Mechanical equipment fault diagnosis method and device constructed based on feature map, equipment and medium

Abstract

The invention discloses a mechanical equipment fault diagnosis method and device constructed based on a feature map, equipment and a medium, and relates to the technical field of mechanical equipment fault diagnosis. The diagnosis method comprises the following steps: S1, acquiring a running signal sample, and respectively extracting original features of multiple domains; s2, setting a standard according to the original features, and screening out an effective feature subset; and S3, setting an edge construction criterion according to the effective feature subset as a node, and calculating a node adjacency matrix. And S4, according to the effective feature subset and the adjacent matrix, constructing an operation feature map of the mechanical equipment. And S5, constructing a graph attention model, optimizing the hyper-parameter combination, and obtaining a to-be-trained diagnosis model. And S6, training the to-be-trained diagnosis model to obtain a trained diagnosis model. And S7, obtaining an operation signal of the to-be-diagnosed mechanical equipment, performing preprocessing and effective feature extraction, and then constructing an operation feature map. And finally, inputting the operation characteristic graph of the mechanical equipment to be diagnosed into the trained diagnosis model to obtain a fault diagnosis result of the equipment.

Description

Technical Field

[0001] The present invention relates to the technical field of mechanical equipment fault diagnosis. Specifically, it relates to a mechanical equipment fault diagnosis method, device, equipment, and medium based on feature map construction. Background Art

[0002] In a mechanical system, the reliability and stability of equipment directly affect the overall performance and safety of the system. With the development of technology, the fault diagnosis of mechanical equipment has become increasingly complex, especially in complex operating environments. During the long-term operation of mechanical equipment, due to the influence of external factors and the degradation of internal components, various fault forms may occur. Many faults are difficult to detect in the early stage. Once they occur, they often trigger larger faults, resulting in equipment downtime, increased maintenance costs, and even threatening the safety of personnel and the environment.

[0003] With the development of sensor technology, it has become easier to obtain equipment data. Therefore, traditional signal excitation model-based fault diagnosis has gradually shifted to data-driven methods. Data-driven fault diagnosis first collects the operating signals of equipment through sensors, then uses signal processing techniques for feature extraction, and finally performs classification and fault diagnosis through machine learning models.

[0004] However, traditional data-driven fault diagnosis has some significant deficiencies: on the one hand, feature extraction usually relies on manual experience and is difficult to comprehensively capture the complex non-linear fault features of equipment. On the other hand, the complex interaction relationships between components in a mechanical system have not been effectively modeled, resulting in the traditional method often ignoring the dependencies between components when identifying faults, affecting the accuracy and robustness of diagnosis. In addition, the fault modes of equipment usually have time-varying and non-linear characteristics, and traditional methods are prone to failure when dealing with these dynamic changes.

[0005] In this context, a new method is needed that can better capture the complex relationships of signals during equipment operation, extract non-linear features in dynamic changes, thereby improving the accuracy, flexibility, and robustness of fault diagnosis, and further enhancing the operating efficiency and maintenance management level of equipment. Summary of the Invention

[0006] The present invention provides a mechanical equipment fault diagnosis method, device, equipment, and medium based on feature map construction to improve at least one of the above technical problems.

[0007] In a first aspect, the present invention provides a mechanical equipment fault diagnosis method based on feature map construction, which includes steps S1 to S7.

[0008] S1. Obtain multiple samples of operating signals and extract original features in multiple domains for each sample respectively.

[0009] S2. Based on the original features of multiple domains of the multiple samples, taking the importance of each feature for data classification as the criterion, filter out the effective feature subset.

[0010] S3. Based on the effective feature subset, construct an adjacency matrix of the edges with the effective features as nodes based on the variability between samples.

[0011] S4. Based on the effective feature subset and the adjacency matrix, construct an operation feature map of the mechanical equipment.

[0012] S5. Construct a graph attention model, and optimize the hyperparameter combination by optimizing the cost with a cost function to obtain the diagnostic model to be trained.

[0013] S6. Based on the operation feature map of the mechanical equipment, train the diagnostic model to be trained to obtain the trained diagnostic model.

[0014] S7. Obtain the operation signal of the mechanical equipment to be diagnosed, perform preprocessing and extract effective features, and then construct an operation feature map. Finally, input the operation feature map of the mechanical equipment to be diagnosed into the trained diagnostic model to obtain the fault diagnosis result of the equipment.

[0015] As a preferred aspect of the present invention, step S2 specifically includes: Calculate the within-class sample variability of the original features of multiple domains of the multiple samples respectively. Among them, . In the formula, is the within-class sample variability, represents the value of the th feature of the th sample, is the value of the th feature of the th sample, represents the distance of the sample in the Euclidean space.

[0016] Calculate the between-class sample variability of the original features of multiple domains of the multiple samples respectively. Among them, . In the formula, is the between-class sample variability, is the variance calculation function, represents the value set of the th feature in all samples.

[0017] Based on the within-class sample variability and the between-class sample variability, construct a feature sensitivity index. Among them, . In the formula, is the feature sensitivity index.

