CN116070131B - Fault diagnosis method and system combining multi-scale interactive graph convolution and contrast pooling - Google Patents
Fault diagnosis method and system combining multi-scale interactive graph convolution and contrast poolingInfo
- Publication number
- CN116070131B CN116070131B CN202211606855.XA CN202211606855A CN116070131B CN 116070131 B CN116070131 B CN 116070131B CN 202211606855 A CN202211606855 A CN 202211606855A CN 116070131 B CN116070131 B CN 116070131B
- Authority
- CN
- China
- Prior art keywords
- graph
- pooling
- fault diagnosis
- layer
- contrast
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M13/00—Testing of machine parts
- G01M13/04—Bearings
- G01M13/045—Acoustic or vibration analysis
Landscapes
- Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Acoustics & Sound (AREA)
- Complex Calculations (AREA)
- Image Analysis (AREA)
Abstract
The invention belongs to the technical field of fault diagnosis, and provides a fault diagnosis method and system combining multi-scale interaction graph convolution and contrast pooling. The fault diagnosis method comprises the steps of obtaining vibration signals of a rolling bearing, obtaining a fault type by adopting a fault diagnosis model, wherein the fault diagnosis model comprises a multi-scale interaction graph, a graph rolling network and a contrast learning reinforced self-attention pooling layer, the process of adopting the fault diagnosis model comprises the steps of calculating node embedding vectors and adjacent matrixes based on the vibration signals of the rolling bearing, constructing a multi-scale interaction graph, extracting graph data features by adopting the graph rolling layer of the graph rolling network based on the node embedding vectors and the adjacent matrixes, carrying out graph structure coarsening and feature dimension reduction on the graph data features by adopting the contrast learning reinforced self-attention pooling layer, and sequentially enabling the coarsened graph data to pass through the graph rolling network readout layer and the full-connection layer to obtain the fault type. The invention improves the accuracy and the robustness of the graph rolling network applied to the fault diagnosis of the rotary machine.
Description
Technical Field
The invention belongs to the technical field of fault diagnosis, and particularly relates to a fault diagnosis method and system combining multi-scale interaction graph convolution and contrast pooling.
Background
The statements in this section merely provide background information related to the present disclosure and may not necessarily constitute prior art.
The rotary machine is widely applied to a plurality of fields such as industrial production, national defense, military and the like, such as fans, engines, turbines and the like. With the development of the 5G technology and the development of the Internet of things, the health state monitoring of mechanical equipment enters a big data era, and data support is provided for a fault diagnosis method based on a deep learning framework. Compared with the traditional machine learning method, the data processing mode of combining the deep learning network with the gradient descent algorithm is more suitable for extracting the characteristic with stronger generalization capability from mass data. Therefore, fault diagnosis methods based on a deep learning framework are a popular field of research.
In the field of deep learning, some network frameworks such as self-encoders, convolutional neural networks, fully-connected networks, and the like have been successfully applied to rotary machine fault diagnosis and demonstrate the powerful fault classification capability of feature extraction. For the deep learning method, factors influencing the feature extraction and fault classification effects can be mainly classified into two points, (1) a description mode of input data, and (2) deep learning network parameters. Aiming at the description mode of the input data, some researches directly take one-dimensional vibration signals as input, convolution or full-connection layers are adopted to extract signal characteristics, a learner firstly extracts time domain characteristics of the vibration signals, then takes characteristic sequences as input to avoid influence of noise in original signals on the characteristic extraction, some researches convert fault diagnosis into image recognition tasks, convert the one-dimensional vibration signals into two-dimensional matrixes according to time sequence relations, adopt two-dimensional convolution to extract signal characteristics, and some students acquire a time-frequency spectrogram of the vibration signals as input by adopting a time-frequency analysis method, so that fault information hidden in a time-frequency domain is extracted by means of the convolution layers. Aiming at the deep learning network parameters (2), the batch normalization and dropout modules are added and deleted by changing the sizes and the layer depths of a convolution layer and a full connection layer so as to obtain different feature extraction capacities. After the rotary machine fails, the waveform, frequency composition and the like of the vibration signal change correspondingly. For the two points mentioned above, most researches adopt the original signals or corresponding time domain, frequency domain and time-frequency domain information as network input, and combine a convolution layer and a full connection layer to realize feature extraction and classification. However, the above researches neglect the extraction of the relevant information features of the frequency domain scale in the signals (for example, the frequency composition changes and the relevant information between different frequency scales also changes when the fault occurs), so that the further improvement of the feature extraction and fault diagnosis performance is limited. Therefore, in order to obtain a richer feature extraction capability, some scholars try to further improve the accuracy and robustness of fault diagnosis by modeling through relevant information.
