Rotary machinery fault diagnosis method and equipment based on multi-graph cooperation of intrinsic model

Abstract

The invention specifically discloses a rotary machine fault diagnosis method and equipment based on multi-graph cooperation of an intrinsic model, and belongs to the technical field of fault diagnosis. According to the method, while multi-manifold structure features of a data set are fully considered, a graph embedding thought is introduced, dimensionality reduction is performed on a fault feature set, and a processed low-dimensional feature set is input into a K-nearest neighbor classifier, so that fault mode identification is realized. According to the method, while high-dimensional nonlinear fault feature information is effectively extracted, the difficulty of equipment fault classification is reduced, and the accuracy of fault identification is improved. Besides, the constructed diagnosis model keeps good fault identification accuracy in different noise environments and training environments, the fault identification capability is effectively improved, and the method has good dimension reduction effect, noise resistance and stability.

Description

Technical Field

[0001] This invention belongs to the field of fault diagnosis technology, specifically a method, device, equipment, and storage medium for fault diagnosis of rotating machinery based on intrinsic model multi-graph collaboration. Background Technology

[0002] Driven by emerging information technologies such as the Internet of Things, big data, artificial intelligence, and cloud computing, the operation and maintenance model of coal mining equipment is constantly developing towards intelligence and smartness. Among them, rotating machinery, as a key component ensuring the stable operation of coal mining equipment, is an indispensable and important part of the equipment. However, due to its complex and variable working environment and its strong correlation with other equipment, any abnormality or failure of rotating machinery will disrupt the orderly operation of the entire coal mine production. Therefore, higher requirements are placed on the real-time status perception of rotating machinery.

[0003] To obtain more accurate and comprehensive fault status information of equipment, it is necessary to extract feature information from multi-channel vibration signals from multiple perspectives, such as the time domain, frequency domain, and time-frequency domain. However, this process inevitably leads to problems such as information redundancy and the "curse of dimensionality." Therefore, the dimensionality reduction process of "refining" high-dimensional fault feature sets has become a key step affecting the accuracy of rotating machinery fault diagnosis.

[0004] However, in real-world operating conditions, the high-dimensional nonlinear fault data collected are often embedded within different manifold structures. Traditional fault diagnosis methods based on data dimensionality reduction mostly rely on a single manifold structure, ignoring the complex and diverse structural features of the dataset. This results in fault diagnosis accuracy that fails to meet practical application requirements. Therefore, this invention introduces the concept of graph embedding and fully considers the multi-manifold structural features of the dataset to reduce the dimensionality of the high-dimensional fault feature set, thereby improving the accuracy of fault mode identification. Summary of the Invention

[0005] In order to solve at least one of the above-mentioned technical problems in the prior art, the present invention provides a method, apparatus, device and storage medium for fault diagnosis of rotating machinery based on intrinsic model multi-graph collaboration.

[0006] The first technical objective of this invention is achieved by the following technical solution: a method for diagnosing rotating machinery faults based on intrinsic model multi-graph collaboration, comprising: Vibration signals generated by rotating machinery under different operating conditions are acquired, and multi-channel feature extraction is performed on the vibration signals to obtain a high-dimensional fault feature set. Intrinsic analysis is performed on the features in the high-dimensional fault feature set to construct an intrinsic model of the features in the high-dimensional fault feature set; Multiple sample feature maps are drawn based on a high-dimensional fault feature set, and the scatter matrix corresponding to the sample feature map of each class is improved by the intrinsic model to obtain the improved scatter matrix of the sample feature map of each class; among them, the types of sample feature maps are: intra-class intrinsic map, inter-class penalty map, nearest neighbor penalty map and far neighbor intrinsic map. Based on the improved divergence matrix, an objective function is constructed, and the dimension-reduced projection matrix is ​​obtained by tuning the parameters of the objective function. By projecting the high-dimensional fault feature set using a dimension reduction projection matrix, a low-dimensional feature set is obtained. The low-dimensional feature set is then input into a k-nearest neighbor classifier for fault classification, and the rotating machinery fault diagnosis results are output.

[0007] Preferably, vibration signals generated by rotating machinery under different operating conditions are collected, and multi-channel feature extraction is performed on the vibration signals to obtain a high-dimensional fault feature set, including: The vibration signals generated by rotating machinery under different operating conditions are collected and preprocessed. Feature extraction of vibration signals is performed from three dimensions: time domain, frequency domain, and time-frequency domain, to obtain an initial high-dimensional fault feature set; The initial high-dimensional fault feature set is normalized to obtain a high-dimensional fault feature set.