[0018] Calculate the mean of the feature sensitivity indices, and filter out the original features whose feature sensitivity indices are less than the mean to obtain the effective feature subset. Among them, . In the formula, is the effective feature subset, is the original feature set, is the set of feature sensitivity indices of all original features, is the number of original features.

[0019] As a preferred aspect of the present invention, step S3 specifically includes: Calculate the Euclidean distance between different nodes according to the effective feature subset. Among them, . In the formula, is the Euclidean distance, and are the serial numbers of the nodes, is the serial number of the sample, is the total number of samples, represents the value of the th sample at node , represents the value of the th sample at node .

[0020] Calculate the cosine similarity between different nodes according to the effective feature subset. Among them, . In the formula, is the cosine similarity, and are the serial numbers of the nodes, is the value of node , is the value of node .

[0021] Calculate the combined similarity between nodes according to the Euclidean distance and the preselected similarity. Among them, . In the formula, is the combined similarity between nodes, is the cosine similarity between node and node , is the value of node and node is the Euclidean distance between nodes.

[0022] Calculate the similarity threshold according to the combined similarity. Among them, . In the formula, is the combined similarity between node and node , is the total number of samples.

[0023] According to the similarity threshold and the effective feature subset, construct an adjacency matrix of edges with effective features as nodes. Among them, . In the formula, is the adjacency matrix of the edges.

[0024] As a preferred aspect of the present invention, step S5 specifically includes: Construct a graph attention model.

[0025] Through a hyperparameter combination optimization method based on Bayesian optimization, optimize the hyperparameter combination of the graph attention model to obtain a diagnostic model to be trained.

[0026] As a preferred aspect of the present invention, through a hyperparameter combination optimization method based on Bayesian optimization, optimize the hyperparameter combination of the graph attention model to obtain a diagnostic model to be trained, which specifically includes: Construct the prior distribution of the objective function: . In the formula, is the objective function, represents a Gaussian process, is the mean function of the Gaussian process, is the covariance function, used to measure the correlation between two different hyperparameter combinations and .

[0027] Based on the prior distribution and historical data, update the posterior distribution of the Gaussian process.

[0028] .

[0029] In the formula, is the posterior distribution model, is the objective function, is the historical data, represents a Gaussian distribution, represents the predicted mean of the objective function at the hyperparameter combination , represents the predicted variance.

[0030] Select an acquisition function. . In the formula, is the expected improvement acquisition function, is the expected value, represents taking the larger value, is the objective function, is the currently known optimal objective function value.

[0031] Based on the evaluation points selected by the acquisition function, the objective function values are then calculated and the results are added to the historical dataset. Then, the Gaussian process model is updated with the new data, and this process is repeated until the stopping criterion is met to obtain the optimal initial parameter combination of the model and acquire the diagnostic model to be trained.

[0032] As a preferred aspect of the present invention, step S1 specifically includes: The operating signals of the equipment under different working conditions are collected by sensors installed on the mechanical equipment as samples to form a basic dataset.

[0033] Through time-domain analysis and frequency-domain analysis, a variety of original feature parameters in multiple domains are extracted from the preprocessed basic dataset to construct a high-dimensional feature set including the operating state of the equipment. Among them, the original features include: mean value, root mean square amplitude, maximum value, minimum value, absolute mean value, root mean square value, standard deviation, variance, peak value, peak-to-peak value, kurtosis, skewness, impulse factor, margin factor, waveform factor, amplitude factor, center frequency, average frequency, root mean square frequency, mean square frequency, frequency variance, and frequency standard deviation.

[0034] In a second aspect, the present invention provides a mechanical equipment fault diagnosis device based on feature map construction, which includes a feature extraction module, a feature screening module, a graph edge construction module, a graph construction module, a model construction module, a model training module, and a fault diagnosis module.

[0035] The feature extraction module is used to obtain multiple samples of the operating signal and extract a variety of original features from each sample respectively.

[0036] The feature screening module is used to screen out an effective feature subset based on the original features in multiple domains of the multiple samples, with the importance of each feature for data classification as the criterion.

[0037] The graph edge construction module is used to construct an adjacency matrix of edges with effective features as nodes based on the effective feature subset and the variability between samples.

[0038] The graph construction module is used to construct an operating feature map of the mechanical equipment according to the effective feature subset and the adjacency matrix.

[0039] The model construction module is used to construct a graph attention model and optimize the hyperparameter combination by optimizing the cost with a cost function to obtain a diagnostic model to be trained.

[0040] The model training module is used to train the diagnostic model to be trained according to the operating feature map of the mechanical equipment to obtain a trained diagnostic model.

[0041] A fault diagnosis module is used to obtain the operating signals of the mechanical equipment to be diagnosed, perform preprocessing and extract effective features, and then construct an operating feature map. Finally, the operating feature map of the mechanical equipment to be diagnosed is input into the trained diagnosis model to obtain the fault diagnosis result of the equipment.

[0042] In a third aspect, the present invention provides a mechanical equipment fault diagnosis device based on feature map construction, which includes a processor, a memory, and a computer program stored in the memory. The computer program can be executed by the processor to implement a mechanical equipment fault diagnosis method based on feature map construction as described in any paragraph of the first aspect.