As a common tool for modeling related information, graph theory is widely applied to fields of natural language processing, image processing and the like in recent years, and the powerful data characterization capability of the graph theory improves the performance of deep learning in tasks in different fields. In recent years, with the advent of graph roll-up networks, deep learning frameworks based on graph theory have also become applied to the field of mechanical fault diagnosis. However, the current fault diagnosis method based on the graph rolling network still has two major problems of (1) most researches are to perform data characterization by constructing interactive graphs of different sample fragments and different sensing signals, but neglecting the extraction of related information among multiple frequency scales of the signals, so that the loss of the related information of the fault is possibly caused, and (2) the robustness of the conventional graph pooling method needs to be enhanced, and a large rise space is provided in the field of fault diagnosis.
Disclosure of Invention
In order to solve the technical problems in the background art, the invention provides a fault diagnosis method and a fault diagnosis system combining multi-scale interaction graph convolution and contrast pooling, which extract fault characteristics hidden in multi-frequency scale related information through a graph convolution network and a constructed multi-scale interaction graph; and (3) through the proposed contrast learning reinforced self-attention pooling layer, layering pooling is carried out on the graph data, and effective fault characterization information is extracted while feature dimension reduction is carried out. Through the operation, the accuracy and the robustness of the graph rolling network applied to the fault diagnosis of the rotating machinery are improved.
In order to achieve the above purpose, the present invention adopts the following technical scheme:
The first aspect of the invention provides a fault diagnosis method combining multi-scale interaction graph convolution and contrast pooling.
The fault diagnosis method combining multi-scale interaction graph convolution and contrast pooling comprises the following steps:
Obtaining a vibration signal of the rolling bearing, and obtaining a fault type by adopting a fault diagnosis model;
the fault diagnosis model comprises a multi-scale interaction graph, a graph convolution network and a contrast learning reinforced self-attention pooling layer;
The fault diagnosis model comprises the steps of calculating node embedding vectors and adjacent matrixes based on vibration signals of a rolling bearing, constructing a multi-scale interaction graph, extracting graph data features by adopting a graph rolling layer of a graph rolling network based on the node embedding vectors and the adjacent matrixes, performing contrast learning to strengthen a self-attention pooling layer, performing graph structure coarsening and feature dimension reduction on the graph data features, and sequentially enabling the coarsened graph data to pass through a graph rolling network reading layer and a full-connection layer to obtain fault types.
Further, the process of using the multi-scale interaction map includes:
Calculating a vibration signal time-frequency spectrogram by adopting continuous wavelet transformation, and normalizing spectrogram values to be within a range of [0,1 ];
Determining the number N of nodes of the multi-scale interaction graph, and equally dividing the obtained time-frequency spectrogram into N sections along the frequency direction to obtain N sections of banded spectrograms;
flattening the obtained banded spectrogram into a one-dimensional vector to become an embedded feature of a node corresponding to the multi-scale interaction graph;
Setting a threshold value, calculating cosine similarity of embedded features of any two nodes, and defining the two nodes as neighbor nodes when the cosine similarity is larger than the threshold value, otherwise, obtaining an adjacent matrix of the nodes without connection relation between the two nodes;
and obtaining an edge index matrix according to the obtained adjacent matrix.
Further, the graph rolling network comprises three graph convolution layers and three readout layers, the output of each graph rolling layer is connected with a contrast learning enhancement self-attention pooling layer, the output of each contrast learning enhancement self-attention pooling layer is connected with one readout layer, and the outputs of the three readout layers are connected with a full connection layer together.
Further, the training of the fault diagnosis model comprises the training of a contrast learning enhanced self-attention pooling layer:
the contrast learning reinforced self-attention pooling layer comprises parallel GCN+ and GCN-, positive correlation is carried out on the obtained positive sample attention score and fault classification precision by applying classification constraint to the GCN+, the GCN-is not constrained, and the obtained negative sample attention score is irrelevant to node contribution degree;
The method comprises the steps of taking a difference between a positive sample attention score and a negative sample attention score as a new positive sample attention score, wherein the negative sample attention score is unchanged, screening and retaining k nodes with the maximum attention score through Top-k criterion and masking operation, discarding the rest nodes to respectively obtain a positive pooling graph and a negative pooling graph, and obtaining a coarsening graph-positive pooling graph which can be most characterized by an original graph by carrying out network training multiple iterations to maximize the difference between the positive sample attention score and the negative sample attention score and obtain the positive sample attention score with higher stability.
Further, the Top-k criterion is that the nodes corresponding to the maximum k attention scores in the attention score sequence of the nodes are reserved, and the rest nodes are discarded.
Further, the fault diagnosis model optimizes parameters of the fault diagnosis model by adopting a classification loss function and a comparison loss function in the training process.
