Rotating machinery fault diagnosis method based on adaptive feature optimization and integrated model

Abstract

The application provides a rotating machinery fault diagnosis method based on adaptive feature optimization and integrated model, and belongs to the field of rotating machinery equipment fault diagnosis, which comprises a data processing module, a feature screening and combination module, a fault category evaluation module, a feature enhancement generation module, a performance monitoring and model protection module and a fault diagnosis module. The basic input of the diagnosis model is obtained by initially processing data and screening features, and after evaluation by the fault category evaluation module, if the performance does not meet the requirements, the feature set is adjusted by the enhanced feature generation module and the performance monitoring and model protection module, and the adjusted features re-enter the fault category evaluation module for re-evaluation. The process can be repeated, optimized in real time, until the diagnosis performance reaches the preset threshold or the iteration termination condition. The method can dynamically optimize features and models according to compound faults and complex working conditions, improve the accuracy and robustness of fault diagnosis, and has a wide application prospect.

Description

Technical Field

[0001] This invention belongs to the field of fault diagnosis of rotating machinery equipment, and specifically relates to a fault diagnosis method for rotating machinery based on adaptive feature optimization and integrated model. Background Technology

[0002] With the increasing complexity and automation of industrial equipment, equipment fault diagnosis plays a crucial role in ensuring system safety and reliability. Vibration signal analysis is an important means of fault detection in rotating machinery (such as bearings and gearboxes), typically involving extracting feature parameters from vibration signals and performing pattern recognition to determine the fault type and severity. Traditional methods mostly determine the feature set and a single classification model (such as RBF or SVM) once during the system design phase, and then apply them in a fixed manner for a long time. However, under complex operating conditions or compound fault scenarios, the pre-fixed features often fail to fully capture key information, leading to a decrease in diagnostic accuracy. In existing technologies, some feature selection methods, such as filtering, wrapping, and embedding, are used to select important features from a large number of features to improve model performance, but these methods are usually static and lack the ability to dynamically adjust during the diagnostic process. When new fault modes or multiple faults occur concurrently, the original features may be insufficient to distinguish different fault categories, and there is a lack of a mechanism to automatically add new feature dimensions to cope with this. At the same time, blindly adding features may lead to redundancy or overfitting, or even degradation of the diagnostic model. Existing methods lack corresponding performance protection and rollback mechanisms to avoid this problem. In summary, existing systems lack performance protection and fallback mechanisms to avoid performance degradation, and there is an urgent need for a diagnostic framework that takes adaptive feature optimization as its core and combines ensemble learning to improve robustness. Summary of the Invention

[0003] To address the aforementioned issues, this invention proposes a rotating machinery fault diagnosis method based on adaptive feature optimization and an integrated model. This method dynamically optimizes the feature set through a closed-loop feedback mechanism and employs an integrated diagnostic model for fault determination, thereby improving diagnostic accuracy and stability.

[0004] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0005] The rotating machinery fault diagnosis method based on adaptive feature optimization and ensemble model comprises six functional modules: data processing, feature selection and combination, fault category assessment, feature enhancement and generation, performance monitoring and model protection, and fault diagnosis. These modules are interconnected through data channels, forming a closed-loop iterative process that enables the system to adaptively optimize the diagnostic process based on diagnostic results and performance feedback. Specifically:

[0006] Data processing module: This module preprocesses the acquired raw vibration signals, dividing the continuous vibration data into time windows of fixed length to extract features. Preferably, a sliding time window of 1024 points is used to segment the vibration signal and extract time-domain and frequency-domain features.

[0007] Feature selection and combination module: This module is used to select a subset of features with high relevance to fault diagnosis from the large number of initial features extracted by the data processing module. It integrates three feature selection strategies—filtering, wrapping, and embedding—and uses a voting fusion mechanism to combine the selection results of each algorithm, ultimately selecting the feature subset with the highest diagnostic value.

[0008] Fault Category Assessment Module: This module performs preliminary fault type determination and performance evaluation based on the currently selected feature set. It monitors the performance metrics (accuracy, recall, F1 score, etc.) output by the fault diagnosis module, compares the performance with preset thresholds to determine the reliability of the diagnosis, and provides error analysis based on feature importance as a basis for feature enhancement. For example, if the F1 score falls below the predetermined threshold, it indicates that the current feature set may be insufficient to support accurate fault identification, and new features need to be introduced.