[0043] In a fourth aspect, the present invention provides a computer-readable storage medium. The computer-readable storage medium includes a stored computer program, wherein when the computer program runs, it controls the device where the computer-readable storage medium is located to execute a mechanical equipment fault diagnosis method based on feature map construction as described in any paragraph of the first aspect.

[0044] By adopting the above technical solutions, the present invention can achieve the following technical effects: The mechanical equipment fault diagnosis method based on feature map construction in the embodiment of the present invention can effectively capture the complex relationships between various components of the mechanical equipment and improve the accuracy of fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the technical solutions of the present invention, the drawings required for the specific embodiments of the present invention will be briefly introduced below. It should be understood that the following drawings only show some specific embodiments of the present invention, and therefore should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can be obtained based on these drawings without creative efforts.

[0046] Figure 1 It is a schematic flow diagram of mechanical equipment fault diagnosis.

[0047] Figure 2 It is a schematic network structure diagram of mechanical equipment fault diagnosis. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0048] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0049] Embodiment 1. Please refer to Figures 1 to 2, the first embodiment of the present invention provides a mechanical equipment fault diagnosis method based on feature graph construction. The flowchart of the mechanical equipment fault diagnosis method is as shown in Figure 2 Figure 1. First, extract effective features from the original operation signals, extract edge information through node features and construct a graph structure, and input the graph data structure into the graph attention model with optimized parameters for fault diagnosis. The mechanical equipment fault diagnosis method mainly consists of two parts: feature graph construction and graph attention network parameter optimization.

[0050] The feature graph construction is specifically as follows: In the feature graph construction stage, first extract the features of the original operation signals through time-frequency analysis, and these features include information in the frequency domain, time domain, and time-frequency domain. Then, use the feature sensitivity index for feature screening to select features with high discrimination for equipment faults. Subsequently, use the KNN algorithm and cosine similarity to construct the edges in the graph to reflect the similarity relationship between features. Each node represents a time-frequency feature, and the edges connect those features with high similarity. In this way, the graph can effectively represent the dependency relationship between various components of the equipment and provide strong support for fault diagnosis.

[0051] The graph attention network parameter optimization is specifically as follows: In the graph attention network parameter optimization stage, use the Bayesian optimization method to adjust the hyperparameters of the model. Bayesian optimization can automatically optimize the selection of hyperparameters through the modeling and probabilistic reasoning of the objective function, avoiding the traditional manual adjustment process. The optimized hyperparameters include the learning rate, attention coefficient, number of layers, etc., and these parameters are crucial for the learning effect and fault diagnosis performance of the model. Through Bayesian optimization, the optimal combination of hyperparameters can be found during the training process to further improve the diagnostic accuracy and robustness of the graph attention network.

[0052] In this embodiment, specifically, the mechanical equipment fault diagnosis method can be executed by a mechanical equipment fault diagnosis device (hereinafter referred to as: diagnosis device). In particular, it is executed by one or more processors in the diagnosis device to implement steps S1 to S7.

[0053] S1. Obtain multiple samples of the operation signals, and extract the original features in multiple domains for each sample respectively. Among them, the multiple samples include healthy samples and fault samples.

[0054] Preferably, step S1 specifically includes steps S11 to S12.

[0055] S11. Collect the operation signals of the equipment under different working conditions as samples through the sensors installed on the mechanical equipment to form a basic data set. Among them, the operation signals include vibration, temperature, pressure, etc.

[0056] S12. Through time-domain analysis and frequency-domain analysis, extract the original feature parameters of multiple domains from the preprocessed basic dataset, and construct a high-dimensional feature set containing the operating state of the device.

[0057] The original signal is usually disturbed by environmental noise. Therefore, before analysis, it is necessary to preprocess the signal. Common preprocessing methods include denoising, normalization, and signal smoothing, etc., to improve the quality and reliability of the data.

[0058] Regarding the problem of traditional feature parameters being selected based on artificial experience, there are unreasonable problems. In the present invention, all 16 time-domain features and 6 frequency-domain features are extracted and combined to form a high-dimensional feature set with a dimension of 22.

[0059] Original feature 1, Mean value : .

[0060] Original feature 2, Root mean square amplitude : .

[0061] Original feature 3, Maximum value : .

[0062] Original feature 4, Minimum value : .

[0063] Original feature 5, Absolute mean : .

[0064] Original feature 6, Root mean square value : .

[0065] Original feature 7, Standard deviation : .

[0066] Original feature 8, Variance : .

[0067] Original feature 9, Peak value : .

[0068] Original feature 10, Peak-to-peak value : .

[0069] Original feature 11, Kurtosis : .

[0070] Original feature 12, Skewness : .

[0071] Original feature 13, Pulse factor :[[]] .

[0072] Original feature 14, Margin factor :[[]] .

[0073] Original feature 15, Waveform factor :[[]] .

[0074] Original feature 16, Amplitude factor :[[]] .

[0075] Original feature 17, Center frequency :[[]] .

[0076] Original feature 18, Average frequency :[[]] .

[0077] Original feature 19, Root mean square frequency :[[]] .