Still further, the classification loss function is:
the contrast loss function is:
Wherein N s is the number of training samples, C is the number of fault categories, p j represents the true category probability of the jth sample, q j represents the predicted probability value of the jth sample belonging to the C category, and l 1,l2,l3 represents the layer indexes corresponding to the three pooling layers respectively; a positive sample attention score representing the i-th sample of the positive pooling network gcn+ output of the l+1 th layer; The negative sample attention score of the i-th sample representing the negative pooling network GCN-output of the l+1-th layer, q +,i representing the class probability of the i-th sample predicted by the method based on the positive pooling graph, q -,i representing the class probability of the i-th sample predicted by the method based on the negative pooling graph.
A second aspect of the invention provides a fault diagnosis system combining multi-scale interactive graph convolution and contrast pooling.
A fault diagnosis system combining multiscale interactive graph convolution and contrast pooling, comprising:
The fault diagnosis module is configured to acquire a vibration signal of the rolling bearing, and acquire a fault type by adopting a fault diagnosis model;
The fault diagnosis module comprises a multi-scale interaction graph, a graph rolling network and a contrast learning strengthening self-attention pooling layer, wherein the fault diagnosis module comprises a node embedding vector and an adjacent matrix which are calculated based on vibration signals of a rolling bearing to construct the multi-scale interaction graph, the graph rolling layer of the graph rolling network is adopted to extract graph data characteristics based on the node embedding vector and the adjacent matrix, the contrast learning strengthening self-attention pooling layer is used for carrying out graph structure coarsening and characteristic dimension reduction on the graph data characteristics, and the coarsened graph data sequentially passes through the graph rolling network reading layer and the full-connection layer to obtain fault types.
A third aspect of the present invention provides a computer-readable storage medium.
A computer readable storage medium having stored thereon a computer program which when executed by a processor performs the steps in a joint multiscale interactive graph convolution and contrast pooling fault diagnosis method as described in the first aspect above.
A fourth aspect of the invention provides a computer device.
A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the steps in the joint multiscale interactive graph convolution and contrast pooling fault diagnosis method according to the first aspect described above when the program is executed.
Compared with the prior art, the invention has the beneficial effects that:
The invention provides a fault diagnosis framework based on a graph convolution neural network, which adopts a three-layer graph convolution layer to extract graph data characteristics, sets a three-layer pooling layer to coarsen a graph structure and reduce dimension of the convolutionally extracted characteristics, uses three-layer reading layers to read pooled graph data as the input of a full-connection layer, and finally outputs data fault types by means of the full-connection layer to complete fault diagnosis tasks.
The multi-scale interaction graph provided by the invention creatively divides a time spectrum into banded spectrums, further converts the banded spectrums into graph data, and constructs a data model for representing multi-frequency scale related information. Meanwhile, the multi-scale interaction graph can be well embedded into a graph convolution neural network, the advantages of deep learning in terms of big data processing are fully exerted, and fault diagnosis knowledge hidden in multi-frequency scale related information is mined.
The contrast learning reinforcement image pooling layer provided by the invention creatively applies a contrast learning framework to image pooling tasks, converts image pooling operation into a two-class approximation task by means of natural contradictory attributes of positive pooled images and negative pooled images when describing original image data, applies classification constraint to the positive pooled images, and simultaneously continuously enlarges the difference between the positive pooled images and the negative pooled images, thereby obtaining image pooling results which have higher robustness and are positively correlated with correct classification, and effectively improving the accuracy and stability of fault diagnosis.
Drawings
The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the invention.
FIG. 1 is a schematic diagram of a fault diagnosis model framework shown in the present invention;
FIG. 2 is a multi-scale interaction graph modeling schematic illustrating the present invention;
FIG. 3 is a schematic diagram of the connection relationship between the adjacency matrix, the edge index matrix and the nodes shown in the present invention;
FIG. 4 is a schematic diagram of the operation of the adjacency matrix mask shown in the present invention;
Fig. 5 is a schematic diagram of a node embedded vector matrix mask in accordance with the present invention.
Detailed Description
The invention will be further described with reference to the drawings and examples.
It should be noted that the following detailed description is illustrative and is intended to provide further explanation of the invention. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.
It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of exemplary embodiments according to the present invention. As used herein, the singular is also intended to include the plural unless the context clearly indicates otherwise, and furthermore, it is to be understood that the terms "comprises" and/or "comprising" when used in this specification are taken to specify the presence of stated features, steps, operations, devices, components, and/or combinations thereof.