[0009] Feature Enhancement Generation Module: This module introduces new features to enrich the feature space when existing features are insufficient to adequately identify faults. It determines the feature generation scheme based on information provided by the fault category assessment module. Feature enhancement factors are generated based on the feature importance analysis results of SHAP values. The newly generated features are then fused with the original features to enhance the subsequent diagnostic model's ability to discriminate difficult samples. Once new features are generated, this module submits them to the fault category assessment module for evaluation.

[0010] The performance monitoring and model protection module monitors the impact of new feature introductions on diagnostic performance and performs a rollback operation when performance is not improved or even degraded. Specifically, after the supplementary feature generation module adds new features and retrains the fault diagnosis model, the rollback mechanism module compares the performance metrics of the old and new models. If the F1 score of the model does not meet the improvement requirements after adding the new feature, or even decreases, a rollback is triggered, restoring the feature set and model to their state before the feature was introduced, thus avoiding interference from invalid features. Simultaneously, the rollback mechanism module can also limit the number of iterations or the size of the feature set, preventing unlimited feature additions that could lead to excessive model complexity or overfitting. This module essentially provides performance protection logic for the entire optimization process, ensuring that feature optimization proceeds in the direction of improving diagnostic performance.

[0011] The fault diagnosis module consists of parallel diagnostic classification models (such as support vector machines, random forests, gradient boosting trees, and shallow neural networks). Each diagnostic classification model independently judges the same input feature and generates its own category probability distribution. The fusion unit adopts a weighted soft voting strategy to combine the outputs of each diagnostic classification model according to the weights learned in advance or online, and calculates the final fault category.

[0012] By organically combining the above modules, this invention forms a closed-loop fault diagnosis process: initially, the basic input for the diagnostic model is obtained through data processing and feature selection. After evaluation by the fault category assessment module, if the performance does not meet the requirements, the feature set is adjusted by the enhanced feature generation module and the performance monitoring and model protection module. The adjusted features are then re-entered into the fault category assessment module for evaluation again. This process can be repeated cyclically and optimized in real time until the diagnostic performance reaches a preset threshold or the iteration terminates.

[0013] The advantages and positive effects of this invention are as follows: Compared with existing technologies, the framework features a highly modular design, with each functional module being independent and decoupled, facilitating adjustments and expansions for different applications; it introduces a feedback mechanism for diagnostic results, enabling dynamic optimization of the feature set and selection of the optimal model based on model performance, significantly improving diagnostic accuracy under complex fault conditions. Simultaneously, this module provides a performance protection rollback mechanism to prevent invalid features or improper adjustments from negatively impacting diagnostic results, thereby ensuring the reliability and stability of the diagnostic process; the entire framework achieves automated feature optimization and model selection, reducing manual intervention and enhancing the intelligence level of equipment health management. Attached Figure Description

[0014] Figure 1 This is a flowchart of a rotating machinery fault diagnosis method based on adaptive feature optimization and ensemble model. Detailed Implementation

[0015] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0016] like Figure 1 As shown, the rotating machinery fault diagnosis method based on adaptive feature optimization and ensemble model includes the following six functional modules: data processing module, feature selection and combination module, fault category assessment module, feature enhancement and generation module, performance monitoring and model protection module, and fault diagnosis module. These modules are interconnected through data channels to form a closed-loop iterative process, enabling the system to adaptively optimize the diagnostic process based on diagnostic results and performance feedback. Specifically:

[0017] Data processing module: This module preprocesses the acquired raw vibration signals, dividing the continuous vibration data into time windows of fixed length to extract features. Preferably, a sliding time window of 1024 points is used to segment the vibration signal and extract time-domain and frequency-domain features.

[0018] Feature selection and combination module: This module is used to select a subset of features with high relevance to fault diagnosis from the large number of initial features extracted by the data processing module. It integrates multiple feature selection strategies, including filtering, wrapping, and embedding, and uses a voting fusion mechanism to combine the selection results of each algorithm, ultimately selecting the feature subset with the highest diagnostic value.