[0078] Original feature 20, Mean square frequency :[[]] .

[0079] Original feature 21, Frequency variance :[[]] .

[0080] Original feature 22, Frequency standard deviation :[[]] .

[0081] Wherein: is the total number of signal sampling points, is the signal value of the th sampling point, represents the absolute value, represents taking the maximum value among all sampling points, represents taking the minimum value among all sampling points, is the total number of frequency components, is the frequency value of the th frequency component, is the amplitude of the th frequency component.

[0082] S2. Based on the original features of multiple domains of the multiple samples, using the importance of each feature for data classification as a criterion, an effective feature subset is screened out. In this embodiment, the screened effective features are used as the nodes of the graph structure.

[0083] The high-dimensional original features extracted in step S1 often contain too much redundant information, and at the same time, the excessive dimension will lead to an increase in the computational burden. Therefore, the present invention further screens the feature dimension by setting a feature sensitivity index and constructing a corresponding screening criterion.

[0084] Preferably, step S21 specifically includes steps S22 to S24.

[0085] S21. Calculate the within-class sample variability of the original features of multiple domains of the multiple samples respectively. Specifically, the small within-class sample variability means that the values of the features change less among the within-class samples.

[0086] 。

[0087] In the formula, is the within-class sample variability, represents the value of the th feature of the th sample, is the value of the th feature of the th sample, represents the distance of the sample in the Euclidean space.

[0088] S22. Calculate the between-class sample variability of the original features of multiple domains of the multiple samples respectively. Specifically, the large between-class sample variability means that the values of the features change greatly among different classes of samples.

[0089] 。

[0090] In the formula, is the between-class sample variability, is the variance calculation function, represents the set of values of the th feature in all samples.

[0091] S23. Construct a feature sensitivity index according to the within-class sample variability and the between-class sample variability. Among them, the feature sensitivity index is the ratio of the within-class sample variability and the between-class sample variability.

[0092] 。

[0093] In the formula, is the feature sensitivity index.

[0094] S24. Calculate the mean of the feature sensitivity indices, and filter out the original features whose feature sensitivity indices are less than the mean to obtain the effective feature subset.

[0095] 。

[0096] Wherein, is the effective feature subset, is the original feature set, is the set of feature sensitivity indices of all original features, is the number of original features (in this embodiment, the original features include 16 time-domain features + 6 frequency-domain features).

[0097] For a set of data containing multiple samples and multiple features, high-quality features have the characteristics of small variability among similar samples and large variability among different samples. Therefore, according to the formula of the feature sensitivity index, the smaller the value of the feature sensitivity index, the better the feature.

[0098] Traditional methods will set up screening criteria based on empirical guidelines to select indicators after quantifying the quality of features, which often results in good performance on a certain data set but poor generalization ability.

[0099] Therefore, the embodiment of the present invention constructs a feature sensitivity index by itself for feature screening. The screening criteria are constructed by calculating the feature sensitivity index values of the feature set, and the effective feature subset after screening. In this embodiment, the effective feature subset is the feature whose feature sensitivity index is less than the average value of the feature sensitivity indices in the original feature set, and the effective features are used as the node information of the graph.

[0100] S3. According to the effective feature subset, based on the variability between samples, construct an adjacency matrix of the edges with the effective features as nodes. In this embodiment, one node is an effective feature in step S2. Preferably, step S3 specifically includes steps S31 to S34.

[0101] S31. According to the effective feature subset, calculate the Euclidean distance between different nodes.

[0102] 。

[0103] Wherein, is the Euclidean distance, and are the serial numbers of the nodes, is the serial number of the sample, is the total number of samples, represents the value of the th sample at node value, Indicates Samples at node The value of .

[0104] S32. Calculate the cosine similarity between different nodes according to the valid feature subset.

[0105] .

[0106] In the formula, is the cosine similarity, and is the node number, For Node The value of For Node The value of .

[0107] S33. Calculate the combined similarity between the nodes according to the Euclidean distance and the pre-selected similarity.

[0108] .

[0109] In the formula, is the combined similarity between nodes, For Node and nodes The cosine similarity between For Node and nodes The Euclidean distance between .

[0110] S34: Calculate a similarity threshold according to the combined similarity.

[0111] .

[0112] In the formula, For Node and nodes The combination similarity between is the total number of samples.

[0113] S35. Construct an adjacency matrix of edges with valid features as nodes according to the similarity threshold.

[0114] .

[0115] In the formula, is the edge adjacency matrix.

[0116] The graph attention network aggregates node features through side information for forward propagation, so the construction of side information is particularly important for the diagnosis result. With Node Whether information can be transmitted between them. In the embodiments of the present invention, the Euclidean distance between nodes is calculated based on node features and the cosine similarity for judgment.

[0117] When two nodes are closer, the Euclidean distance between them in space is smaller, and at the same time, their cosine similarity is higher. Therefore, after mapping the Euclidean distance, the greater the combined similarity obtained, the higher the association between the two, and thus the more worthy of information transmission.

[0118] Specifically, whether information can be transmitted is determined by setting a threshold. In the embodiments of the present invention, the mean value is used as the threshold, and two nodes with a combined similarity greater than the mean value are connected to construct an adjacency matrix of edges.