It is noted that the flowcharts and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of methods and systems according to various embodiments of the present disclosure. It should be noted that each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the logical functions specified in the various embodiments. It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the flowchart illustrations and/or block diagrams, and combinations of blocks in the flowchart illustrations and/or block diagrams, can be implemented by special purpose hardware-based systems which perform the specified functions or operations, or combinations of special purpose hardware and computer instructions.
Example 1
The embodiment provides a fault diagnosis method combining multi-scale interaction graph rolling and contrast pooling, which is applied to a server for illustration, and it can be understood that the method can also be applied to a terminal, can also be applied to a system and a terminal, and can be realized through interaction of the terminal and the server. The server can be an independent physical server, a server cluster or a distributed system formed by a plurality of physical servers, and can also be a cloud server for providing cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network servers, cloud communication, middleware services, domain name services, security services CDNs, basic cloud computing services such as big data and artificial intelligent platforms and the like. The terminal may be, but is not limited to, a smart phone, a tablet computer, a notebook computer, a desktop computer, a smart speaker, a smart watch, etc. The terminal and the server may be directly or indirectly connected through wired or wireless communication, and the present application is not limited herein. In this embodiment, the method includes the steps of:
Obtaining a vibration signal of the rolling bearing, and obtaining a fault type by adopting a fault diagnosis model;
the fault diagnosis model comprises a multi-scale interaction graph, a graph convolution network and a contrast learning reinforced self-attention pooling layer;
The fault diagnosis model comprises the steps of calculating node embedding vectors and adjacent matrixes based on vibration signals of a rolling bearing, constructing a multi-scale interaction graph, extracting graph data features by adopting a graph rolling layer of a graph rolling network based on the node embedding vectors and the adjacent matrixes, performing contrast learning to strengthen a self-attention pooling layer, performing graph structure coarsening and feature dimension reduction on the graph data features, and sequentially enabling the coarsened graph data to pass through a graph rolling network reading layer and a full-connection layer to obtain fault types.
The present embodiments will now be described more fully hereinafter with reference to the accompanying drawings:
the fault diagnosis model provided by the embodiment comprises (1) a multi-scale interaction graph, (2) a graph rolling network and (3) a contrast learning reinforcement pooling layer. Fig. 1 shows the overall network architecture of the present embodiment.
1. Multi-scale interaction graph
FIG. 2 is a schematic diagram of a multi-scale interaction graph modeling. The multi-scale interaction graph modeling mainly comprises the following steps:
(1) Calculating a vibration signal time-frequency spectrogram by adopting Continuous Wavelet Transform (CWT), and normalizing spectrogram values to be within a range of [0,1 ];
(2) Determining the number N of nodes of the multi-scale interaction graph, and equally dividing the obtained time-frequency spectrogram into N sections along the frequency direction to obtain N sections of banded spectrograms;
(3) Flattening the obtained banded spectrogram into a one-dimensional vector to become an embedded characteristic of a corresponding node;
(4) Setting a threshold epsilon, calculating cosine similarity of embedded features of any two nodes, and defining that the two nodes are neighbor nodes when the cosine similarity is larger than epsilon, otherwise, the two nodes have no connection relationship. Obtaining an adjacent matrix of the node according to the processing;
(5) And obtaining an edge index matrix according to the obtained adjacent matrix.
The formula involved in step (4) is as follows:
Wherein, the Representing the neighbor nodes of the node v i, x i,xj is the embedded vector of the node v i and the node v j, epsilon-radius (v i) is the neighbor search function, the return value { v j,j∈N:cos(xi,xj) > epsilon } of which is the neighbor node set of the node v i, and cos (x i,xj) represents the cosine similarity of x i,xj.
Because the built multi-scale interaction graph is an undirected and unauthorized graph, the modeling only needs to consider node embedded vectors and edge index matrixes, and therefore, the description and the construction of the graph are realized through the steps. In order to illustrate the node connection relationship in the graph, the correspondence relationship between the adjacent matrix and the edge index matrix is illustrated in fig. 3.
2. Graph rolling network
The network frame of the fault diagnosis method provided by the embodiment comprises three graph roll layers (GCN 1, GCN2 and GCN 3) which are used for extracting features from the built multi-scale interactive graph, each graph roll layer is subjected to graph structure coarsening and feature dimension reduction through a pooling layer (the pooling layer is a contrast learning enhancement graph pooling layer provided by the invention, the specific function of the module is described in detail in the next section), then, the coarsened graph is next to a Readout layer (Readout 1, readout2 and Readout 3) which converts the graph topological structure and node embedded vector into data which can be directly processed by a full-connection layer (MLP), and the full-connection layer (MLP) receives the data read by the three Readout layers and performs fault classification on the data to complete the fault diagnosis task. The formula involved in the readout layer processing data is as follows:
wherein, r i represents the characteristics read by the ith reading layer, N represents the number of nodes of the graph data to be processed by the reading layer; Representing the embedded vector of the nth node of the graph data after being processed by the ith convolution layer, max representing the maximum value operation, CONCAT representing the spliced row vector operation. As can be seen from equation (2), the readout layer takes the average value and the maximum value of all node embedded vectors of the graph data to be processed as the readout characteristics to be input into the full connection layer.