[0019] Fault Category Assessment Module: This module performs preliminary fault type determination and performance evaluation based on the currently selected feature set. It monitors the performance metrics (accuracy, recall, F1 score, etc.) output by the fault diagnosis module, compares the performance with preset thresholds to determine the reliability of the diagnosis, and provides error analysis based on feature importance as a basis for feature enhancement. For example, if key metrics such as the F1 score fall below the predetermined threshold, it indicates that the current feature set may be insufficient to support accurate fault identification, and new features need to be introduced.

[0020] Feature Enhancement Generation Module: This module introduces new features to enrich the feature space when existing features are insufficient to adequately identify faults. It determines the feature generation scheme based on information provided by the fault category assessment module. Feature enhancement factors are generated based on the feature importance analysis results of SHAP values. The newly generated features are then fused with the original features to enhance the subsequent diagnostic model's ability to discriminate difficult samples. Once new features are generated, this module submits them to the fault category assessment module for evaluation.

[0021] The performance monitoring and model protection module monitors the impact of new feature introductions on diagnostic performance and performs a rollback operation when performance is not improved or even degraded. Specifically, after the supplementary feature generation module adds new features and retrains the fault diagnosis model, the rollback mechanism module compares the performance metrics of the old and new models. If the F1 score of the model does not meet the improvement requirements after adding the new feature, or even decreases, a rollback is triggered, restoring the feature set and model to their state before the feature was introduced, thus avoiding interference from invalid features. Simultaneously, the rollback mechanism module can also limit the number of iterations or the size of the feature set, preventing unlimited feature additions that could lead to excessive model complexity or overfitting. This module essentially provides performance protection logic for the entire optimization process, ensuring that feature optimization proceeds in the direction of improving diagnostic performance.

[0022] The fault diagnosis module consists of parallel diagnostic classification models (such as support vector machines, random forests, gradient boosting trees, shallow neural networks, etc.). Each diagnostic classification model independently judges the same input feature and generates its own category probability distribution. The fusion unit adopts a weighted soft voting strategy to combine the outputs of each diagnostic classification model according to the weights learned in advance or online, and calculates the final fault category.

[0023] By organically combining the above modules, this invention forms a closed-loop fault diagnosis process: initially, the basic input for the diagnostic model is obtained through data processing and feature selection. After evaluation by the fault category assessment module, if the performance does not meet the requirements, the feature set is adjusted by the enhanced feature generation module and the performance monitoring and model protection module. The adjusted features are then re-entered into the fault category assessment module for evaluation again. This process can be repeated cyclically and optimized in real time until the diagnostic performance reaches a preset threshold or the iteration terminates.

[0024] Data Processing Module: This module is responsible for preprocessing and feature extraction of the vibration signals acquired online by the rotating machinery to generate the initial time-frequency domain feature set required for subsequent diagnosis. Its specific implementation is as follows:

[0025] Step 1: Use a length of The sliding window is segmented, with the window shift preferably having 50% overlap (i.e., 512 points), to obtain discrete signal segments. .

[0026] Step 2: For each signal segment The time-domain features listed in formulas (1)-(12) are calculated to form a time-domain feature vector. .

[0027] Step 3: Use the same signal Performing a discrete Fourier transform yields And calculate the power spectrum Its total energy Then, frequency domain feature vectors are extracted according to formulas (13)-(16). .

[0028] Step 4: Convert the time-domain feature vector With frequency domain eigenvectors Cascaded, forming the initial feature vector of this signal segment. Repeat the above operation for all signal segments to construct the initial feature pool; This module will then... The output is sent to the "Feature Filtering and Combination Module" for further processing.

[0029] (1)

[0030] (2)

[0031] (3)

[0032] (4)

[0033] (5)

[0034] (6)

[0035] (7)

[0036] (8)

[0037] (9)

[0038] (10)

[0039] (11)

[0040] (12)

[0041] (13)

[0042] (14)

[0043] (15)

[0044] , (16)

[0045] Feature Filtering and Combination Module: This module employs three strategies: filtering, wrapping, and embedded methods. The features are filtered and fused through voting to form a preliminary feature subset. Specifically, it includes:

[0046] Step 1: Input the initial feature set Target fault label set .

