Rotating machine fault diagnosis method and system

Abstract

The invention discloses a rotating machine fault diagnosis method and system, and belongs to the technical field of fault diagnosis. The method comprises the following steps: acquiring a vibration signal of the rotating machine; inputting the vibration signal of the rotating machine into a pre-constructed fault diagnosis model, and outputting a fault diagnosis result of the rotating machine; wherein the fault diagnosis model comprises a metagraph construction layer, a hierarchical attention aggregation layer and a classification layer which are connected in sequence, and in the metagraph construction layer, a signal metagraph is constructed according to a vibration signal of the rotating machine; in the hierarchical attention aggregation layer, performing dual-channel signal feature extraction on the signal element graph to obtain a node feature matrix of the signal element graph; performing multi-channel signal feature extraction on the node feature matrix of the signal element graph to obtain a multi-channel signal feature matrix; and in the classification layer, classifying the multi-channel signal features to obtain a fault diagnosis result of the rotating machinery. The method can achieve the time-space fusion of the multi-channel vibration signals, and is higher in classification accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of fault diagnosis, and in particular to a rotating machinery fault diagnosis method and system. Background Art

[0002] With the continuous development of intelligent diagnosis technology, shallow neural network models such as back propagation (BP) neural network and radial basis function network (RBFN) have achieved certain results in mechanical fault diagnosis due to their low computational overhead, strong interpretability and efficient modeling capabilities for small and medium-sized data. The development of deep neural network models such as convolutional neural network (CNN) and recurrent neural network (RNN) has further promoted the improvement of intelligent diagnosis technology.

[0003] The aforementioned intelligent diagnostic algorithms have some signal feature extraction capabilities, but they face a bottleneck in fusing multi-channel sensor signals. Complex mechanical failures are often caused by the interaction of multiple subsystems, and relying solely on single-channel vibration signals is difficult to fully characterize the system's degradation state. Therefore, there is an urgent need to achieve targeted enhancement of fault-sensitive features by integrating multi-channel vibration signals from complex equipment. Summary of the Invention

[0004] The purpose of the present invention is to overcome the deficiencies in the prior art and to provide a rotating machinery fault diagnosis method and system that can solve the problem of insufficient ability of a single-channel vibration signal to characterize fault characteristics when facing complex faults.

[0005] To achieve the above object, the present invention is implemented by adopting the following technical solutions:

[0006] In one aspect, the present invention provides a rotating machinery fault diagnosis method, comprising:

[0007] Obtain vibration signals of rotating machinery;

[0008] Inputting the vibration signal of the rotating machinery into a pre-built fault diagnosis model and outputting a fault diagnosis result of the rotating machinery;

[0009] The fault diagnosis model includes a meta-graph construction layer, a hierarchical attention aggregation layer, and a classification layer connected in sequence. The processing steps of the fault diagnosis model include:

[0010] In the metagraph construction layer, a signal isomorphic graph is constructed according to the vibration signal of the rotating machinery, and a signal metagraph is constructed according to the signal isomorphic graph;

[0011] In the hierarchical attention aggregation layer, dual-channel signal feature extraction is performed on the signal meta-graph to obtain a node feature matrix of the signal meta-graph; multi-channel signal feature extraction is performed on the node feature matrix of the signal meta-graph to obtain a multi-channel signal feature matrix;

[0012] In the classification layer, the multi-channel signal features are classified to obtain the fault diagnosis result of the rotating machinery.

[0013] Optionally, constructing a signal isomerism graph according to the vibration signal of the rotating machinery includes:

[0014] Extracting frequency domain features of the rotating machinery vibration signal;

[0015] The K-nearest neighbor algorithm is used to calculate the Euclidean distance of the amplitude at the frequency point corresponding to the frequency domain feature;

[0016] According to the Euclidean distance, the vibration signals of the first K rotating machines with the closest Euclidean distance are obtained;

[0017] The vibration signals of the first K rotating machines closest to each other are used as heterogeneous nodes to build edge connections and obtain a signal heterogeneous graph.

