix.

[0065] Finally, A and D are processed by the fully connected layer and the Sigmoid function to obtain the classification information and weighted to obtain the final result. The classification coding matrix The formula is Where W A , W D is the weight matrix of the fully connected layer. The Sigmoid function is a binary classification function. Composite faults use hot encoding labels, so we only need to determine whether a certain fault occurs and then comprehensively obtain the composite fault type.

[0066] The entire process provided by this invention can be divided into two steps. The first step is to preprocess the data and extract features. The second step is to input the extracted features into a composite fault decoupling classifier to train the model. The first step mainly utilizes the short-time Fourier transform and a three-layer two-dimensional neural network, while the second step uses a multi-head attention mechanism and a neural network. Ultimately, the end-to-end decoupling classification task of composite faults in rotating machinery is achieved.

[0067] 1. There are two steps to troubleshooting:

[0068] (1) Data preprocessing and feature extraction

[0069] Short-time Fourier transform was used to process the data, and a three-layer two-dimensional convolutional neural network was used for feature extraction.

[0070] (2) Classification of composite fault decoupling

[0071] 1. The extracted features are processed in three steps: ① Pre-classification using a multi-head attention mechanism to obtain content-aware category information; ② Further correction of the content-aware category information using dynamic and static graph neural networks; ③ The classification information obtained from the first two steps is weighted and summed to obtain the final classification hot encoding matrix.

[0072] 2. The multi-head attention mechanism is used to pre-classify the extracted features. The attention mechanism can spontaneously pay attention to features at different positions according to the embedding of the label, and the advantage of the multi-head mechanism is that it can extract features from different aspects.

[0073] 3. Faults in current rotating machinery, such as gears and bearings, can occur simultaneously, or one fault can occur alongside another. This indicates that these faults are not independent but rather correlated and induce each other. Here, we use graph neural networks to refine the pre-classification results based on the correlation between faults.

[0074] 4. The Sigmoid function is rarely used for compound faults. Here, the Sigmoid function is used for binary classification. Each type of fault is judged as either present or absent, that is, output 1 or 0. In this way, no matter how the faults are combined, they are not considered as a new fault.

[0075] Vibration signals collected from rotating machinery are typically one-dimensional, but fault characteristics are more prominent in the frequency domain than in the time domain. Therefore, a short-time Fourier transform (SFT) is performed on the signal to produce a two-dimensional signal that contains both time and frequency domain information. This allows for more effective extraction of distinguishing features during the next step of feature extraction. The advantage of the multi-head attention mechanism is that it adaptively focuses on different regions of the embedded label, resulting in high final classification accuracy. By modifying the pre-classified signal through a graph neural network, the correlation between faults plays a role in the diagnosis of complex rotating machinery faults, allowing for more accurate identification of complex fault types.

[0076] The present invention also provides a rotating machinery system fault diagnosis system, comprising:

[0077] The acquisition module is used to obtain the original signal of the rotating mechanical system.

[0078] The feature extraction module is used to extract features from the original signal using short-time Fourier transform and convolutional neural network to obtain feature information.

[0079] The pre-classification module is used to pre-classify the feature information using a multi-head attention mechanism and a feedforward connection neural network to obtain category perception information.

[0080] A classification correction module is used to perform classification correction on the category perception information using a graph neural network to obtain correction information; the graph neural network includes a static graph neural network and a dynamic graph neural network.

[0081] The weighting and activation module is used to weight and activate the category perception information and the correction information to obtain a classification coding matrix; the classification coding matrix is used to determine the type of composite fault.

[0082] As an optional implementation, the feature extraction module specifically includes:

[0083] The short-time Fourier transform unit is used to perform short-time Fourier transform on the original signal to obtain a transformed signal.

[0084] A feature extraction unit is used to extract features from the transformed signal using a convolutional neural network to obtain feature information; the convolutional neural network includes a two-dimensional convolution layer, a batch normalization layer, a rule activation function and a maximum pooling layer connected in sequence.

[0085] As an optional implementation, the classification correction module specifically includes:

[0086] The static graph neural network processing unit is used to input the category perception information into the static graph neural network to obtain a static graph neural network output result.

[0087] The global maximum pooling unit is used to perform global maximum pooling on the output result of the static graph neural network using the maximum pooling layer to obtain global feature information.

[0088] A dynamic category perception and activation unit is used to perform dynamic category perception and activation using the dynamic graph neural network according to the global feature information to obtain correction information.

[0089] As an optional implementation, the weighted and activated module specifically includes:

[0090] The weighted and activated unit is used to weight and activate the category perception information and the correction information using a fully connected layer and a Sigmoid function to obtain a classification coding matrix.

[0091] The present invention further provides an electronic device, comprising:

[0092] one or more processors;

[0093] a storage device having one or more programs stored thereon;

[0094] When the one or more programs are executed by the one or more processors, the one or more processors are caused to implement the above method.

[0095] The present invention also provides a storage medium having a computer program stored thereon, wherein the computer program implements the above method when executed by a processor.

[0096] In the process of composite fault diagnosis, the present invention incorporates the correlation between different single faults and the relationship between composite faults, thereby improving the accuracy of composite fault identification in rotating machinery systems. The present invention proposes a model based on the label correlation established by the attention mechanism and the graph neural network, and obtains the fault signal in the time-frequency domain through short-time Fourier transform to extract features that are more relevant to fault classification. The multi-head attention module can adaptively focus on features related to the label, and its parallelism is very high, making model training easy. The graph neural network can establish a label correlation matrix to add some empirical knowledge to make the result classification more accurate.