[0047] Step 2: Filtering

[0048] (1) Pearson correlation coefficient: According to absolute value Before selecting in descending order One feature;

[0049] (2) Mutual information method: ,according to Before preserving descending order One characteristic.

[0050] Step 3: Request for Evidence (RFE)

[0051] Using random forest as the base model, the iterative steps are as follows:

[0052] (1) Train a random forest using the current feature set;

[0053] (2) Calculate the importance of the feature based on the reduction in impurity (Gini) at node splitting;

[0054] (3) Eliminate the least important features in each round until only the remaining features remain. indivual.

[0055] Step 4: Embedded Filtering: according to Before the election item.

[0056] Step 5: Voting Integration

[0057] (1) Each method outputs its filtering result list;

[0058] (2) Count the votes for each feature (1 point is added for each occurrence);

[0059] (3) The one with the highest number of votes Features (preferred) )constitute .

[0060] Step 6: Put The data is passed to the "Fault Category Assessment Module" for preliminary diagnosis.

[0061] Fault Category Assessment Module: Based on the preliminary screening features mentioned above, and using the fault diagnosis module for assessment, the module provides the assessment results. The specific steps are as follows:

[0062] Step 1: Calculate the macro average based on the validation set ;

[0063] Step 2: If It directly outputs the diagnostic results;

[0064] Step 3: Otherwise, filter based on the confusion matrix. The list of categories is used as the categories to be enhanced, and the feature enhancement generation module is triggered.

[0065] Feature enhancement generation module: for fault categories with low F1 scores ( Feature importance analysis is performed using SHAP values, and the specific steps are as follows.

[0066] Step 1: Input the set of fault categories that need enhancement ,in Indicate category F1 score; current preliminary feature subset ; SHAP value matrix of the ensemble model trained on the validation set , , .

[0067] Step 2: Backup Feature Pool Preparation: Maintain a candidate feature library in the system, including but not limited to: wavelet packet sub-band energy; main peaks of the envelope spectrum; short-time Fourier transform (STFT) band energy; variational mode decomposition (VMD) mode energy and other statistical features.

[0068] Step 3: SHAP targeted selection, for each fault category :

[0069] (1) Set the current feature set All features according to Sort in descending order and select the first few. The most critical feature set ;

[0070] (2) In the spare feature pool In the process, each candidate feature is calculated. and The correlation (such as Pearson correlation coefficient or mutual information) is taken from the highest correlation values. (Preferred) ) features, to obtain the incremental feature set .

[0071] Step 4: Merge the incremental features of all weak classes: ,right Each feature is normalized to zero-mean and unit variance to obtain This forms an enhanced feature set: .

[0072] Step 5: Output the enhanced feature set Return to the "Feature Selection and Combination Module" or directly transfer to the "Fault Category Evaluation Module" for retraining and evaluation.

[0073] Performance Monitoring and Model Protection Module: This module uses an enhanced feature set in the "Fault Category Assessment Module". After training and evaluating the ensemble model, the system is responsible for comparing the performance of the new and old models and deciding whether to roll back or continue iterating based on the results. The specific monitoring steps are as follows.

[0074] Step 1: Input

[0075] (1) Preliminary optimal state: Feature set: Corresponding ensemble model: Macro average F1: ;

[0076] (2) Current enhancement state: Feature set: ;based on The trained ensemble model: .

[0077] Step 2: Performance Evaluation

[0078] (1) Calculate the macro-average F1 of the two models on the same validation set: , ;

[0079] (2) Calculate performance gain: .

[0080] Step 3: Rollback and Update Logic

[0081] (1) If (The new model did not bring any improvement)

[0082] 1) Rollback operation: , , ;

[0083] 2) Set an end flag to terminate feature addition and model iteration, and proceed to the output module;

[0084] (2) Otherwise (Performance improvements in the new model)

[0085] 1) Update to optimal state: , ,

[0086] 2) Threshold determination: If If the maximum number of iterations or the maximum dimension is reached, set the end flag and enter the output module; otherwise, if the number of iterations or the feature dimension has not reached the upper limit, return to the "Feature Enhancement Generation Module" to continue the next round of iterations; otherwise, if the maximum number of iterations or the maximum dimension is reached, set the end flag and enter the output module.