[0018] Optionally, the calculation of the Euclidean distance of the amplitude at the frequency point corresponding to the frequency domain feature includes:

[0019] ;

[0020] ;

[0021] in, express Channel frequency domain signal; express Channel frequency domain signal; express Channel time domain signal; Indicates frequency; express Channel time domain signal; express Channel and The Euclidean distance of the amplitude of the channel signal samples at the corresponding frequency points; express Channel No. Frequency domain characteristics of signal samples; express Channel No. Frequency domain characteristics of signal samples; Indicates the total number of signal samples; represents the real part; represents the imaginary part; Represents an imaginary unit.

[0022] Optionally, the adjacency matrix of the signal heterogeneous graph is expressed as:

[0023] ;

[0024] in, Represents the adjacency matrix of the signal heterogeneous graph; 、 Respectively represent the specific value of the mth row and nth column and the specific value of the nth row and mth column in the adjacency matrix; Indicates the threshold value; Indicates Channel and The Euclidean distance of the amplitude of the channel signal at the corresponding frequency point; Indicated as conforming The first K units in the condition are connected by edges; k represents a meta-graph node.

[0025] Optionally, constructing a signal metagraph according to the signal heterogeneous graph includes:

[0026] On the basis of the signal heterogeneous graph, constructing a meta-path instance that is closed from beginning to end;

[0027] Embed the preset meta-path into the heterogeneous node matrix of the signal heterogeneous graph to obtain the feature representation of each meta-path instance;

[0028] The feature representation of each meta-path instance is used as the edge feature to obtain the signal meta-graph.

[0029] Optionally, dual-channel signal feature fusion is performed on the signal metagraph to obtain a node feature matrix of the signal metagraph, including:

[0030] Input the input edge features of each meta-graph node in the signal meta-graph into the attention linear transformation layer, and output the attention weight of each input edge feature;

[0031] The attention weight of each incoming edge feature is processed by the softmax function to obtain the attention coefficient of each incoming edge feature;

[0032] Perform weighted fusion of each incoming edge feature and the attention coefficient of each incoming edge feature to obtain the fused meta-graph node feature;

[0033] According to the fused meta-graph node features, the node feature matrix of the signal meta-graph is obtained.

[0034] Optionally, obtaining the node feature matrix of the signal element graph includes:

[0035] ;

[0036] ;

[0037] ;

[0038] in, represents the attention coefficient of the lth input edge; represents the attention weight of the lth input edge; represents the lth incoming edge; represents the softmax activation function; Represents the features of the k-th meta-graph node; The node feature matrix representing the signal element graph; Represents feature dimension; Indicates the number of incoming edges of each metagraph node; The number of nodes representing the signal metagraph.

[0039] Optionally, performing multi-channel signal feature extraction on the node feature matrix of the signal element graph to obtain the multi-channel signal feature matrix includes:

[0040] Input the node feature matrix of the signal meta-graph into the attention linear transformation layer, and output the attention weight of the meta-graph node;

[0041] The attention weight of the meta-graph node is processed by the softmax function to obtain the attention coefficient of the meta-graph node;

[0042] The node feature matrix of the signal meta-graph and the attention coefficient of the meta-graph node are weightedly fused to obtain the multi-channel signal feature matrix of the meta-graph node.

[0043] Optionally, acquiring the multi-channel signal feature matrix includes:

[0044] ;

[0045] ;

[0046] in, represents the attention coefficient of the kth meta-graph node; The attention weight of the k-th meta-graph node; The node feature matrix representing the signal element graph; represents the softmax activation function; The number of nodes representing the signal metagraph; Represents the multi-channel signal feature matrix.