[0008] Preferably, intrinsic analysis is performed on the features of the high-dimensional fault feature set to construct an intrinsic model of the features in the high-dimensional fault feature set, including: The features in the high-dimensional fault feature set are spanned into a subspace that does not lose the original topological structure features. The formula is expressed as: In the formula, x 1~ x n Represent each feature, The mean of all features. The number of features; There exists a unique projection onto the subspace orthogonal projection operator on , making each feature Decomposed into: in, , For subspace The left singular eigenvector matrix corresponding to the non-zero singular values ​​after singular value decomposition; Features generated from vibration signals acquired under any type of working condition Decompose it into: in, For unit array; For the first The first characteristic of vibration signal generation under similar working conditions One feature; For the first Orthogonal projection operator of the subspace spanned by the characteristic span of vibration signals collected under similar working conditions; Based on characteristics and characteristics The decomposition of the equation yields the intrinsic model, expressed as follows: in, The common component of all features For the first Common components of vibration signals collected under similar working conditions For the first Non-common components among the features generated by vibration signals collected under similar working conditions; , These are the intrinsic commonality matrix and the intrinsic non-commonality matrix of the same type, respectively.

[0009] Preferably, multiple sample feature maps are drawn based on a high-dimensional fault feature set, and the divergence matrix corresponding to each type of sample feature map is improved using an intrinsic model to obtain the improved divergence matrix of each type of sample feature map, including: Draw the intra-class eigenmap, and construct the adjacency matrix and intra-class scatter matrix of the intra-class eigenmap; By embedding the intrinsic model into the intra-class scatter matrix and incorporating the adjacency matrix of the intra-class intrinsic graph, an improved intra-class scatter matrix is ​​obtained. Draw the inter-class penalty graph, and construct the adjacency matrix and scatter matrix of the inter-class penalty graph; By embedding the intrinsic model into the inter-class scatter matrix and incorporating the adjacency matrix of the inter-class penalty graph, an improved inter-class scatter matrix is ​​obtained. Draw the nearest neighbor penalty graph and construct the nearest neighbor scatter matrix and nearest neighbor weight matrix of the nearest neighbor penalty graph; By embedding the intrinsic model into the nearest neighbor divergence matrix and incorporating the nearest neighbor weight matrix of the nearest neighbor penalty graph, an improved nearest neighbor divergence matrix is ​​obtained. Draw the distant neighbor eigenmap, and construct the distant neighbor scatter matrix and distant neighbor weight matrix of the distant neighbor eigenmap; By embedding the intrinsic model into the distant neighbor divergence matrix and incorporating the distant neighbor weight matrix of the distant neighbor eigenmap, an improved distant neighbor divergence matrix is ​​obtained.

[0010] Preferably, based on the improved divergence matrix, an objective function is constructed, and by tuning the parameters of the objective function, a dimension-reduced projection matrix is ​​obtained, including: To achieve the goal of clustering within classes and dispersing between classes, we aim to minimize the improved intra-class scatter matrix and maximize the improved inter-class scatter matrix. To reduce confusion between dissimilar nearest neighbors and dispersion between similar distant neighbors, we maximize the improvement of the nearest neighbor scatter matrix and minimize the improvement of the distant neighbor scatter matrix. Based on the above objectives, the constructed objective function formula is expressed as follows: in, To improve the intra-class scatter matrix, To improve the inter-class scatter matrix, To improve the nearest neighbor scatter matrix, To improve the far neighbor scatter matrix; As a regulating factor; By using the Lagrange multiplier method, the objective function is transformed into a problem of solving generalized eigenvalues. The eigenvector with the largest eigenvalue is selected to form a dimension-reduced projection matrix.

[0011] Preferably, the high-dimensional fault feature set is projected using a dimension-reduction projection matrix to obtain a low-dimensional feature set. This low-dimensional feature set is then input into a k-nearest neighbor classifier for fault classification, outputting the rotating machinery fault diagnosis results, including: By using a dimension reduction projection matrix, the high-dimensional fault feature set is projected and transformed into a low-dimensional feature set. The low-dimensional feature set is input into the k-nearest neighbor classifier for fault classification, and the output is the fault diagnosis result of rotating machinery.

[0012] Preferably, the high-dimensional fault feature set is projected using a dimension reduction projection matrix to transform it into a low-dimensional feature set, including: Based on the dimension-reduced projection matrix, a projection transformation model is constructed; the formula for the projection transformation model is expressed as: Where A is the dimension-reduced projection matrix, X is the high-dimensional fault feature set, and Y is the low-dimensional feature set.

[0013] The second technical objective of this invention is achieved by the following technical solution: a rotating machinery fault diagnosis device based on intrinsic model multi-graph collaboration, comprising: The data acquisition module is used to acquire vibration signals generated by rotating machinery under different types of operating conditions, and to extract multi-channel features from the vibration signals to obtain a high-dimensional fault feature set. The intrinsic model building module is used to perform intrinsic analysis on the features in the high-dimensional fault feature set and build an intrinsic model of the features in the high-dimensional fault feature set. The sample feature map drawing module is used to draw various sample feature maps based on a high-dimensional fault feature set, and improve the scatter matrix corresponding to each class of sample feature maps through an intrinsic model to obtain the improved scatter matrix of each class of sample feature maps; among them, the types of sample feature maps are: intra-class intrinsic maps, inter-class penalty maps, nearest neighbor penalty maps, and far neighbor intrinsic maps. The dimension reduction projection matrix calculation module is used to construct an objective function based on the improved scatter matrix, and obtain the dimension reduction projection matrix by optimizing the parameters of the objective function. The fault diagnosis module is used to project a high-dimensional fault feature set using a dimension-reduction projection matrix to obtain a low-dimensional feature set. The low-dimensional feature set is then input into a k-nearest neighbor classifier for fault classification, and the fault diagnosis results for rotating machinery are output.

[0014] The third technical objective of the present invention is achieved by the following technical solution: a computer device, including an input / output unit, a memory and a processor, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the processor performs the steps in the aforementioned technical solution method.

[0015] The fourth technical objective of this invention is achieved by the following technical solution: a storage medium storing computer-readable instructions, which, when executed by one or more processors, cause the one or more processors to perform the steps in the aforementioned technical solution method.

[0016] Compared with the prior art, the beneficial effects of the present invention are: This invention addresses the difficulties in fault classification and identification caused by the "curse of dimensionality" in fault feature sets by proposing a multi-graph collaborative fault diagnosis method for rotating machinery based on intrinsic models. This method, while fully considering the multi-manifold structure characteristics of the dataset, introduces graph embedding to reduce the dimensionality of the fault feature set. The processed low-dimensional feature set is then input into a K-nearest neighbor classifier to identify fault modes. This method effectively extracts high-dimensional nonlinear fault feature information while reducing the difficulty of equipment fault classification and improving the accuracy of fault identification. Furthermore, the constructed diagnostic model maintains good fault identification accuracy in different noise environments and training environments, effectively improving fault identification capabilities and demonstrating good dimensionality reduction, noise resistance, and stability. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating a rotating machinery fault diagnosis method based on intrinsic model multi-graph collaboration provided by the present invention.

[0019] Figure 2 This is a logical schematic diagram of a rotating machinery fault diagnosis method based on intrinsic model multi-graph collaboration provided by the present invention.

[0020] Figure 3 This is a schematic diagram of the intraclass intrinsic graph in a rotating machinery fault diagnosis method based on intrinsic model multi-graph collaboration provided by the present invention.

[0021] Figure 4 This is a schematic diagram of the inter-class penalty graph in a rotating machinery fault diagnosis method based on intrinsic model multi-graph collaboration provided by the present invention.

[0022] Figure 5 This is a schematic diagram of the nearest neighbor penalty graph in a rotating machinery fault diagnosis method based on intrinsic model multi-graph collaboration provided by the present invention.

[0023] Figure 6 This is a schematic diagram of the distant neighbor eigengraph in a rotating machinery fault diagnosis method based on intrinsic model multi-graph collaboration provided by the present invention.

[0024] Figure 7 This is a schematic diagram of the structure of a rotating machinery fault diagnosis device based on intrinsic model multi-graph collaboration according to the present invention.

[0025] Figure 8 This is a schematic diagram of the structure of a non-transitory computer-readable storage medium storing computer instructions provided by the present invention. Detailed Implementation

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other implementation methods obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] It should be noted that the structures, proportions, sizes, etc., shown in the accompanying drawings of this specification are only for the purpose of assisting those skilled in the art in understanding and reading the content disclosed in the specification, and are not intended to limit the conditions under which the present invention can be implemented. Therefore, they have no substantial technical significance. Any modifications to the structure, changes in the proportional relationships, or adjustments to the size, without affecting the effects and objectives that the present invention can produce, should fall within the scope of the technical content disclosed in the present invention. It should be noted that in this specification, relational terms such as "first" and "second" are only used to distinguish one entity from several other entities, and do not necessarily require or imply any actual relationship or order between these entities.

[0028] This invention provides an embodiment: such as Figure 1 and Figure 2 As shown, this invention provides a method for fault diagnosis of rotating machinery based on intrinsic model multi-graph collaboration, including: S110: Acquire vibration signals generated by rotating machinery under different types of operating conditions, extract multi-channel features from the vibration signals, and obtain a high-dimensional fault feature set.

[0029] In this embodiment of the invention, rotating machinery is illustrated using a rotor as an example. Specifically, the invention employs a rotor fault test bench to conduct experiments, collecting vibration signals generated by the rotor under five common operating conditions: rotor imbalance, rotor misalignment, rubbing, looseness, and normal operation. The sensors used in this experiment include two X-axis accelerometers and two Y-axis accelerometers. The sampling frequency of each sensor is 8 kHz, with 250,000 sampling points. Through a rotating machinery fault diagnosis method based on intrinsic model multi-graph collaboration, fault mode identification and classification are accurately performed on the fault feature set, improving the accuracy of fault diagnosis.