[0087] Step 4: Iteration Limit and Protection: Maximum Number of Iterations Used to avoid infinite loops; maximum feature dimension To prevent overfitting and computational bottlenecks caused by too many features. If any upper limit is triggered and performance still fails to meet the target, then the current upper limit is used. , Output the optimal result.

[0088] The fault diagnosis module includes the following steps:

[0089] Step 1: Diagnostic classification model training: For diagnostic classification models such as Support Vector Machine, Random Forest, Gradient Boosting Tree, and Shallow Neural Network, train them on the validation set to obtain classification probabilities. and F1 score .

[0090] Step 2: Weight Normalization: Calculate the weights .

[0091] Step 3: Soft Voting Fusion: Fusion Probability and with Determine the fault type and output the diagnostic accuracy.

[0092] In this embodiment, the fault diagnosis of rotating machinery (rolling bearings) is used as an application example. The Case Western Reserve University (CWRU) bearing fault public dataset is selected, and the specific data selection is shown in Table 1.

[0093] Table 1 Data Table

[0094] 1 Normal — — 99.mat Baseline data, 2HP, 1750rpm 2 Inner ring fault 0.007″ (IR007) 0.007″ Inner circle 107.mat 2HP, 1750rpm 3 Rolling element failure 0.007″ (BF007) 0.007″ Rolling body 120.mat 2HP, 1750rpm 4 Outer ring fault 0.007″ (OR007) 0.007″ Outer circle @6:00 132.mat 2HP, 1750rpm 5 Inner ring fault 0.014″ (IR014) 0.014″ Inner circle 171.mat 2HP, 1750rpm 6 Rolling element failure 0.014″ (BF014) 0.014″ Rolling body 187.mat 2HP, 1750rpm 7 Outer ring fault 0.014″ (OR014) 0.014″ Outer circle @6:00 199.mat 2HP, 1750rpm 8 Inner ring fault 0.021″ (IR021) 0.021″ Inner circle 211.mat 2HP, 1750rpm 9 Rolling element failure 0.021″ (BF021) 0.021″ Rolling body 224.mat 2HP, 1750rpm 10 Outer ring fault 0.021″ (OR021) 0.021″ Outer circle @6:00 236.mat 2HP, 1750rpm

[0095] Data processing module: Perform the process on all sample segments, using a sliding window of length 1024 points for segmentation, with a preferred window overlap of 50% (i.e., 512 points), resulting in 3074 sample segments. Calculate the time-domain features (12) and frequency-domain features (4) according to formulas (1)-(16); construct the initial output feature pool. .

[0096] Feature filtering and combination module: targeting The filtering method (mutual information), wrapping method (random forest), and embedding method (Lasso regression) are combined with voting to select the top 10 features to form an initial feature subset. .

[0097] Fault Category Assessment Module: Inputting the

S6 Fault Diagnosis Module

[0098] Table 2 Diagnostic performance indicators for each category

[0099] 1 0.98233 0.96864 0.97544 2 0.98611 0.91026 0.94667 3 0.81429 0.89062 0.85075 4 1 1 1 5 0.95775 0.95775 0.95775 6 0.81944 0.89394 0.85507 7 0.83099 0.80822 0.81944 8 1 1 1 9 0.77465 0.74324 0.75862 10 0.95775 1 0.97842

[0100] Feature enhancement generation module: for Within each category: Select by SHAP value The first 6 features are selected, and the 4 features with the highest correlation are selected from the backup feature pool (energy of each subband of the wavelet packet and the main peak of the envelope); Each feature is normalized to zero-mean and unit variance to obtain : {'WP3_E_band1'}, {'WP3_E_band2'}, {'EnvE_1_2kHz'}, {'EnvE_2_3kHz'}, and with fusion The feature dimension changed from 10 to 14.