[0047] In another aspect, the present invention provides a rotating machinery fault diagnosis system, comprising:

[0048] The signal acquisition module is used to: acquire the vibration signal of the rotating machinery;

[0049] a fault diagnosis module, configured to: input the vibration signal of the rotating machinery into a pre-built fault diagnosis model and output a fault diagnosis result of the rotating machinery;

[0050] The fault diagnosis model includes a meta-graph construction layer, a hierarchical attention aggregation layer, and a classification layer connected in sequence. The processing steps of the fault diagnosis model include:

[0051] In the metagraph construction layer, a signal isomorphic graph is constructed according to the vibration signal of the rotating machinery, and a signal metagraph is constructed according to the signal isomorphic graph;

[0052] In the hierarchical attention aggregation layer, dual-channel signal feature extraction is performed on the signal meta-graph to obtain a node feature matrix of the signal meta-graph; multi-channel signal feature extraction is performed on the node feature matrix of the signal meta-graph to obtain a multi-channel signal feature matrix;

[0053] In the classification layer, the multi-channel signal features are classified to obtain the fault diagnosis result of the rotating machinery.

[0054] Compared with the prior art, the present invention has the following beneficial effects:

[0055] The present invention can realize the spatiotemporal fusion of multi-channel vibration signals. Compared with single-channel vibration signals as input, it has a stronger ability to extract the frequency domain characteristics of multi-channel signal faults, a higher classification accuracy, and a certain degree of noise resistance. It can be oriented to multi-channel vibration signals, based on meta-paths and fused multi-level attention mechanisms, and is more noise-resistant and interpretable. It can more effectively improve the fault diagnosis accuracy and generalization ability of the model in a strong interference environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 FIG. 1 is a flow chart of a rotating machinery fault diagnosis method according to an embodiment of the present invention;

[0057] Figure 2 FIG2 is a schematic diagram of a construction process of a signal isomerization graph according to an embodiment of the present invention;

[0058] Figure 3 FIG. 1 is a flow chart showing the construction of a signal element graph in an embodiment of the present invention;

[0059] Figure 4 FIG. 1 is a flow chart of dual-channel signal feature extraction in one embodiment of the present invention;

[0060] Figure 5 FIG. 1 is a flow chart of multi-channel signal feature extraction in one embodiment of the present invention;

[0061] Figure 6The figure shows the classification accuracy of single-channel vibration signals and multi-channel vibration signals as samples under different signal-to-noise ratios in one embodiment of the present invention;

[0062] Figure 7 Shown is the Hilbert envelope spectrum of fault features learned when single-channel vibration signals and multi-channel vibration signals are used as samples at different signal-to-noise ratios in one embodiment of the present invention;

[0063] Figure 8 FIG. 4 shows a comparison of classification performances under different signal-to-noise ratios in one embodiment between the present invention and the prior art. DETAILED DESCRIPTION

[0064] The technical solution of the present invention is described in detail below through the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations on the technical solution of the present invention. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.

[0065] The term "and / or" simply describes a relationship between related objects, indicating that three possible relationships exist. For example, "A and / or B" can mean: A exists alone, A and B exist simultaneously, or B exists alone. Additionally, the character " / " generally indicates an "or" relationship between the related objects.

[0066] Example 1

[0067] like Figure 1 As shown, this embodiment introduces a rotating machinery fault diagnosis method, including the following steps:

[0068] Step 1: Obtain the vibration signal of the rotating machinery.

[0069] Step 2: Input the vibration signal of the rotating machinery into the pre-built fault diagnosis model and output the fault diagnosis results of the rotating machinery, specifically:

[0070] This embodiment uses a heterogeneous graph neural network (MMA-GNN) model for subsequent processing. The MMA-GNN model includes a meta-graph construction layer, a hierarchical attention aggregation layer, and a classification layer connected in sequence. Its processing steps include:

[0071] In the metagraph construction layer, a signal heterogeneous graph is constructed based on the vibration signal of the rotating machinery. Specifically:

[0072] like Figure 2 The figure shows the construction process of the signal heterogeneity graph. The vibration signals at different positions of the mechanical equipment are collected by multi-channel sensors, such as Figure 1 As shown in the figure, the i-th signal sample collected by the u-th sensor is represented by ui. Signal samples collected by different sensors represent different channels. The collected vibration signal is truncated into signal samples of the same length to extract the frequency domain features of the rotating machinery vibration signal. Then, the KNN algorithm (K-Nearest Neighbors) is used to construct a heterogeneous graph. Specifically, the K-nearest neighbor algorithm is used to calculate the Euclidean distance of the amplitude at the frequency point corresponding to the frequency domain feature. The formula is:

[0073] ;

[0074] ;

[0075] in, express Channel frequency domain signal; express Channel frequency domain signal; express Channel time domain signal; Indicates frequency; express Channel time domain signal; express Channel and The Euclidean distance of the amplitude of the channel signal samples at the corresponding frequency points; express Channel No. Frequency domain characteristics of signal samples; express Channel No. Frequency domain characteristics of signal samples; Indicates the total number of signal samples; represents the real part; represents the imaginary part; represents an imaginary unit;

[0076] The first K vibration signals of rotating machinery with the closest Euclidean distance are selected according to the Euclidean distance, and edge connections are constructed between the first K signal samples of different channels, that is, a signal heterogeneous graph is obtained. The edges in the signal heterogeneous graph only represent the connection relationship between multi-channel vibration signals and do not have feature representation. The adjacency matrix of the signal heterogeneous graph is expressed as:

[0077] ;

[0078] in, Represents the adjacency matrix of the signal heterogeneous graph; 、 Respectively represent the specific value of the mth row and nth column and the specific value of the nth row and mth column in the adjacency matrix; Indicates the threshold value; express Channel and The Euclidean distance of the amplitude of the channel signal at the corresponding frequency point; Indicated as conforming The first K units in the condition are connected by edges; k represents a metagraph node;

[0079] This process aims to capture the similarities in the frequency domain characteristics of signal samples, thereby constructing a heterogeneous graph composed of multi-channel vibration signals. The intercepted multi-channel vibration signal samples are the heterogeneous nodes in the graph, and the original signal sample information is the node features of the heterogeneous graph. This heterogeneous graph construction method effectively preserves the diversity and complexity of multi-channel vibration signals.

[0080] Construct a signal metagraph based on the signal heterogeneous graph, specifically:

[0081] like Figure 3 The figure shows the signal metagraph construction process. Based on the preset metapaths, the signal heterogeneous graph is further converted into a signal metagraph. Multiple closed metapath instances are defined within the signal heterogeneous graph to represent semantic relationship paths between signal node samples from different channels. The preset metapaths are embedded into the heterogeneous graph node feature matrix of the signal heterogeneous graph, resulting in a feature representation for each metapath instance. The feature representation of each metapath instance is then used as an edge feature to construct the signal metagraph. The initially constructed signal metagraph is a multi-directed graph, with multiple directed edges connecting metagraph nodes. Edges are metapath instances, and edge features are the feature representations of metapath instances. Initially, metagraph nodes are featureless. When edge features are subsequently aggregated, they are assigned to the nodes. The metagraph node type is determined by the channel signal sample types defined at the beginning and end of the metapaths. The metagraph edge features incorporate vibration signals collected by two sensors, integrating cross-channel information and providing rich semantic information for subsequent feature aggregation and fault diagnosis.

[0082] In the hierarchical attention aggregation layer, dual-channel signal feature extraction is performed on the signal meta-graph to obtain the node feature matrix of the signal meta-graph, specifically:

[0083] Figure 4 shows the process for extracting dual-channel signal features. The attention mechanism within the signal metagraph aggregates all edge features pointing to a node in the metagraph and assigns the aggregated features to the metagraph node. This design integrates the characteristics of dual-channel sensor signals and, by adaptively focusing on the effective information conveyed by different edges, aims to improve the discriminative and representative capabilities of signal features, providing more discriminative feature inputs for subsequent pattern recognition, thereby effectively supporting improved classification accuracy.