[0097] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0098] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only intended to help understand the method and core concept of the present invention. At the same time, those skilled in the art will find that the specific implementation methods and application scopes may vary based on the concept of the present invention. In summary, the contents of this specification should not be construed as limiting the present invention.

Claims

1. A method for diagnosing faults in a rotating machinery system, characterized in that: include: Obtain the original signal of the rotating mechanical system; Performing feature extraction on the original signal using short-time Fourier transform and convolutional neural network to obtain feature information; The feature information is pre-classified using a multi-head attention mechanism and a feedforward neural network to obtain category perception information. The input feature information and label space are sequentially passed through the multi-head attention mechanism and the feedforward neural network FFN to iteratively update the label space layer by layer. The iterative formula is as follows: the multi-head attention mechanism can adaptively focus on the local area related to the fault category in the feature to obtain category perception information. The formula of the multi-head attention mechanism is: Q i =Multihead(Q i-1 ,K,V) where Q i is the i-th query label, Q i-1 is the i-1th query label, K is the output matrix of the feature extraction module, that is, the extracted features, and V is the positional encoding embedding matrix added on the basis of K; Performing classification correction on the category perception information using a graph neural network to obtain corrected information; The graph neural network includes a static graph neural network and a dynamic graph neural network, specifically comprising: inputting the category perception information into the static graph neural network to obtain a static graph neural network output result; performing global maximum pooling on the static graph neural network output result using a maximum pooling layer to obtain global feature information; and performing dynamic category perception and activation using the dynamic graph neural network based on the global feature information to obtain correction information; The category perception information and the correction information are weighted and activated to obtain a classification coding matrix; the classification coding matrix is used to determine the type of composite fault.

2. The rotating machinery system fault diagnosis method according to claim 1, characterized in that: The original signal is subjected to feature extraction using short-time Fourier transform and convolutional neural network to obtain feature information, specifically including: Performing short-time Fourier transform on the original signal to obtain a transformed signal; The transformed signal is subjected to feature extraction using a convolutional neural network to obtain feature information; the convolutional neural network includes a two-dimensional convolutional layer, a batch normalization layer, a rule activation function, and a maximum pooling layer connected in sequence.

3. The rotating machinery system fault diagnosis method according to claim 1, characterized in that: The category perception information and the correction information are weighted and activated to obtain a classification coding matrix, which specifically includes: The category perception information and the correction information are weighted and activated using a fully connected layer and a Sigmoid function to obtain a classification coding matrix.

4. A rotating machinery system fault diagnosis system, characterized in that: include: An acquisition module, used to acquire the original signal of the rotating mechanical system; A feature extraction module is used to extract features from the original signal using short-time Fourier transform and convolutional neural network to obtain feature information; The pre-classification module is used to pre-classify the feature information using a multi-head attention mechanism and a feedforward neural network to obtain category perception information. The input feature information and label space will be sequentially passed through the multi-head attention mechanism and the feedforward neural network FFN to iteratively update the label space layer by layer. The iterative formula is as follows: the multi-head attention mechanism can adaptively focus on the local area related to the fault category in the feature to obtain category perception information. The formula of the multi-head attention mechanism is: Q i =Multihead(Q i-1 ,K,V) where Q i is the i-th query label, Q i-1 is the i-1th query label, K is the output matrix of the feature extraction module, that is, the extracted features, and V is the positional encoding embedding matrix added on the basis of K; A classification correction module, configured to perform classification correction on the category perception information using a graph neural network to obtain correction information; The graph neural network includes a static graph neural network and a dynamic graph neural network; The classification correction module specifically includes: a static graph neural network processing unit, which is used to input the category perception information into the static graph neural network to obtain a static graph neural network output result; a global maximum pooling unit, which is used to perform global maximum pooling on the static graph neural network output result using a maximum pooling layer to obtain global feature information; a dynamic category perception and activation unit, which is used to perform dynamic category perception and activation using the dynamic graph neural network based on the global feature information to obtain correction information; The weighting and activation module is used to weight and activate the category perception information and the correction information to obtain a classification coding matrix; the classification coding matrix is used to determine the type of composite fault.

5. The rotating machinery system fault diagnosis system according to claim 4, characterized in that: Feature extraction module, specifically including: A short-time Fourier transform unit, configured to perform a short-time Fourier transform on the original signal to obtain a transformed signal; A feature extraction unit is used to extract features from the transformed signal using a convolutional neural network to obtain feature information; the convolutional neural network includes a two-dimensional convolution layer, a batch normalization layer, a rule activation function and a maximum pooling layer connected in sequence.

6. The rotating machinery system fault diagnosis system according to claim 4, characterized in that: Weighted and activated modules, including: The weighted and activated unit is used to weight and activate the category perception information and the correction information using a fully connected layer and a Sigmoid function to obtain a classification coding matrix.

7. An electronic device, characterized in that: include: one or more processors; a storage device having one or more programs stored thereon; When the one or more programs are executed by the one or more processors, the one or more processors are enabled to implement the method according to any one of claims 1 to 3.

8. A storage medium, characterized in that: A computer program is stored thereon, wherein when the computer program is executed by a processor, the method according to any one of claims 1 to 3 is implemented.

Images