[0101] Performance monitoring and model protection module: based on Train the ensemble model and compute new As shown in Table 3, it can be seen that... and all categories Enhancement complete. Utilizing the optimal feature set. ; Optimal ensemble model Output fault category; overall diagnostic accuracy: 98.2%.

[0102] Table 3 F1 Tables for Each Category

[0103] <![CDATA[Category 1: F1 best =0.9754]]> <![CDATA[Category 1: F1 cur =1.0000]]> <![CDATA[Category 2: F1 best =0.9467]]> <![CDATA[Category 2: F1 cur =1.0000]]> <![CDATA[Category 3: F1 best =0.8507]]> <![CDATA[Category 3: F1 cur =1.0000]]> <![CDATA[Category 4: F1 best =1.0000]]> <![CDATA[Category 4: F1 cur =1.0000]]> <![CDATA[Category 5: F1 best =0.9577]]> <![CDATA[Category 5: F1 cur =0.9930]]> <![CDATA[Category 6: F1 best =0.8551]]> <![CDATA[Class 6: F1 cur =0.9796]]> <![CDATA[Class 7: F1 best =0.8194]]> <![CDATA[Class 7: F1 cur =0.9929]]> <![CDATA[Class 8: F1 best =1.0000]]> <![CDATA[Class 8: F1 cur =1.0000]]> <![CDATA[Class 9: F1 best =0.7586]]> <![CDATA[Class 9: F1 cur =0.9787]]> <![CDATA[Category 10: F1 best =0.9784]]> <![CDATA[Class 10: F1 cur =0.9857]]>

[0104] This embodiment fully demonstrates the feasibility and effectiveness of the present invention. In practical applications, the feature set and base learner combination can be adjusted according to specific working conditions and requirements. The adaptive architecture of the present invention has good flexibility and robustness, and is suitable for intelligent health management of various equipment such as rotating machinery.

[0105] It should be emphasized that the embodiments described in this invention are illustrative rather than limiting. Therefore, this invention includes, but is not limited to, the embodiments described in the specific implementation. Any other implementations derived by those skilled in the art based on the technical solutions of this invention are also within the scope of protection of this invention.

Claims

1. A method for fault diagnosis of rotating machinery based on adaptive feature optimization and ensemble model, characterized in that, This method consists of six main functional modules: data processing, feature selection and combination, fault category assessment, feature enhancement and generation, performance monitoring and model protection, and fault diagnosis. These modules are interconnected via data channels, forming a closed-loop iterative process. This allows the system to adaptively optimize the diagnostic process based on diagnostic results and performance feedback. Specifically: Data processing module: used to preprocess the acquired raw vibration signal, divide the continuous vibration data into time windows of fixed length to extract features; use a sliding time window of 1024 points to segment the vibration signal and extract time domain features and frequency domain features; Feature selection and combination module: This module is used to select the subset of features with high relevance to fault diagnosis from a large number of initial features extracted from the data processing module. It integrates multiple feature selection strategies, including filtering, wrapping, and embedding, and adopts a voting fusion mechanism to integrate the selection results of each algorithm to select the feature subset with the most diagnostic value. Fault Category Assessment Module: Used to make preliminary judgments and performance evaluations on fault types based on the currently selected feature set, monitor the performance indicators output by the fault diagnosis module, including accuracy, recall, and F1 score, compare the performance with preset thresholds to determine whether the diagnosis is reliable; and provide error analysis based on feature importance as a basis for feature enhancement. Feature enhancement generation module: When existing features are deemed insufficient to fully identify faults, new features are introduced to enrich the feature space; a feature generation scheme is determined based on information provided by the fault category evaluation module; for fault categories with F1 scores below a preset threshold, key features in the current feature set are determined based on SHAP value analysis, and the correlation between each candidate feature and the key feature is calculated in the backup feature pool, selecting the candidate feature with the highest correlation as the incremental feature; the incremental feature is fused with the original feature to obtain an enhanced feature set, thereby enhancing the ability of the subsequent diagnostic model to distinguish difficult samples; once a new feature is generated, it is sent to the fault category evaluation module for evaluation; The performance monitoring and model protection module monitors the impact of new feature introductions on diagnostic performance and performs a rollback operation when performance is found to be unimproved or decreased. Specifically, after the supplementary feature generation module adds new features and retrains the fault diagnosis model, the rollback mechanism module compares the performance metrics of the old and new models. If the new feature results in a performance gain greater than zero but not reaching a preset threshold, it triggers a return to the feature enhancement generation module to continue the next round of feature addition. If it is found that the F1 score of the model does not meet the improvement requirements or even decreases after adding new features, a rollback is triggered, restoring the feature set and model to the state before the introduction of the new features to avoid interference from invalid features. At the same time, the rollback mechanism module can limit the number of iterations or the size of the feature set, thereby preventing the unlimited increase of features from causing excessive model complexity or overfitting. The fault diagnosis module consists of several diagnostic classification models operating in parallel, including support vector machines, random forests, gradient boosting trees, and shallow neural networks. Each diagnostic classification model independently judges the same input feature and generates its own category probability distribution. The fusion unit adopts a weighted soft voting strategy to combine the outputs of each diagnostic classification model according to the weights learned in advance or online, and calculates the final fault category.