[0084] Specifically, first for each target node in the signal metagraph , through a learnable attention mechanism for all its input edges Perform importance evaluation, input each input edge feature vector into the trainable attention linear transformation layer, and output the attention weight with dimension 1 ; Then the attention coefficient of each edge after normalization is obtained through the softmax function , ensuring that the sum of all weights is 1; finally, based on the calculated attention coefficient, all input edge features are weighted fused to finally generate the fused meta-graph node features containing multi-source information , according to the fused meta-graph node features, the feature matrix of each node of the signal meta-graph is obtained ,Right now:

[0085] ;

[0086] ;

[0087] ;

[0088] in, represents the attention coefficient of the lth input edge; represents the attention weight of the lth input edge; represents the lth incoming edge; represents the softmax activation function; Represents the features of the k-th meta-graph node; The node feature matrix representing the signal element graph; Represents feature dimension; Indicates the number of incoming edges of each metagraph node; The number of nodes representing the signal metagraph;

[0089] Perform multi-channel signal feature extraction on the dual-channel signal feature matrix to obtain a multi-channel signal feature matrix, specifically:

[0090] like Figure 5 The figure shows the multi-channel signal feature extraction process. The inter-metagraph attention mechanism aggregates the dual-channel vibration signal features from multiple metagraphs. This fusion module aims to effectively integrate multi-channel vibration signals, breaking through the limitations of dual-channel vibration signals. More importantly, the attention mechanism dynamically distinguishes the importance of different sensor channels, thereby improving the discriminative ability of downstream classification tasks.

[0091] Specifically, taking the constructed signal metagraphs Metagraph1 and Metagraph2 as input, their corresponding node feature matrices are: 、 , ( , k is the number of nodes in the signal element graph, is the node feature dimension). First, the two node feature matrices are projected through a trainable attention linear transformation layer, and the adaptive attention weights of each node feature matrix are calculated respectively. ; The attention weights are then normalized by the softmax function to generate the standardized attention coefficient , the attention coefficient can quantitatively characterize the contribution of each sensor channel signal feature in the target node feature fusion process; finally, the attention-aware weighted fusion strategy is adopted to perform matrix multiplication operation on the attention coefficient and the node feature matrix to generate a fusion feature matrix with multi-channel perception capability, namely the multi-channel signal feature matrix The multi-channel signal feature matrix That is, the graph node feature matrix finally used for classification.

[0092] In the classification layer, the multi-channel signal feature matrix is ​​classified to obtain the fault diagnosis results of the rotating machinery. Classification is performed through the classifier.

[0093] In this embodiment, it is possible to face multi-channel vibration signals, based on meta-paths, and integrate multi-level attention mechanisms, and to realize the construction of a basic platform for cross-channel feature interaction. At the data fusion level, the concept of meta-paths is introduced, and a meta-graph construction method is proposed to establish a topological association structure for multi-channel vibration signals; multi-channel vibration signal features are aggregated based on a hierarchical attention mechanism, and a two-layer attention framework is designed in the feature aggregation process, aiming to realize a feature optimization mechanism from local to global.

[0094] Example 2

[0095] Based on Example 1, this example introduces an experimental example of a rotating machinery fault diagnosis method:

[0096] To comprehensively evaluate the performance advantages of the MMA-GNN model in this embodiment in the field of fault diagnosis, a series of validation studies were conducted. First, they demonstrated that multi-channel vibration signal fusion offers superior classification performance and interpretability compared to single-channel vibration signals. Second, they compared the MMA-GNN model with five models: a graph convolutional network (GCN), a graph attention network (GAT), a graph sample and aggregated (graph SAGE), a CNN, and a long short-term memory (LSTM) model, based on two key dimensions: overall performance and noise immunity.