[0119] S4. Construct an operation feature graph of the mechanical equipment according to the effective feature subset and the adjacency matrix. Specifically, when constructing the graph structure, each node represents an effective feature, and the edge is used to represent the association between features.

[0120] S5. Construct a graph attention model, and optimize the cost with a cost function for hyperparameter combination optimization to obtain a diagnostic model to be trained. Preferably, step S5 specifically includes steps S51 to S52.

[0121] S51. Construct a graph attention model (abbreviation: GAT).

[0122] The calculation model of the graph attention model is: .

[0123] In the formula, is the updated feature representation of node , is a non-linear activation function, is the number of neighbor nodes, is the attention coefficient of node to node , is a learnable weight matrix, is the feature vector of node .

[0124] The calculation model of the attention coefficient is: .

[0125] In the formula, is the exponential function, is the LeakyReLU activation function, is a learnable attention mechanism weight vector, denotes transpose, is the eigenvector of node and is the eigenvector of node and denotes the concatenation operation.

[0126] S52. Optimize the hyperparameter combination of the graph attention model through a hyperparameter combination optimization method based on Bayesian optimization to obtain a diagnostic model to be trained. In this embodiment, in order to improve the performance of the graph attention model in fault diagnosis, the present invention designs a hyperparameter combination optimization method for the graph attention model based on Bayesian optimization. By optimizing the learning rate , the number of attention heads , the number of hidden nodes and the dropout rate and other key parameters, ensure that the model has the best diagnostic effect under different datasets and working conditions.

[0127] Preferably, the specific steps of step S52 include steps S521 to step A524.

[0128] S521. Construct the prior distribution of the objective function: In this embodiment, assume that the prior distribution of the objective function is a Gaussian process. The prior distribution of the Gaussian process is defined by the mean function and the covariance function .

[0129] .

[0130] In the formula, is the objective function, representing the performance index (such as the loss function value) of the graph attention model under the hyperparameter combination , represents the Gaussian process, used to model the objective function, is the mean function of the Gaussian process, representing the expected value of the objective function at the hyperparameter combination , is the covariance function (kernel function), used to measure the correlation between two different hyperparameter combinations and (such as the radial basis function RBF), and are two different points in the hyperparameter space (i.e., two different hyperparameter combinations).

[0131] S522. Update the posterior distribution of the Gaussian process based on the prior distribution and historical data.

[0132] .

[0133] In the formula, is the posterior distribution model, is the objective function, is the historical data, including the evaluated hyperparameter combinations and their corresponding objective function values , represents a Gaussian distribution, represents the predicted mean of the objective function at the hyperparameter combination , represents the predicted variance.

[0134] Specifically, collect the preliminary evaluation results of the objective function, denoted as . Update the posterior distribution of the Gaussian process with these data.

[0135] S523. Select an acquisition function.

[0136]

[0137] In the formula, is the expected improvement acquisition function, used to balance exploration (selecting points with high uncertainty) and exploitation (selecting points near the current optimum), is the expected value, calculated based on the posterior distribution of the Gaussian process, represents taking the larger value to screen the improved part, is the objective function, is the currently known optimal objective function value (minimum loss).

[0138] In each iteration, select an acquisition function to determine the next evaluation point. The acquisition function measures the balance between exploration and exploitation. In the embodiment of the present invention, the expected improvement (EI) is used as the acquisition function.

[0139] S524. Based on the evaluation points selected by the acquisition function (i.e., points in the hyperparameter space, representing hyperparameter combinations), then calculate the objective function value and add the result to the historical dataset. Then, update the Gaussian process model with the new data, repeat this process until the stopping criterion is met, obtain the best initial parameter combination of the model, and obtain the diagnostic model to be trained.

[0140] In an alternative embodiment, based on the above process, the initial learning rate , the number of head nodes , the hidden layer factor , the forgetting rate of the four parameters are used as the factors to be optimized. The minimum loss is used as the objective function.

[0141] The minimum loss objective function is as follows: .

[0142] The minimum loss objective function is optimized as: .

[0143] In the formula, is the objective function, represents the loss function, represents the range, represents finding the combination of hyperparameters that minimizes the objective function, represents the graph attention model under the combination of hyperparameters The calculated loss value (in this embodiment: cross - entropy loss).

[0144] In the classification task, cross - entropy is used as the loss function to measure the difference between the probability distribution of the model output and the true label. Finally, the calculation result of the objective function is fed back to the acquisition function for adaptive parameter optimization. By maximizing the EI value, the optimal next set of hyperparameters is selected for training, and the iteration continues until the optimal solution is found to obtain the best initial parameter combination of the model.

[0145] S6. Train the to - be - trained diagnostic model according to the operation characteristic graph of the mechanical equipment to obtain the trained diagnostic model. Specifically, using the feature structure graph to train the graph attention model is a conventional technical means for those skilled in the art, and the present invention will not elaborate on this.