2. The rotating machinery fault diagnosis method based on adaptive feature optimization and ensemble model according to claim 1, characterized in that, The data processing module includes the following steps: Step 2.1: Use a length of Segmenting with a sliding window and a 50% overlap window shift yields discrete signal segments. ,in Indicates the sampling point number within the signal segment; Step 2.2: For each signal segment Calculate the time-domain features to form a time-domain feature vector. These include: mean, variance, standard deviation, root mean square value, peak-to-peak value, absolute peak value, kurtosis, skewness, kurtosis factor, margin factor, impulse factor, and peak factor. Step 2.3: Combine the same signal segment Performing a discrete Fourier transform yields ,in, Indicates the first The complex spectral coefficients corresponding to each frequency point Indicate the frequency point number; and calculate the power spectrum. Its total energy Then extract the frequency domain feature vector. This includes: spectral energy, average spectral energy, spectral entropy, and spectral variance; Step 2.4: Convert the time-domain feature vector With frequency domain eigenvectors Cascaded, forming the initial feature vector of this signal segment. Repeat steps 2.2 and 2.3 for all signal segments to construct the initial feature pool. ,in, Indicates the first The initial feature vector corresponding to each signal segment Indicates the signal segment number. This represents the total number of segments.

3. The rotating machinery fault diagnosis method based on adaptive feature optimization and ensemble model according to claim 2, characterized in that, Feature filtering and combination module: Employs three strategies: filtering, wrapping, and embedded. The features are filtered and fused through voting to form a preliminary feature subset. Specifically, it includes the following steps: Step 3.1: Input the initial feature pool Target fault label set , d Indicates the fault label number; Step 3.2: Filtering Step 3.2.1: Pearson correlation coefficient: ,in, Representing candidate features The Pearson correlation coefficient, Indicates fault label, Representing candidate features With fault labels covariance, Representing candidate features standard deviation Indicates fault label Standard deviation; in terms of absolute value Before selecting in descending order One characteristic, This is a preset threshold; Step 3.2.2: Mutual Information Method: ,in, Representing candidate features With fault labels mutual information, Representing candidate features With fault labels The joint probability, Representing candidate features The marginal probability, Indicates fault label Marginal probabilities; according to Before preserving descending order One characteristic, This is a preset threshold; Step 3.3: Wrap-up Filtering Using random forest as the base model, the iterative steps are as follows: Step 3.3.1: Train a random forest using the current feature set; Step 3.3.2: Calculate feature importance based on the Gini coefficient for the reduction in impurity at node splits; Step 3.3.3: In each round, eliminate the least important features until only the following remain. indivual, This is a preset threshold; Step 3.4: Embedded Filtering: in, Represents the sample feature matrix. Indicates fault label, This represents the coefficient vector corresponding to the candidate features. Represents the regularization parameter. This represents the model fitting error term. Represents the L1 regularization term, according to Before the election Item, among which, Representing candidate features The corresponding coefficients, This is a preset threshold; Step 3.5: Voting Integration Step 3.5.1: Output the filtering result list for each of the three strategies described above; Step 3.5.2: Count the votes for each feature; Step 3.5.3: The candidate with the highest number of votes... Each feature constitutes ; Step 3.6: [The sentence is incomplete and requires more context to be translated accurately.] The data is passed to the "Fault Category Assessment Module" for preliminary diagnosis.

4. The rotating machinery fault diagnosis method based on adaptive feature optimization and ensemble model according to claim 3, characterized in that, Fault Category Assessment Module: Based on the feature selection and combination module, and using the fault diagnosis module for evaluation, the evaluation results are given, specifically including the following steps; Step 4.1: Calculate the macro average of the F1 score based on the validation set. ; Step 4.2: If It directly outputs the diagnostic results; Step 4.3: Otherwise, filter based on the confusion matrix. The list of categories is used as the categories to be enhanced, and the feature enhancement generation module is triggered.

5. The rotating machinery fault diagnosis method based on adaptive feature optimization and ensemble model according to claim 1, characterized in that, The feature enhancement generation module also includes: Construct a backup feature pool Maintain a candidate feature library in the system, which includes: wavelet packet subband energy; envelope spectrum peak value; short-time Fourier transform (STFT) band energy; variational mode decomposition (VMD).

6. The rotating machinery fault diagnosis method based on adaptive feature optimization and ensemble model according to claim 4, characterized in that, The performance monitoring and model protection module includes: Enhanced feature sets are used in the "Fault Category Assessment Module". After training and evaluating the ensemble model, the responsible steps are to compare the performance of the new and old models and decide whether to roll back or continue iterating based on the results. Step 6.1: Input Step 6.1.1: Preliminary optimal state: Set of characteristics of the optimal state: The optimal state corresponds to the ensemble model: The optimal state corresponds to the ensemble model. : ; Step 6.1.2: Current Enhancement State: Enhancement State Feature Set: ;based on Training of augmented state ensemble model: Enhanced state integration model : ; Step 6.2: Performance Evaluation Step 6.2.1: Calculate the ensemble model corresponding to the optimal state on the same validation set. : Enhanced state integration model : ; Step 6.2.2: Calculate the performance gain: ; Step 6.3: Rollback and Update Logic Step 6.3.1: If The new model did not bring any improvement; A1) Rollback operation: , , ; A2) Set the end flag to terminate feature addition and model iteration, and proceed to the output module; Step 6.3.2: Otherwise The new model has improved performance; B1) Update to the optimal state: , , ; B2) Threshold determination: If If the number of additions or feature dimensions has not reached the upper limit, then set the end flag and enter the output module; otherwise, if the number of additions or feature dimensions has not reached the upper limit, return to the "feature enhancement generation module" to continue the next round of additions. Step 6.4: Iteration Limit and Protection: Maximum Number of Iterations Used to avoid infinite loops; maximum feature dimension To prevent overfitting and computational bottlenecks caused by too many features, if any upper limit is triggered and performance still fails to meet the target, the current upper limit will be used. , Output the optimal result.

7. The rotating machinery fault diagnosis method based on adaptive feature optimization and ensemble model according to claim 1, characterized in that, The fault diagnosis module includes the following steps: Step 7.1: Diagnostic Classification Model Training: For the Support Vector Machine, Random Forest, Gradient Boosting Tree, and Shallow Neural Network diagnostic classification models, train them on the validation set to obtain classification probabilities. and F1 score ,in, Indicates the first The category failure probability output by a diagnostic classification model. Indicates the first The F1 score of each diagnostic classification model on the validation set; Step 7.2: Weight Normalization: Calculate the weights ,in, Indicates the first The fusion weights of a diagnostic classification model This represents the index of the diagnostic classification models involved in the fusion. This represents the sum of the F1 scores of all diagnostic classification models involved in the fusion. Step 7.3: Soft Voting Fusion: Fusion Probability ,in, Indicates the first The fusion probability of class-specific faults Indicates the first The output of the diagnostic classification model is the first Class failure probability; and in Determine the fault type and output the diagnostic accuracy rate. Indicates the fault category in the final output. This means that the category with the highest fusion probability among all fault categories is selected as the final judgment result.

Images