[0097] Analysis of the superiority of multi-channel vibration signal fusion. As shown in Figure 6, compared with single-channel vibration signal as input, multi-channel vibration signal fusion has higher classification accuracy. Complex rotating machinery fault signals contain more frequency components, and the Hilbert envelope spectrum has better frequency generalization ability. Therefore, this application uses the Hilbert envelope spectrum to analyze the interpretability of the MMT-GNN model. Figure 7 is the Hilbert envelope spectrum of the model learning fault features under different SNRs. The upper part is the feature map learned by the model when a single-channel vibration signal is used as input, and the lower part is the feature map learned by the model when a multi-channel vibration signal is used as input. The following three conclusions can be drawn from the figure:

[0098] ① It can accurately capture and deeply mine fault characteristics. The fault characteristic frequency in the figure is clearly shown as 107.36Hz. When the multi-channel vibration signal is input into the MMA-GNN model, the model demonstrates excellent learning ability. Compared with inputting a single-channel vibration signal, the multi-channel vibration signal enables the model to not only accurately capture this critical fault frequency, but also further identify and learn the harmonic components of this frequency. These harmonic components often contain richer fault information, which is of great significance for deeply understanding the nature of the fault and improving the accuracy of fault diagnosis. This shows that the fusion of multi-channel vibration signals provides the model with more comprehensive and richer information, allowing the model to more deeply mine fault characteristics.

[0099] ② Stronger signal screening and noise suppression capabilities. The MMA-GNN model demonstrates powerful signal screening capabilities when processing multi-channel vibration signals. It effectively blocks frequency components irrelevant to fault diagnosis, which may originate from normal mechanical operating noise, environmental interference, or other non-fault-related factors. Through this precise feature extraction and noise suppression, the MMA-GNN model ensures that the learned feature maps are highly focused on fault-related signals, significantly improving the specificity and discriminability of fault signatures.

[0100] ③ Enhanced feature interaction and aggregation. The MMA-GNN model's deep processing capabilities for multi-channel vibration signals fully demonstrate its strength in feature interaction and fault feature aggregation. Through complex internal mechanisms, the model enables effective information exchange and fusion between vibration signals from different sensors. This interaction not only enhances the correlation between individual signals but also further highlights and strengthens fault features through the combined effects of multi-channel vibration signals. However, due to the lack of this interaction, single-channel vibration signals may not have clear fault features, making effective aggregation and strengthening difficult.

[0101] Overall performance analysis of the model. The network structure parameters of the five-category model are shown in Table 1, and the classification performance under noise-free conditions is shown in Table 2.

[0102] Table 1 Network structure parameters of each model

[0103]

[0104] Table 2 Classification results of each model under noise-free conditions

[0105]

[0106] Under ideal, noise-free conditions, experimental results show that the MMA-GNN model achieves a perfect 100% classification accuracy across all metrics. The three graph neural network models—the GCN, GAT, and Graph SAGE—also perform well, achieving classification accuracy above 98%. However, the classification performance of two deep network models, the 1D CNN and LSTM models, is less than satisfactory and relatively poor. The experimental results strongly demonstrate that, under noise-free conditions, inputting multi-channel vibration signals into graph neural network models can achieve exceptional classification performance. In contrast, the 1D CNN and LSTM models are significantly less capable of integrating multi-channel signal features. The unique topological structure of neural networks gives them a significant inherent advantage over CNN and LSTM models in integrating multi-channel vibration signals. This advantage enables graph neural network models to achieve more accurate and efficient feature extraction, a key step.

[0107] Noise immunity analysis. As shown in Figure 8, we simulated a noise signal by adding additive white Gaussian noise proportional to the clean signal. We then selected fault signals at six signal-to-noise ratios (SNRs) of 6dB, 4dB, 2dB, -2dB, -4dB, and -6dB to evaluate the noise immunity of the five deep models.

[0108] Experimental results show that the accuracy of the GCN, GAT, and Graph SAGE models is lower than that of the MMA-GNN model under different SNR conditions. Compared with other graph neural network models, the MMA-GNN model demonstrates superior performance in processing multi-channel vibration signals thanks to its unique meta-graph construction strategy. The introduction of the meta-graph enables the model to accurately capture and fuse key information from multi-channel vibration signals, significantly improving the interaction efficiency between signals even in complex environments with strong noise interference. In addition, the attention mechanism embedded in the MMA-GNN model can automatically identify and enhance fault features in vibration signals while effectively suppressing the interference of noise signals. This mechanism not only gives the model strong noise resistance but also greatly improves its performance in classification tasks such as fault diagnosis.

[0109] Example 3

[0110] Based on Example 1, this embodiment introduces a fault diagnosis system for rotating machinery, including:

[0111] The signal acquisition module is used to: acquire the vibration signal of the rotating machinery;

[0112] a fault diagnosis module, configured to: input the vibration signal of the rotating machinery into a pre-built fault diagnosis model and output a fault diagnosis result of the rotating machinery;

[0113] The fault diagnosis model includes a meta-graph construction layer, a hierarchical attention aggregation layer, and a classification layer connected in sequence. The processing steps of the fault diagnosis model include:

[0114] In the metagraph construction layer, a signal isomorphic graph is constructed according to the vibration signal of the rotating machinery, and a signal metagraph is constructed according to the signal isomorphic graph;

[0115] In the hierarchical attention aggregation layer, dual-channel signal feature extraction is performed on the signal meta-graph to obtain a node feature matrix of the signal meta-graph; multi-channel signal feature extraction is performed on the node feature matrix of the signal meta-graph to obtain a multi-channel signal feature matrix;

[0116] In the classification layer, the multi-channel signal features are classified to obtain the fault diagnosis result of the rotating machinery.

[0117] The specific functional implementation of each of the above modules can be found in the relevant content of the method in Example 1 and will not be elaborated on here.

[0118] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0119] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1A device that provides the functions specified in a block or multiple blocks.

[0120] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0121] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0122] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the purpose of the present invention and the claims, which are all protected by the present invention.

Claims

1. A method for diagnosing a rotating machinery fault, characterized in that: include: Obtain vibration signals of rotating machinery; Inputting the vibration signal of the rotating machinery into a pre-built fault diagnosis model and outputting a fault diagnosis result of the rotating machinery; The fault diagnosis model includes a meta-graph construction layer, a hierarchical attention aggregation layer, and a classification layer connected in sequence. The processing steps of the fault diagnosis model include: In the metagraph construction layer, a signal isomorphic graph is constructed according to the vibration signal of the rotating machinery, and a signal metagraph is constructed according to the signal isomorphic graph; In the hierarchical attention aggregation layer, dual-channel signal feature extraction is performed on the signal meta-graph to obtain a node feature matrix of the signal meta-graph; multi-channel signal feature extraction is performed on the node feature matrix of the signal meta-graph to obtain a multi-channel signal feature matrix; In the classification layer, the multi-channel signal features are classified to obtain the fault diagnosis result of the rotating machinery.

2. The rotating machinery fault diagnosis method according to claim 1, characterized in that: Constructing a signal isomerism graph according to the vibration signal of the rotating machinery, including: Extracting frequency domain features of the rotating machinery vibration signal; The K-nearest neighbor algorithm is used to calculate the Euclidean distance of the amplitude at the frequency point corresponding to the frequency domain feature; According to the Euclidean distance, the vibration signals of the first K rotating machines with the closest Euclidean distance are obtained; The vibration signals of the first K rotating machines closest to each other are used as heterogeneous nodes to build edge connections and obtain a signal heterogeneous graph.

3. The rotating machinery fault diagnosis method according to claim 2, characterized in that: The calculation of the Euclidean distance of the amplitude at the frequency point corresponding to the frequency domain feature includes: ； ； in, express Channel frequency domain signal; express Channel frequency domain signal; express Channel time domain signal; Indicates frequency; express Channel time domain signal; express Channel and The Euclidean distance of the amplitude of the channel signal samples at the corresponding frequency points; express Channel No. Frequency domain characteristics of signal samples; express Channel No. Frequency domain characteristics of signal samples; Indicates the total number of signal samples; represents the real part; represents the imaginary part; Represents an imaginary unit.

4. The rotating machinery fault diagnosis method according to claim 2, characterized in that: The adjacency matrix of the signal heterogeneous graph is expressed as: ； in, Represents the adjacency matrix of the signal heterogeneous graph; 、 Respectively represent the specific value of the mth row and nth column and the specific value of the nth row and mth column in the adjacency matrix; Indicates the threshold value; express Channel and The Euclidean distance of the amplitude of the channel signal at the corresponding frequency point; Indicated as conforming The first K units in the condition are connected by edges; k represents a meta-graph node.

5. The rotating machinery fault diagnosis method according to claim 1, characterized in that: Constructing a signal metagraph according to the signal heterogeneous graph includes: On the basis of the signal heterogeneous graph, constructing a meta-path instance that is closed from beginning to end; Embed the preset meta-path into the heterogeneous node matrix of the signal heterogeneous graph to obtain the feature representation of each meta-path instance; The feature representation of each meta-path instance is used as the edge feature to obtain the signal meta-graph.

6. The rotating machinery fault diagnosis method according to claim 1, characterized in that: Performing dual-channel signal feature fusion on the signal element graph to obtain a node feature matrix of the signal element graph, including: Input the input edge features of each meta-graph node in the signal meta-graph into the attention linear transformation layer, and output the attention weight of each input edge feature; The attention weight of each incoming edge feature is processed by the softmax function to obtain the attention coefficient of each incoming edge feature; Perform weighted fusion of each incoming edge feature and the attention coefficient of each incoming edge feature to obtain the fused meta-graph node feature; According to the fused meta-graph node features, the node feature matrix of the signal meta-graph is obtained.

7. The rotating machinery fault diagnosis method according to claim 6, characterized in that: The acquisition of the node feature matrix of the signal element graph includes: ； ； ； in, represents the attention coefficient of the lth input edge; represents the attention weight of the lth input edge; represents the lth incoming edge; represents the softmax activation function; Represents the features of the k-th meta-graph node; The node feature matrix representing the signal element graph; Represents feature dimension; Indicates the number of incoming edges of each metagraph node; The number of nodes representing the signal metagraph.

8. The rotating machinery fault diagnosis method according to claim 1, characterized in that: Performing multi-channel signal feature extraction on the node feature matrix of the signal element graph to obtain a multi-channel signal feature matrix, including: Input the node feature matrix of the signal meta-graph into the attention linear transformation layer, and output the attention weight of the meta-graph node; The attention weight of the meta-graph node is processed by the softmax function to obtain the attention coefficient of the meta-graph node; The node feature matrix of the signal meta-graph and the attention coefficient of the meta-graph node are weightedly fused to obtain the multi-channel signal feature matrix of the meta-graph node.

9. The rotating machinery fault diagnosis method according to claim 8, characterized in that: The acquisition of the multi-channel signal feature matrix includes: ； ； in, represents the attention coefficient of the kth meta-graph node; The attention weight of the k-th meta-graph node; The node feature matrix representing the signal element graph; represents the softmax activation function; The number of nodes representing the signal metagraph; Represents the multi-channel signal feature matrix.

10. A rotating machinery fault diagnosis system, characterized in that: include: The signal acquisition module is used to: acquire the vibration signal of the rotating machinery; a fault diagnosis module, configured to: input the vibration signal of the rotating machinery into a pre-built fault diagnosis model and output a fault diagnosis result of the rotating machinery; The fault diagnosis model includes a meta-graph construction layer, a hierarchical attention aggregation layer, and a classification layer connected in sequence. The processing steps of the fault diagnosis model include: In the metagraph construction layer, a signal isomorphic graph is constructed according to the vibration signal of the rotating machinery, and a signal metagraph is constructed according to the signal isomorphic graph; In the hierarchical attention aggregation layer, dual-channel signal feature extraction is performed on the signal meta-graph to obtain a node feature matrix of the signal meta-graph; multi-channel signal feature extraction is performed on the node feature matrix of the signal meta-graph to obtain a multi-channel signal feature matrix; In the classification layer, the multi-channel signal features are classified to obtain the fault diagnosis result of the rotating machinery.

Images