[0146] S7. Obtain the operation signal of the mechanical equipment to be diagnosed, perform pre - processing and extract effective features, and then construct it into an operation characteristic graph. Finally, input the operation characteristic graph of the mechanical equipment to be diagnosed into the trained diagnostic model to obtain the fault diagnosis result of the equipment.

[0147] Specifically, after training the diagnostic model, the operation signal of the equipment obtained in real - time can be constructed into an operation characteristic graph and input into the diagnostic model to obtain the fault diagnosis result of the operation state of the equipment.

[0148] In summary, the overall process of the mechanical equipment fault diagnosis method based on feature graph construction of the present invention is as follows: Collect original signal data: Collect the operation signals such as vibration, temperature, and pressure of the equipment under different working conditions through sensors installed on the mechanical equipment to form a basic data set. Extract time - domain and frequency - domain features: Use time - domain analysis and frequency - domain analysis to extract various feature parameters from the processed signals to construct a high - dimensional feature set including the operation state of the equipment.

[0149] Screen effective features and construct nodes: Based on the feature sensitivity index, perform feature screening on the high-dimensional feature set, retain the features with high discrimination for fault diagnosis, and use these features as the node information of the graph structure.

[0150] Construct edge information: Combine the Euclidean distance and cosine similarity, calculate the combined similarity between the screened feature nodes, determine the edge weights between the nodes, and construct the adjacency matrix of the graph according to the set threshold.

[0151] Construct a feature graph: According to the node information and the adjacency matrix, construct the feature graph of the equipment operation signal.

[0152] Construct the diagnostic model to be trained: Construct a graph attention model (GAT model), and use the Bayesian optimization method to automatically adjust the hyperparameters of the GAT model, such as the learning rate, the number of attention heads, the number of hidden nodes, and the dropout rate, to improve the diagnostic accuracy and robustness of the model.

[0153] Train the diagnostic model to be trained: Use the healthy data and fault data to train the optimized GAT model so that it can identify different fault types and fault severities.

[0154] Use the trained model for prediction: Construct a graph structure according to the signal data to be predicted, then input the graph structure into the trained model to obtain the fault diagnosis result of the mechanical equipment. And according to the output result of the model, analyze the health status of the equipment, identify the fault type, fault location and its severity, form a fault diagnosis report, and provide decision support for equipment maintenance. Specifically, input the constructed graph structure into the GAT model, aggregate the node features through the attention mechanism, and capture the complex correlation relationships between the features.

[0155] Specifically, the existing graph-based equipment fault diagnosis methods have the following problems: 1. The selection of node features often relies on experience and may not be able to comprehensively capture all relevant information. 2. The edge information cannot fully reflect the complex dependency relationships between equipment components, resulting in unsatisfactory performance of the model when dealing with specific fault modes. 3. The hyperparameter settings of the graph attention network are inappropriate, resulting in the model being unable to effectively learn the complex relationships between features, or overfitting to the training data, affecting the accuracy and robustness of the diagnosis. Therefore, how to optimize the construction of node features and edge information, and improve the performance of the graph neural network by adjusting hyperparameters has become the key direction for improving the fault diagnosis method. The present invention proposes a mechanical equipment fault diagnosis method based on feature graph construction for the graph construction problem and model parameter combination problem in the above diagnostic process.

[0156] An embodiment of the present invention designs a method for constructing a feature map, which constructs a graph structure by using time-frequency feature screening and similarity evaluation, so as to more accurately reflect the fault correlation relationship of equipment and improve the diagnostic effect. In addition, through Bayesian optimization, automatic adjustment of hyperparameters is realized, avoiding the problems of long time consumption and easy to fall into local optimum of traditional parameter tuning methods, and enhancing the robustness and adaptability of the model. In summary, the mechanical equipment fault diagnosis method based on feature map construction in the embodiment of the present invention can effectively capture the complex relationships among various components of mechanical equipment and improve the accuracy of fault diagnosis.

[0157] It can be understood that the diagnostic device can be an electronic device with computing performance such as a portable notebook computer, a desktop computer, a server, a smart phone or a tablet computer.

[0158] Embodiment 2: The present invention provides a mechanical equipment fault diagnosis device based on feature map construction, which includes a feature extraction module, a feature screening module, a graph edge construction module, a graph construction module, a model construction module, a model training module and a fault diagnosis module.

[0159] The feature extraction module is used to obtain multiple samples of the operating signal and extract the original features in multiple domains for each sample respectively.

[0160] The feature screening module is used to screen out an effective feature subset according to the original features in multiple domains of the multiple samples, taking the importance of each feature for data classification as the criterion.

[0161] The graph edge construction module is used to construct an adjacency matrix of edges with effective features as nodes based on the variability between samples according to the effective feature subset.

[0162] The graph construction module is used to construct an operating feature map of the mechanical equipment according to the effective feature subset and the adjacency matrix.

[0163] The model construction module is used to construct a graph attention model and optimize the cost with a cost function for hyperparameter combination optimization to obtain a diagnostic model to be trained.

[0164] The model training module is used to train the diagnostic model to be trained according to the operating feature map of the mechanical equipment to obtain a trained diagnostic model.