The full connection layer in the network structure is composed of three layers, and finally, the full connection layer is output after being activated by a softmax function, the number of output nodes is the number of fault types, and the output of each node represents the probability value that the current data is the fault.
The calculation formula when the three convolutional layers participate in forward propagation of the network is as follows:
H0＝ReLU([A0XW0]) (3)
H1＝ReLU([CSPool(A0,H0)·W1]) (4)
H2＝ReLU([CSPool(A1,H1)·W2]) (5)
The method comprises the steps of H 0,H1,H2, reLU, CSPool, W 0,W1,W2, X, A 0,H0, A 1,H1 and A 1,H1, wherein H 0,H1,H2 is a node embedded vector matrix after forward propagation, reLU is a nonlinear activation unit, the processing flow of the module is specifically given in the next section, W 0,W1,W2 represents a1 st, 2 nd and 3 rd layer convolution kernel, X represents a node embedded vector matrix before pooling, A 0,H0 represents an adjacent matrix and a node embedded vector matrix of original graph data, and A 1,H1 represents an adjacent matrix and a node embedded vector matrix of graph data after a first layer pooling layer and a second layer convolution layer.
3. Contrast learning reinforcement pooling layer
As shown in the contrast learning enhanced self-attention pooling of fig. 1, consists essentially of a graph roll-up network (GCN +、GCN-) comprising two parallel lines. The module aims to obtain two groups of node attention scores, namely a positive sample attention score and a negative sample attention score, by means of GCN +、GCN-, apply classification constraint to GCN +, enable positive sample attention score to be positively correlated with fault classification accuracy (the node embedding vector pair correctly performs fault classification contribution degree is higher, the node attention score is higher), meanwhile, the GCN - is not constrained, the obtained negative sample attention score is irrelevant to the node contribution degree, then, the difference between the positive sample attention score and the negative sample attention score serves as a new positive sample attention score, the negative sample attention score is unchanged, the positive sample attention score and the negative sample attention score are subjected to Top-k criterion and masking operation, k nodes with the largest attention score are screened and reserved, the rest nodes are discarded, a positive pooling graph and a negative pooling graph are respectively obtained, the difference between the positive sample attention score and the negative sample attention score is maximized through network training multiple rounds of iteration, the positive sample attention score and the negative sample attention score are obtained, and the original map representing the maximum dimension-roughened graph is obtained when the map embedding vector is reduced.
The Top-k criterion in the pooling process is that the nodes corresponding to the maximum k attention scores in the attention score sequence of the nodes are reserved, and the rest nodes are discarded.
The mask operation in the pooling process aims at the object being an adjacency matrix and a node embedding vector matrix of the graph data, as shown in fig. 4 and 5, and taking the graph data comprising five nodes as an example, it is assumed that after Top-k screening, nodes 1, 3 and 5 are the first three nodes with the largest attention score, and the three nodes are reserved and discarded for nodes 2 and 4. Since the graph data is the adjacency matrix and the node embedded vector matrix to participate in the operation when participating in the forward transfer of the neural network, the rejected node does not participate in the operation, so the masking operation effectively covers the second row, the fourth row, the second column and the fourth column corresponding to the nodes 2 and 4 in the adjacency matrix and the second row and the fourth row in the node embedded vector matrix, so that the rejected node does not participate in the subsequent operation.
In the contrast learning reinforced self-attention pooling layer, the calculation formulas of the positive and negative attention scores are as follows:
Wherein, the Representing positive and negative attention scores of layer 1 and sigma representing sigmoid functions, and H l,Al representing node embedding vector matrix and adjacency matrix of layer 1 graph data, respectively.
The node screening formula based on the Top-k criterion is as follows:
Wherein, the Respectively representing node index values corresponding to the maximum k attention scores obtained through Topk-k screening, wherein N represents the number of nodes of the original graph data, and ratio pool represents the pooling proportion.
The formula for the masking operation is as follows:
Wherein the method comprises the steps of Embedding a vector matrix for the adjacency matrix and the nodes of the calculated layer 1 positive pooling graph; and embedding a vector matrix for the adjacency matrix and the nodes of the calculated layer 1+1 negative pooling graph.