[0030] In this embodiment, vibration signals under five common rotor operating conditions are collected according to the aforementioned method. By preprocessing the vibration signals to remove noise and other interference components, the influence of the surrounding environment and the system's own vibration is reduced, ensuring the reliability and accuracy of signal acquisition.

[0031] Feature extraction of vibration signals is performed in three dimensions: time domain, frequency domain, and time-frequency domain. Each dimension contains multiple different feature types, as shown in Table 1. Feature extraction of vibration signals is performed according to the various feature types shown in Table 1 to obtain an initial high-dimensional fault feature set.

[0032] Table 1 Feature Type Table By performing standard normalization on the extracted initial high-dimensional fault feature set, the centering and scaling of the fault feature set are completed, resulting in the final high-dimensional fault feature set. The normalization process transforms the data into a standard normal distribution with a mean of 0 and a standard deviation of 1, thus resolving the analytical bias caused by differences in the dimensions and ranges of different sample features.

[0033] In an embodiment of the present invention, the high-dimensional fault feature set is divided according to a predetermined ratio to form a training set and a test set for the high-dimensional fault feature set. By dividing the training set and the test set, strong support is provided for the subsequent training and performance evaluation of the fault diagnosis model. Specifically, the ratio of the training set to the test set is 3:7.

[0034] S120: Perform intrinsic analysis on the features in the high-dimensional fault feature set and construct an intrinsic model of the features in the high-dimensional fault feature set.

[0035] In this embodiment, the structural features of the high-dimensional fault feature set are fully considered, the features are projected, an intrinsic model is constructed, and the intrinsic commonality matrix and intrinsic non-commonality matrix of the same type are obtained.

[0036] Specifically, the features in the high-dimensional fault feature set are spanned into a subspace that does not lose the original topological structure features. The formula is expressed as: In the formula, x 1~ x n Represent each feature, The mean of all features. The number of features; There exists a unique projection onto the subspace orthogonal projection operator on , making each feature Decomposed into: in, , For subspace The left singular eigenvector matrix corresponding to the non-zero singular values ​​after singular value decomposition; Features generated from vibration signals acquired under any type of working condition Decompose it into: in, For unit array; For the first The first characteristic of vibration signal generation under similar working conditions One feature; For the first Orthogonal projection operator of the subspace spanned by the characteristic span of vibration signals collected under similar working conditions; Based on characteristics and characteristics The decomposition of the equation yields the intrinsic model, expressed as follows: in, The common component of all features For the first Common components of vibration signals collected under similar working conditions For the first Non-common components among the features generated by vibration signals collected under similar working conditions; , These are the intrinsic commonality matrix and the intrinsic non-commonality matrix of the same type, respectively.

[0037] S130: Based on the high-dimensional fault feature set, draw various sample feature maps, and improve the scatter matrix corresponding to each type of sample feature map through the intrinsic model to obtain the improved scatter matrix of each type of sample feature map.

[0038] In this embodiment, to enhance the ability to mine equipment fault feature information, a graph embedding concept is introduced, embedding the constructed intrinsic model into the divergence matrix of the sample feature map. This process involves drawing four sample feature maps: intra-class intrinsic map, inter-class penalty map, nearest neighbor penalty map, and far neighbor intrinsic map; simultaneously, four divergence matrices are established: intra-class divergence matrix, inter-class divergence matrix, nearest neighbor divergence matrix, and far neighbor divergence matrix.

[0039] Specifically, drawing as follows Figure 3 The intra-class intrinsic eigenmap shown assigns weights to adjacent samples of the same feature class based on label information. This ensures that adjacent samples of the same class remain close after being mapped to the low-dimensional space, thus achieving greater intra-class clustering of samples of the same feature class. An adjacency matrix is ​​constructed based on the intra-class intrinsic eigenmap. The intrinsic model is embedded into the intra-class scatter matrix, using non-common components of the same class. Replace similar samples By incorporating the adjacency matrix of the intra-class eigenmap, an improved intra-class scatter matrix is ​​obtained. The formula for the improved intra-class scatter matrix is ​​expressed as: in, To improve the intra-class scatter matrix; Number of operating condition categories; The number of features of the same type; For the first projection The vibration signal generated during operation under certain conditions was extracted as follows: The inherent non-common components of each characteristic , The projection matrix; For the first The vibration signal generated during operation under certain conditions was extracted as follows: The inherent non-common components of each characteristic , For the corresponding number Orthogonal projection operators for various working conditions; ; It is a diagonal matrix, and its diagonal elements are... ; For Laplace matrix, , .in, The first vibration signal extracted during operation. The adjacency matrix of the intrinsic non-common components of each feature, whose elements are: In the formula, for of The neighborhood of non-common components within a class is formed by their nearest neighbors. The settings are based on specific experimental data.

[0040] Drawing as Figure 4 The inter-class penalty graph shown assigns weights to adjacent out-of-class feature samples based on label information, ensuring that adjacent out-of-class samples in the high-dimensional space are far apart after being mapped to the low-dimensional space, thus achieving greater dispersion of out-of-class feature samples. The adjacency matrix and inter-class scatter matrix of the inter-class penalty graph are constructed, and the intrinsic model is embedded into the inter-class scatter matrix, using common components of the same class. The mean of the class is replaced with the mean of the class in the between-class scatter matrix, and the adjacency matrix of the between-class penalty graph is used. By integrating, we obtain the improved inter-class scatter matrix.

[0041] Specifically, improve the inter-class scatter matrix The formula is expressed as: In the formula: For the first projection The mean of the intrinsic common components of the sample class. ; For the first i The mean value of the intrinsic common components of the features extracted from the vibration signals generated under various operating conditions. ; ; It is a diagonal matrix, and its diagonal elements are... ; For a matrix, ; Let be the adjacency matrix of the common components, and its elements are: In the formula, for of A neighborhood consisting of 1 nearest neighbors The settings are based on specific experimental data.

[0042] Drawing as Figure 5 The nearest neighbor penalty graph shown assigns greater weight to out-of-class nearest neighbor samples based on sample label information. This keeps out-of-class nearest neighbor samples away from confusing samples, increasing the distance between them and thus mitigating confusion and improving the discriminative power between them. The nearest neighbor scatter matrix and nearest neighbor weight matrix of the nearest neighbor penalty graph are constructed as follows: Embedding the intrinsic model into the nearest neighbor divergence matrix, and the nearest neighbor weight matrix of the nearest neighbor penalty graph. By incorporating, an improved nearest-neighbor scatter matrix is ​​obtained. .

[0043] Improved nearest neighbor scatter matrix The formula is expressed as: In the formula: It is a diagonal matrix, and the diagonal elements are ; For a matrix, ; in, The matrix is ​​the nearest neighbor weight matrix of the mean of the intrinsic common components. In the formula, , For the first i The mean value of the intrinsic common components of the characteristics extracted from the vibration signals generated under various operating conditions; The square of the mean Euclidean distance among all features; for of A neighborhood consisting of 1 nearest neighbors The settings are based on specific experimental data.

[0044] Drawing as Figure 6 The distant neighbor intrinsic map shown assigns greater weights to similar distant neighbor feature samples based on sample label information, causing similar distant neighbor samples to aggregate towards the cluster center and enhancing the ability of similar feature samples to cluster more effectively. The distant neighbor scatter matrix and distant neighbor weight matrix of the distant neighbor intrinsic map are constructed. Embed the intrinsic model into the distant neighbor scatter matrix, and use the distant neighbor weight matrix of the distant neighbor eigenmap. By incorporating, we obtain the improved distant neighbor scatter matrix. .

[0045] Improved far neighbor scatter matrix The formula is expressed as: In the formula: It is a diagonal matrix, and its diagonal elements are... ; For Laplace matrix, , where the matrix For the first The nearest neighbor weight matrix of the intrinsic non-common components of the vibration signals extracted under similar working conditions has the following elements: In the formula, , For the first i The vibration signal generated during operation under various working conditions was extracted as follows: i The inherent non-common components of each characteristic; For the first m The square of the mean Euclidean distance between the features extracted from the vibration signals generated under one type of operating condition and the features extracted from the vibration signals generated under other types of operating conditions. for of The neighborhood of a class consists of the distant neighbors of its non-common components. The settings are based on specific experimental data.

[0046] S140: Based on the improved divergence matrix, construct the objective function, and obtain the dimension-reduced projection matrix by optimizing the parameters of the objective function.

[0047] To achieve the goal of clustering within classes and dispersing between classes, the objective function needs to minimize the intra-class scatter matrix. And maximize the inter-class scatter matrix To reduce confusion between dissimilar nearest neighbors and dispersion between similar distant neighbors, it is necessary to maximize the dissimilar nearest neighbor scatter matrix. And minimize the scatter matrix of similar distant neighbors. Based on the above analysis, the objective function formula constructed in this invention is expressed as: In the formula, This is an adjustment factor used to adjust the contribution rate of different scatter matrices. Using the Lagrange multiplier method, it is transformed into a problem of solving generalized eigenvalues, i.e. During the solution process, through... , , , Perform parameter tuning and sort the eigenvalues ​​in descending order, taking the maximum value. Each eigenvalue corresponds to an eigenvector, thus yielding the optimal dimensionality reduction projection matrix. .

[0048] S150: Project the high-dimensional fault feature set using a dimension reduction projection matrix to obtain a low-dimensional feature set. Input the low-dimensional feature set into a k-nearest neighbor classifier for fault classification and output the fault diagnosis results of rotating machinery.

[0049] In this embodiment, based on the dimension-reduced projection matrix , training set and test set Substitution Projection mapping is performed separately to obtain low-dimensional feature sets. and The low-dimensional feature set after dimensionality reduction and projection. and enter The nearest neighbor classifier classifies faults and outputs the identification results.

[0050] Optionally, after inputting the projected low-dimensional feature set into the classifier for fault classification, the method further includes: calculating the separability index of the model, and conducting stability and noise resistance experiments. To verify the applicability of the fault diagnosis model, this embodiment selects four algorithms—LPP, LDA, IDA, and DGDP—for comparative experimental analysis with this method.

[0051] In this embodiment, a separability index is introduced to evaluate the effectiveness of the diagnostic model in reducing the dimensionality of the fault feature set. This is used to calculate the separability between different fault categories in a low-dimensional test set after dimensionality reduction. Separability index A higher value indicates lower intra-class dispersion, while higher inter-class dispersion indicates better dimensionality reduction performance of the algorithm. The formula for calculating the separability index is: In the formula, The average dispersion among the various categories; Number of categories; For the first Prior probability of a class , For the first i The number of features of a class The total number of characteristics; For the first The dispersion of the class.

[0052] By conducting comparative experiments with four algorithms—LPP, LDA, IDA, and DGDP—and the method of this invention, the dimensionality reduction separability indices of each algorithm are shown in Table 2. It is evident that the MCIM algorithm of this invention has a significantly higher separability index than the other algorithms. This indicates that after processing the sample data using this algorithm, similar data are more compactly clustered, while dissimilar data are more dispersed; that is, the dimensionality reduction effect of this algorithm is better than other algorithms.

[0053] Table 2 Separability Indicators In this embodiment, to evaluate the noise resistance of the diagnostic model, random noise with different signal-to-noise ratios (SNRs) of -4dB, -2dB, 0dB, 2dB, and 4dB was added to the collected fault dataset for experimental testing. The noise-added fault dataset was then input into LPP, LDA, IDA, DGDP, and the MCIM diagnostic model of this invention. The identification accuracy of each algorithm under different SNRs was obtained, as shown in Table 3. According to the identification accuracy results of each diagnostic model in Table 3, it can be seen that as the SNR of the added random noise gradually increases, the identification accuracy of the MCIM algorithm of this invention also gradually increases, indicating better noise resistance. This shows that compared to the other four algorithms, this algorithm can still maintain good fault identification capability under different levels of noise.

[0054] Table 3 Noise Resistance Test In this embodiment, to evaluate the stability of the diagnostic model, the training and test samples were divided into groups of 10 / 90, 20 / 80, 30 / 70, 40 / 60, 50 / 50, 60 / 40, 70 / 30, 80 / 20, and 90 / 10, respectively. The extracted fault feature sets were then input into LPP, LDA, IDA, DGDP, and the MCIM diagnostic model of this invention. The identification accuracy of each diagnostic model under different ratios of training and test samples was obtained, as shown in Table 4. According to the identification accuracy results of each diagnostic model in Table 4, it can be seen that compared to the other four algorithms, the identification accuracy of the MCIM algorithm of this invention gradually increases and remains stable as the ratio of training samples gradually increases, demonstrating good fault identification capability under different training sample ratios.

[0055] Table 4 Stability Experiment This invention addresses the difficulties in fault classification and identification caused by the "curse of dimensionality" in fault feature sets by proposing a multi-graph collaborative fault diagnosis method for rotating machinery based on intrinsic models. This method, while fully considering the multi-manifold structure characteristics of the dataset, introduces graph embedding to reduce the dimensionality of the fault feature set. The processed low-dimensional feature set is then input into a K-nearest neighbor classifier to identify fault modes. This method effectively extracts high-dimensional nonlinear fault feature information while reducing the difficulty of equipment fault classification and improving the accuracy of fault identification. Furthermore, the constructed diagnostic model maintains good fault identification accuracy in different noise environments and training environments, effectively improving fault identification capabilities and demonstrating good dimensionality reduction, noise resistance, and stability.

[0056] like Figure 7 As shown, this invention proposes a rotating machinery fault diagnosis device based on intrinsic model multi-graph collaboration, comprising: The data acquisition module 710 is used to acquire vibration signals generated by rotating machinery under different types of working conditions, perform multi-channel feature extraction on the vibration signals, and obtain a high-dimensional fault feature set. The intrinsic model construction module 720 is used to perform intrinsic analysis on the features in the high-dimensional fault feature set and construct the intrinsic model of the features in the high-dimensional fault feature set. The sample feature map drawing module 730 is used to draw various sample feature maps based on a high-dimensional fault feature set, and improve the scatter matrix corresponding to each class of sample feature maps through an intrinsic model to obtain the improved scatter matrix of each class of sample feature maps; among them, the types of sample feature maps are: intra-class intrinsic maps, inter-class penalty maps, nearest neighbor penalty maps, and far neighbor intrinsic maps. The dimension reduction projection matrix calculation module 740 is used to construct an objective function based on the improved scatter matrix, and obtain the dimension reduction projection matrix by optimizing the parameters of the objective function; The fault diagnosis module 750 is used to project a high-dimensional fault feature set using a dimension reduction projection matrix to obtain a low-dimensional feature set. The low-dimensional feature set is then input into a k-nearest neighbor classifier for fault classification, and the fault diagnosis results for rotating machinery are output.

[0057] To implement the embodiments, the present invention also proposes an electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of the method described above.

[0058] like Figure 8As shown, the non-transitory computer-readable storage medium includes a memory 810 for instructions and an interface 830, the instructions of which can be executed by a processor 820 to complete the method. Optionally, the storage medium can be a non-transitory computer-readable storage medium, such as a read-only memory (ROM), random access memory (RAM), read-only optical disc (CD-ROM), magnetic tape, floppy disk, and optical data storage device.

[0059] To implement the embodiments, the present invention also proposes a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs rotating machinery fault diagnosis as described in the embodiments of the present invention.

[0060] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for fault diagnosis of rotating machinery based on intrinsic model multi-graph collaboration, characterized in that, include: Vibration signals generated by rotating machinery under different operating conditions are acquired, and multi-channel feature extraction is performed on the vibration signals to obtain a high-dimensional fault feature set. Intrinsic analysis is performed on the features in the high-dimensional fault feature set to construct an intrinsic model of the features in the high-dimensional fault feature set; Based on the high-dimensional fault feature set, various sample feature maps are drawn, and the scatter matrix corresponding to the sample feature map of each class is improved by the intrinsic model to obtain the improved scatter matrix of the sample feature map of each class; wherein, the types of sample feature maps are: intra-class intrinsic map, inter-class penalty map, nearest neighbor penalty map and far neighbor intrinsic map. Based on the improved divergence matrix, an objective function is constructed, and by optimizing the parameters of the objective function, a dimension-reduced projection matrix is ​​obtained. The high-dimensional fault feature set is projected using the dimensionality reduction projection matrix to obtain a low-dimensional feature set. The low-dimensional feature set is then input into a k-nearest neighbor classifier for fault classification, and the rotating machinery fault diagnosis result is output.

2. The rotating machinery fault diagnosis method based on intrinsic model multi-graph collaboration according to claim 1, characterized in that, Vibration signals generated by rotating machinery under different operating conditions are collected, and multi-channel feature extraction is performed on the vibration signals to obtain a high-dimensional fault feature set, including: Vibration signals generated by rotating machinery under different operating conditions are collected, and the vibration signals are preprocessed. Feature extraction is performed on the vibration signal from three dimensions: time domain, frequency domain, and time-frequency domain, to obtain an initial high-dimensional fault feature set; The initial high-dimensional fault feature set is normalized to obtain the high-dimensional fault feature set.

3. The rotating machinery fault diagnosis method based on intrinsic model multi-graph collaboration according to claim 2, characterized in that, The intrinsic analysis of the features in the high-dimensional fault feature set is performed to construct an intrinsic model of the features in the high-dimensional fault feature set, including: The features in the high-dimensional fault feature set are spanned into a subspace that does not lose the original topological structure features. The formula is expressed as: In the formula, x 1~ x n Represent each feature, The mean of all features. The number of features; There exists a unique projection onto the subspace. orthogonal projection operator on , making each feature Decomposed into: in, , For subspace The left singular eigenvector matrix corresponding to the non-zero singular values ​​after singular value decomposition; Features generated from vibration signals acquired under any type of working condition Decompose it into: in, For unit array; For the first The first characteristic of vibration signal generation under similar working conditions One feature; For the first Orthogonal projection operator of the subspace spanned by the characteristic span of vibration signals collected under similar working conditions; Based on characteristics and characteristics The decomposition of the equation yields the intrinsic model, expressed as follows: in, The common component of all features For the first Common components of vibration signals collected under similar working conditions For the first Non-common components among the features generated by vibration signals collected under similar working conditions; , These are the intrinsic commonality matrix and the intrinsic non-commonality matrix of the same type, respectively.

4. The rotating machinery fault diagnosis method based on intrinsic model multi-graph collaboration according to claim 3, characterized in that, Based on the high-dimensional fault feature set, various sample feature maps are drawn, and the divergence matrix corresponding to each type of sample feature map is improved using the intrinsic model to obtain the improved divergence matrix of each type of sample feature map, including: Draw the intra-class eigenmap, and construct the adjacency matrix and intra-class scatter matrix of the intra-class eigenmap; The intrinsic model is embedded into the intra-class scatter matrix, and the adjacency matrix of the intra-class intrinsic graph is incorporated to obtain the improved intra-class scatter matrix. Draw the inter-class penalty graph, and construct the adjacency matrix and inter-class scatter matrix of the inter-class penalty graph; The intrinsic model is embedded into the inter-class scatter matrix, and the adjacency matrix of the inter-class penalty graph is incorporated to obtain the improved inter-class scatter matrix. Draw a nearest neighbor penalty graph and construct the nearest neighbor scatter matrix and nearest neighbor weight matrix of the nearest neighbor penalty graph; The intrinsic model is embedded into the nearest neighbor divergence matrix, and the nearest neighbor weight matrix of the nearest neighbor penalty graph is incorporated to obtain the improved nearest neighbor divergence matrix. Draw the distant neighbor eigenmap, and construct the distant neighbor scatter matrix and distant neighbor weight matrix of the distant neighbor eigenmap; The intrinsic model is embedded into the distant neighbor divergence matrix, and the distant neighbor weight matrix of the distant neighbor eigenmap is incorporated to obtain the improved distant neighbor divergence matrix.

5. The rotating machinery fault diagnosis method based on intrinsic model multi-graph collaboration according to claim 4, characterized in that, Based on the improved divergence matrix, an objective function is constructed. By fine-tuning the parameters of the objective function, a dimension-reduced projection matrix is ​​obtained, including: To achieve the goal of clustering within classes and dispersing between classes, the improved intra-class scatter matrix is ​​minimized, and the improved inter-class scatter matrix is ​​maximized. To reduce confusion between dissimilar nearest neighbors and dispersion between similar distant neighbors, the improved nearest neighbor scatter matrix is ​​maximized and the improved distant neighbor scatter matrix is ​​minimized. Based on the above objectives, the constructed objective function formula is expressed as follows: in, To improve the intra-class scatter matrix, To improve the inter-class scatter matrix, To improve the nearest neighbor scatter matrix, To improve the far neighbor scatter matrix; As a regulating factor; Using the Lagrange multiplier method, the objective function is transformed into a problem of solving generalized eigenvalues. The eigenvector with the largest eigenvalue is selected to form a dimension-reduced projection matrix.

6. The rotating machinery fault diagnosis method based on intrinsic model multi-graph collaboration according to claim 5, characterized in that, The high-dimensional fault feature set is projected using the dimensionality reduction projection matrix to obtain a low-dimensional feature set. This low-dimensional feature set is then input into a k-nearest neighbor classifier for fault classification, outputting rotating machinery fault diagnosis results, including: The high-dimensional fault feature set is projected using the dimensionality reduction projection matrix, thereby converting the high-dimensional fault feature set into a low-dimensional feature set. The low-dimensional feature set is input into a k-nearest neighbor classifier for fault classification, and the fault diagnosis results of rotating machinery are output.

7. The rotating machinery fault diagnosis method based on intrinsic model multi-graph collaboration according to claim 6, characterized in that, The high-dimensional fault feature set is projected using the dimensionality reduction projection matrix, transforming it into a low-dimensional feature set, including: Based on the reduced-dimensional projection matrix, a projection transformation model is constructed; the formula for the projection transformation model is expressed as: Where A is the dimension-reduced projection matrix, X is the high-dimensional fault feature set, and Y is the low-dimensional feature set.

8. A rotating machinery fault diagnosis device based on intrinsic model multi-graph collaboration, characterized in that, include: The data acquisition module is used to acquire vibration signals generated by rotating machinery under different types of operating conditions, and to perform multi-channel feature extraction on the vibration signals to obtain a high-dimensional fault feature set. An intrinsic model construction module is used to perform intrinsic analysis on the features in the high-dimensional fault feature set and construct an intrinsic model of the features in the high-dimensional fault feature set. The sample feature map drawing module is used to draw various sample feature maps based on the high-dimensional fault feature set, and improve the scatter matrix corresponding to the sample feature map of each class through the intrinsic model to obtain the improved scatter matrix of the sample feature map of each class; wherein, the types of sample feature maps are: intra-class intrinsic map, inter-class penalty map, nearest neighbor penalty map and far neighbor intrinsic map. The dimension reduction projection matrix calculation module is used to construct an objective function based on the improved scatter matrix, and obtain the dimension reduction projection matrix by optimizing the parameters of the objective function; The fault diagnosis module is used to project the high-dimensional fault feature set through the dimensionality reduction projection matrix to obtain a low-dimensional feature set, input the low-dimensional feature set into a k-nearest neighbor classifier for fault classification, and output the rotating machinery fault diagnosis result.

9. A computer device, characterized in that, The method includes an input / output unit, a memory, and a processor, wherein the memory stores computer-readable instructions that, when executed by the processor, cause the processor to perform the steps of the method as described in any one of claims 1 to 7.

10. A storage medium storing computer-readable instructions, characterized in that, When the computer-readable instructions are executed by one or more processors, they cause the one or more processors to perform the steps in the method as described in any one of claims 1 to 7.

Images