[0165] The fault diagnosis module is used to obtain the operating signal of the mechanical equipment to be diagnosed, perform preprocessing and extract effective features, and then construct an operating feature map. Finally, the operating feature map of the mechanical equipment to be diagnosed is input into the trained diagnostic model to obtain the fault diagnosis result of the equipment.

[0166] Embodiment 3. The present invention provides a mechanical equipment fault diagnosis device based on feature map construction, which includes a processor, a memory, and a computer program stored in the memory. The computer program can be executed by the processor to implement a mechanical equipment fault diagnosis method based on feature map construction as described in any paragraph of Embodiment 1.

[0167] Embodiment 4. The present invention provides a computer-readable storage medium. The computer-readable storage medium includes a stored computer program, wherein when the computer program runs, it controls the device where the computer-readable storage medium is located to execute a mechanical equipment fault diagnosis method based on feature map construction as described in any paragraph of Embodiment 1.

[0168] In several embodiments provided by the embodiments of the present invention, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device and method embodiments described above are only illustrative. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architectures, functions, and operations of devices, methods, and computer program products according to multiple embodiments of the present invention. In this regard, each block in the flowchart or block diagram can represent a module, a program segment, or a part of code, and the module, program segment, or part of code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than marked in the accompanying drawings. For example, two consecutive blocks can actually be executed substantially in parallel, and they can sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, as well as the combination of blocks in the block diagram and / or flowchart, can be implemented by a dedicated hardware-based system for performing the specified functions or actions, or can be implemented by a combination of dedicated hardware and computer instructions.

[0169] In addition, the functional modules in each embodiment of the present invention can be integrated together to form an independent part, or each module can exist alone, or two or more modules can be integrated to form an independent part.

[0170] When the above-mentioned function is implemented in the form of a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, an electronic device, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories, random access memories, magnetic disks, or optical discs that can store program codes. It should be noted that in this article, the terms "including", "comprising", or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article, or device. Without further limitation, an element defined by the statement "including one..." does not exclude the existence of another identical element in the process, method, article, or device including the said element.

[0171] The terms used in the embodiments of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The singular forms "a", "the", and "said" used in the embodiments of the present invention and the appended claims are also intended to include the plural forms unless the context clearly dictates otherwise.

[0172] It should be understood that the term "and / or" used herein is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone. In addition, the character " / " in this article generally represents an "or" relationship between the associated objects before and after.

[0173] Depending on the context, the word "if" as used herein can be interpreted as "when", "while", "in response to determining", or "in response to detecting". Similarly, depending on the context, the phrase "if determined" or "if detected (stated condition or event)" can be interpreted as "when determined", "in response to determining", "when detected (stated condition or event)", or "in response to detecting (stated condition or event)".

[0174] The "first / second" mentioned in the embodiments is only used to distinguish similar objects and does not represent a specific order for the objects. It can be understood that the "first / second" can be interchanged in a specific order or sequence when permitted. It should be understood that the objects distinguished by the "first / second" can be interchanged under appropriate circumstances so that the embodiments described herein can be implemented in an order other than those illustrated or described herein.

[0175] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A mechanical equipment fault diagnosis method based on feature map construction, characterized in that Including: Obtain multiple samples of the operating signal, and extract the original features in multiple domains for each sample respectively; According to the original features in multiple domains of the multiple samples, taking the importance of each feature for data classification as the criterion, filter out the effective feature subset; According to the effective feature subset, based on the variability between samples, construct an adjacency matrix of edges with effective features as nodes; According to the effective feature subset and the adjacency matrix, construct an operating feature map of the mechanical equipment; Construct a graph attention model, and optimize the cost with a cost function for hyperparameter combination optimization to obtain a diagnostic model to be trained; According to the operating feature map of the mechanical equipment, train the diagnostic model to be trained to obtain a trained diagnostic model; Obtain the operating signal of the mechanical equipment to be diagnosed, perform preprocessing and extract effective features, and then construct an operating feature map; finally, input the operating feature map of the mechanical equipment to be diagnosed into the trained diagnostic model to obtain the fault diagnosis result of the equipment.

2. The mechanical equipment fault diagnosis method based on feature map construction according to claim 1, characterized in that According to the original features in multiple domains of the multiple samples, taking the importance of each feature for data classification as the criterion, filter out the effective feature subset, specifically including: Calculate the within-class sample variability of the original features in multiple domains of the multiple samples respectively; among them, ; in the formula, is the within-class sample variability, represents the -th sample's -th feature value, is the -th sample's -th feature value, represents the distance of the sample in the Euclidean space; Calculate the heterogeneous sample variability of the original features of multiple domains of the multiple samples respectively; wherein, ; in the formula, is the heterogeneous sample variability, is the variance calculation function, represents the set of values of the feature in all samples; Construct a feature sensitivity index based on the within-class sample variability and the between-class sample variability; wherein, ; where, is the feature sensitivity index; Calculate the mean of the feature sensitivity indices, and filter out the original features whose feature sensitivity indices are less than the mean to obtain the effective feature subset; where, ; In the formula, is the effective feature subset, is the original feature set, is the set of feature sensitivity indices of all original features, is the number of original features.

3. A mechanical equipment fault diagnosis method based on feature map construction according to claim 1, characterized in that, According to the effective feature subset, based on the variability between samples, construct an adjacency matrix of edges with effective features as nodes, specifically including: Calculate the Euclidean distance between different nodes according to the effective feature subset; where, ; In the formula, is the Euclidean distance, and are the serial numbers of the nodes, is the serial number of the sample, is the total number of samples, represents the value of the th sample at node ; represents the value of the th sample at node ; Calculate the cosine similarity between different nodes according to the effective feature subset; where, ; In the formula, is the cosine similarity, and are the serial numbers of the nodes, is the value of node is the value of node According to the Euclidean distance and the pre-selected similarity, the combined similarity between the nodes is calculated; wherein, ; In the formula, is the combined similarity between nodes, For Node and nodes The cosine similarity between is the Euclidean distance between node and node ; Calculate a similarity threshold according to the combined similarity; wherein, ; in the formula, is the node and the node the combined similarity between them, is the total number of samples; Construct an adjacency matrix of edges with effective features as nodes according to the similarity threshold and the effective feature subset; where ; In the formula, is the adjacency matrix of the edges.

4. A mechanical equipment fault diagnosis method based on feature map construction according to any one of claims 1 to 3, characterized in that Construct a graph attention model, and optimize the cost with a cost function for hyperparameter combination optimization to obtain a diagnostic model to be trained, specifically including: Construct a graph attention model; Through a hyperparameter combination optimization method based on Bayesian optimization, optimize the hyperparameter combination of the graph attention model to obtain a diagnostic model to be trained.

5. A mechanical equipment fault diagnosis method based on feature map construction according to claim 4, characterized in that Through a hyperparameter combination optimization method based on Bayesian optimization, optimize the hyperparameter combination of the graph attention model to obtain a diagnostic model to be trained, specifically including: Construct the prior distribution of the objective function: ; In the formula, is the objective function, represents the Gaussian process, is the mean function of the Gaussian process, is the covariance function, used to measure the correlation between two different hyperparameter combinations and ; Based on the prior distribution and historical data, update the posterior distribution of the Gaussian process; ； In the formula, is the posterior distribution model, is the objective function, is the historical data, represents the Gaussian distribution, represents the predicted mean of the objective function at the hyperparameter combination and represents the predicted variance; Select the acquisition function; ; where, is the expected improved acquisition function, is the expected value, represents taking the larger value, is the objective function, is the currently known optimal objective function value; Based on the evaluation points selected by the acquisition function, then calculate the objective function value and add the result to the historical dataset; then, update the Gaussian process model with the new data, repeat this process until the stopping criterion is met, obtain the best initial parameter combination of the model, and obtain a diagnostic model to be trained.

6. A mechanical equipment fault diagnosis method based on feature map construction according to any one of claims 1 to 3, characterized in that Obtain multiple samples of the operating signal, and extract the original features in multiple domains for each sample respectively, specifically including: Collect the operating signals of the equipment under different working conditions as samples through sensors installed on the mechanical equipment to form a basic dataset; Through time-domain analysis and frequency-domain analysis, extract the original feature parameters in multiple domains from the preprocessed basic dataset to construct a high-dimensional feature set containing the operating state of the equipment; among them, the original features include: mean value, root mean square amplitude, maximum value, minimum value, absolute mean value, root mean square value, standard deviation, variance, peak value, peak-to-peak value, kurtosis, skewness, impulse factor, margin factor, waveform factor, amplitude factor, center frequency, average frequency, root mean square frequency, mean square frequency, frequency variance and frequency standard deviation.

7. A mechanical equipment fault diagnosis device based on feature map construction, characterized in that, Including: A feature extraction module, which is used to obtain multiple samples of the operating signal and extract the original features in multiple domains for each sample respectively; A feature screening module, configured to screen out an effective feature subset according to the original features of multiple domains of the multiple samples, taking the importance of each feature for data classification as a criterion; A graph edge construction module, configured to construct an adjacency matrix of edges with effective features as nodes based on the variability between samples according to the effective feature subset; A graph construction module, configured to construct an operation feature graph of a mechanical device according to the effective feature subset and the adjacency matrix; A model construction module, configured to construct a graph attention model, and optimize the cost with a cost function for hyperparameter combination optimization to obtain a diagnostic model to be trained; A model training module, configured to train the diagnostic model to be trained according to the operation feature graph of the mechanical device to obtain a trained diagnostic model; A fault diagnosis module, configured to obtain the operation signal of the mechanical device to be diagnosed, perform preprocessing and extract effective features, and then construct an operation feature graph; finally, input the operation feature graph of the mechanical device to be diagnosed into the trained diagnostic model to obtain the fault diagnosis result of the device.

8. A mechanical equipment fault diagnosis device based on the construction of a feature map, characterized in that, It includes a processor, a memory, and a computer program stored in the memory; the computer program can be executed by the processor to implement a mechanical device fault diagnosis method based on feature graph construction according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein when the computer program runs, it controls the device where the computer-readable storage medium is located to execute a mechanical device fault diagnosis method based on feature graph construction according to any one of claims 1 to 6.

Images