4. Loss function
Based on the above modules, the invention contains two kinds of loss functions, namely classification loss L cls and comparison loss L cor. The calculation formula of the two loss functions is as follows:
Wherein N s is the number of training samples, C is the number of fault categories, p j represents the true category probability of the jth sample, q j represents the probability value of the jth sample predicted by the invention belonging to the C-th category, and l 1,l2,l3 represents the layer indexes corresponding to the three pooling layers respectively; A positive sample attention score representing the i-th sample output by the layer l+1 positive pooling network GCN +; Representing the negative sample attention score of the ith sample output by the negative pooling network GCN - of the l+1 layer, q +,i representing the class probability of the ith sample predicted by the method based on the positive pooling graph, and q -,i representing the class probability of the ith sample predicted by the method based on the negative pooling graph. Specifically, based on the above loss function, each parameter is updated according to the following formula.
The training procedure of the method according to the present embodiment is as follows.
Based on the training process, network parameters except theta pool- learn fault classification knowledge through classification loss L cls, meanwhile, the loss function can restrict positive sample attention scores output by GCN + to be positively correlated with correct fault classification, and through comparison loss L cor, similarity between prediction results output based on negative pooling graphs and prediction results output based on positive pooling graphs is reduced, and because positive pooling graphs are obtained based on positive sample attention scores which are obtained by differentiating original positive sample attention scores and negative sample attention scores (see formula (6)), the difference between the positive and negative pooling output attention scores is continuously increased based on comparison loss and formula (6), so that the output positive sample attention scores are more stable and are positively correlated with correct fault classification.
Example two
The embodiment provides a fault diagnosis system combining multi-scale interaction graph convolution and contrast pooling.
A fault diagnosis system combining multiscale interactive graph convolution and contrast pooling, comprising:
The fault diagnosis module is configured to acquire a vibration signal of the rolling bearing, and acquire a fault type by adopting a fault diagnosis model;
The fault diagnosis module comprises a multi-scale interaction graph, a graph rolling network and a contrast learning strengthening self-attention pooling layer, wherein the fault diagnosis module comprises a node embedding vector and an adjacent matrix which are calculated based on vibration signals of a rolling bearing to construct the multi-scale interaction graph, the graph rolling layer of the graph rolling network is adopted to extract graph data characteristics based on the node embedding vector and the adjacent matrix, the contrast learning strengthening self-attention pooling layer is used for carrying out graph structure coarsening and characteristic dimension reduction on the graph data characteristics, and the coarsened graph data sequentially passes through the graph rolling network reading layer and the full-connection layer to obtain fault types.
It should be noted that the fault diagnosis module, the model building and processing module are the same as the examples and the application scenarios implemented by the steps in the first embodiment, but are not limited to the disclosure of the first embodiment. It should be noted that the modules described above may be implemented as part of a system in a computer system, such as a set of computer-executable instructions.
Example III
The present embodiment provides a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, implements the steps in the joint multiscale interactive graph convolution and contrast pooling fault diagnosis method described in the above embodiment.
Example IV
The present embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, where the processor implements the steps in the method for fault diagnosis in combination with multi-scale interactive graph convolution and contrast pooling as described in the above embodiment.
It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of a hardware embodiment, a software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, magnetic disk storage, optical storage, and the like) having computer-usable program code embodied therein.
The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
Those skilled in the art will appreciate that implementing all or part of the above-described methods in accordance with the embodiments may be accomplished by way of a computer program stored on a computer readable storage medium, which when executed may comprise the steps of the embodiments of the methods described above. The storage medium may be a magnetic disk, an optical disc, a Read-Only Memory (ROM), a Random access Memory (Random AccessMemory, RAM), or the like.
The above description is only of the preferred embodiments of the present invention and is not intended to limit the present invention, but various modifications and variations can be made to the present invention by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (10)
1. The fault diagnosis method combining multi-scale interaction graph convolution and contrast pooling is characterized by comprising the following steps of:
Obtaining a vibration signal of the rolling bearing, and obtaining a fault type by adopting a fault diagnosis model;
the fault diagnosis model comprises a multi-scale interaction graph, a graph convolution network and a contrast learning reinforced self-attention pooling layer;
The fault diagnosis model comprises the steps of calculating node embedding vectors and adjacent matrixes based on vibration signals of a rolling bearing, constructing a multi-scale interaction graph, extracting graph data features by adopting a graph rolling layer of a graph rolling network based on the node embedding vectors and the adjacent matrixes, performing contrast learning to strengthen a self-attention pooling layer, performing graph structure coarsening and feature dimension reduction on the graph data features, and sequentially enabling the coarsened graph data to pass through a graph rolling network reading layer and a full-connection layer to obtain fault types.
2. The method for fault diagnosis in combination with multi-scale interactive graph convolution and contrast pooling according to claim 1, wherein the process of constructing the multi-scale interactive graph comprises:
Calculating a vibration signal time-frequency spectrogram by adopting continuous wavelet transformation, and normalizing spectrogram values to be within a range of [0,1 ];
Determining the number N of nodes of the multi-scale interaction graph, and equally dividing the obtained time-frequency spectrogram into N sections along the frequency direction to obtain N sections of banded spectrograms;
flattening the obtained banded spectrogram into a one-dimensional vector to become an embedded feature of a node corresponding to the multi-scale interaction graph;
Setting a threshold value, calculating cosine similarity of embedded features of any two nodes, and defining the two nodes as neighbor nodes when the cosine similarity is larger than the threshold value, otherwise, obtaining an adjacent matrix of the nodes without connection relation between the two nodes;
and obtaining an edge index matrix according to the obtained adjacent matrix.
3. The method of claim 1, wherein the graph rolling network comprises three graph convolution layers and three readout layers, the output of each graph rolling layer is connected with one contrast learning enhanced self-attention pooling layer, the output of each contrast learning enhanced self-attention pooling layer is connected with one readout layer, and the outputs of the three readout layers are commonly connected with one fully connected layer.
4. The method of combined multiscale interactive graph convolution and contrast pooling fault diagnosis of claim 1, wherein the training of the fault diagnosis model comprises training of a contrast learning enhanced self-attention pooling layer:
the contrast learning reinforced self-attention pooling layer comprises parallel GCN+ and GCN-, positive correlation is carried out on the obtained positive sample attention score and fault classification precision by applying classification constraint to the GCN+, the GCN-is not constrained, and the obtained negative sample attention score is irrelevant to node contribution degree;
The method comprises the steps of taking a difference between a positive sample attention score and a negative sample attention score as a new positive sample attention score, wherein the negative sample attention score is unchanged, screening and retaining k nodes with the maximum attention score through Top-k criterion and masking operation, discarding the rest nodes to respectively obtain a positive pooling graph and a negative pooling graph, and obtaining a coarsening graph-positive pooling graph which can be most characterized by an original graph by carrying out network training multiple iterations to maximize the difference between the positive sample attention score and the negative sample attention score and obtain the positive sample attention score with higher stability.
5. The method for fault diagnosis combined with multi-scale interactive graph convolution and contrast pooling as claimed in claim 4, wherein the Top-k criterion is to reserve the nodes corresponding to the largest k attention scores in the attention score sequence of the nodes, and discard the rest nodes.
6. The combined multiscale interactive graph convolution and contrast pooling fault diagnosis method of claim 1, wherein the fault diagnosis model optimizes parameters of the fault diagnosis model using classification loss functions and contrast loss functions during training.
7. The method for fault diagnosis in combination with multi-scale interactive graph convolution and contrast pooling according to claim 6, wherein the classification loss function is:
the contrast loss function is:
Wherein, the The number of training samples; The number of fault categories; Represents the first True class probability of the individual samples; Representing predicted first The samples belong toProbability values for the individual categories; respectively representing layer indexes corresponding to the three pooling layers; Represents the first Layer of the forward pooled network GCN+ outputPositive sample attention scores for the individual samples; Represents the first Layer negative pooling network GCN-outputNegative sample attention scores for the individual samples; Representing the first of the method predictions based on the forward pooling graph Class probability of individual samples; representing the first of the method predictions based on the negative pooling graph Class probability of individual samples.
8. The fault diagnosis system combining multi-scale interaction graph convolution and contrast pooling is characterized by comprising:
The fault diagnosis module is configured to acquire a vibration signal of the rolling bearing, and acquire a fault type by adopting a fault diagnosis model;
The fault diagnosis module comprises a multi-scale interaction graph, a graph rolling network and a contrast learning strengthening self-attention pooling layer, wherein the fault diagnosis module comprises a node embedding vector and an adjacent matrix which are calculated based on vibration signals of a rolling bearing to construct the multi-scale interaction graph, the graph rolling layer of the graph rolling network is adopted to extract graph data characteristics based on the node embedding vector and the adjacent matrix, the contrast learning strengthening self-attention pooling layer is used for carrying out graph structure coarsening and characteristic dimension reduction on the graph data characteristics, and the coarsened graph data sequentially passes through the graph rolling network reading layer and the full-connection layer to obtain fault types.
9. A computer readable storage medium having stored thereon a computer program, which when executed by a processor implements the steps of the joint multiscale interactive graph convolution and contrast pooling fault diagnosis method according to any of claims 1-7.
10. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the steps in the combined multiscale interactive graph convolution and contrast pooling fault diagnosis method of any one of claims 1-7 when the program is executed by the processor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211606855.XA CN116070131B (en) | 2022-12-14 | 2022-12-14 | Fault diagnosis method and system combining multi-scale interactive graph convolution and contrast pooling |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211606855.XA CN116070131B (en) | 2022-12-14 | 2022-12-14 | Fault diagnosis method and system combining multi-scale interactive graph convolution and contrast pooling |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116070131A CN116070131A (en) | 2023-05-05 |
| CN116070131B true CN116070131B (en) | 2025-07-15 |
Family
ID=86181207
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211606855.XA Active CN116070131B (en) | 2022-12-14 | 2022-12-14 | Fault diagnosis method and system combining multi-scale interactive graph convolution and contrast pooling |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116070131B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116720074B (en) * | 2023-06-16 | 2025-08-22 | 安徽大学 | Fault diagnosis method and system based on wavelet graph convolution |
| CN117454145B (en) * | 2023-10-23 | 2025-02-07 | 四川大学 | A method for aircraft engine fault diagnosis based on multimodal sensors |
| CN118981684B (en) * | 2024-08-15 | 2025-04-15 | 北京建筑大学 | A method and system for out-of-distribution fault detection based on energy propagation and graph learning |
| CN118862950B (en) * | 2024-09-29 | 2025-03-07 | 烟台大学 | Industrial equipment fault diagnosis method and system based on graph structure joint optimization |
| CN120744461B (en) * | 2025-08-20 | 2026-01-09 | 浪潮通用软件有限公司 | A device fault diagnosis method based on multi-scale graph convolution |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US12175666B2 (en) * | 2021-02-01 | 2024-12-24 | PathAI, Inc. | Systems and methods for classifying biomedical image data using a graph neural network |
| CN114860246A (en) * | 2022-03-29 | 2022-08-05 | 湖北大学 | A method for predicting class interactions in software systems based on graph convolutional neural networks |
| CN115082142B (en) * | 2022-05-10 | 2024-04-30 | 华南理工大学 | A recommendation method, device and medium based on heterogeneous relationship graph neural network |
-
2022
- 2022-12-14 CN CN202211606855.XA patent/CN116070131B/en active Active
Non-Patent Citations (1)
| Title |
|---|
| A multi-scale graph convolutional network with contrastive-learning enhanced self-attention pooling for intelligent fault diagnosis of gearbox;Zixu Chen 等;《Measurement》;20240315;1-14 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116070131A (en) | 2023-05-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN116070131B (en) | Fault diagnosis method and system combining multi-scale interactive graph convolution and contrast pooling | |
| CN112991354B (en) | A semantic segmentation method for high-resolution remote sensing images based on deep learning | |
| CN112084331B (en) | Text processing, model training method, device, computer equipment and storage medium | |
| CN112734775B (en) | Image labeling, image semantic segmentation and model training methods and devices | |
| CN111723732B (en) | Optical remote sensing image change detection method, storage medium and computing equipment | |
| CN113779581B (en) | Robust detection method and system for lightweight high-precision malicious software identification model | |
| CN111160533A (en) | Neural network acceleration method based on cross-resolution knowledge distillation | |
| US20210182666A1 (en) | Weight data storage method and neural network processor based on the method | |
| CN107945210B (en) | Target tracking method based on deep learning and environment self-adaption | |
| CN111429977B (en) | Novel molecular similarity search algorithm based on attention of graph structure | |
| CN114913379A (en) | Remote sensing image small sample scene classification method based on multi-task dynamic contrast learning | |
| CN105469376A (en) | Method and device for determining picture similarity | |
| CN116362325B (en) | A Lightweight Application Method for Power Image Recognition Model Based on Model Compression | |
| CN114841307A (en) | Training method for binaryzation target detection neural network structure and model | |
| CN111783688B (en) | A classification method of remote sensing image scene based on convolutional neural network | |
| CN113435321A (en) | Method, system and equipment for evaluating state of main shaft bearing and readable storage medium | |
| CN114996495A (en) | Single-sample image segmentation method and device based on multiple prototypes and iterative enhancement | |
| CN111694977A (en) | Vehicle image retrieval method based on data enhancement | |
| CN117036368A (en) | Image data processing method, device, computer equipment and storage medium | |
| CN119167411A (en) | Countermeasures and defense methods for multi-modal large models | |
| CN112307048B (en) | Semantic matching model training method, matching method, device, equipment and storage medium | |
| Shen et al. | Efficient deep structure learning for resource-limited IoT devices | |
| CN119719675B (en) | A multimodal social relationship extraction method based on hypergraph attention neural network | |
| CN114764919B (en) | A deep learning-based method for pedestrian attribute recognition | |
| CN117173790B (en) | Gait recognition method and device based on causal intervention mechanism |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |