Bearing fault recognition method and system based on abnormal sample suppression and multi-domain features

Abstract

This invention relates to the field of bearing fault identification. To address the problems of existing technologies where anomaly suppression relies on fixed thresholds and multi-domain feature selection lacks operational stability and redundancy constraints, this invention provides a bearing fault identification method and system based on anomaly suppression and multi-domain features. The bearing fault identification method based on anomaly suppression and multi-domain features includes: cleaning the sample set; extracting candidate features from the cleaned sample set; constructing a multi-objective feature scoring function; forming a multi-domain core feature subset from candidate features with scores not lower than the feature selection threshold; training a fault diagnosis model; preprocessing and feature extraction of the bearing vibration signal under the diagnostic operating condition to obtain multi-domain core features; and then processing these features through the trained fault diagnosis model to obtain the corresponding fault type. This method can still stably output bearing fault diagnosis results with high accuracy and low false alarm rate even under noise and fluctuating operating conditions.

Description

Technical Field

[0001] This invention relates to the field of bearing fault identification, and in particular to a bearing fault identification method and system based on abnormal sample suppression and multi-domain features. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] Rotating machinery bearings are critical components in high-speed train traction systems, electric motor transmission systems, and wind turbine generators, and their operating condition directly affects the safety and reliability of the entire system. With increasingly complex operating conditions and longer service life, bearings are prone to failures such as rolling element spalling, inner ring cracks, and outer ring pitting. Failure to identify these issues promptly and accurately can lead to significant economic losses or even safety accidents.

[0004] Existing bearing fault diagnosis methods in engineering practice mainly fall into two categories: one is experience-based rule-based diagnosis, such as spectral feature recognition and envelope demodulation, but this type of method largely relies on experience, resulting in poor reliability of fault diagnosis results; the other is intelligent diagnosis methods based on machine learning or deep learning, such as support vector machines, random forests, and convolutional neural networks. In this type of intelligent diagnosis method, the diagnostic performance is highly dependent on the quality of training samples, feature construction methods, and classification model parameter settings.

[0005] Vibration signals collected at engineering sites often contain anomalous samples due to factors such as improper installation, loose sensors, and short-term impact interference. Existing technologies typically employ anomaly detection algorithms such as isolated forests and local outliers, using manual experience to set fixed thresholds for anomaly removal, or simply limiting the anomaly removal rate to a certain percentage. These methods struggle to balance overall recognition accuracy with the performance of each fault category, especially in imbalanced scenarios. Fixed thresholds can lead to over-purging of a few fault types, weakening the diagnostic model's ability to identify key fault types. To improve diagnostic accuracy, researchers generally construct numerous features in the time and frequency domains and use methods such as correlation coefficients, principal component analysis, and minimum redundancy maximum correlation for feature selection. These methods often rely on single correlation or variance indices, failing to adequately consider the stability of features under multiple operating conditions and the redundancy between features. Furthermore, the thresholds often depend on manual experience, making it difficult for feature subsets to maintain robustness under different operating conditions and limiting generalization performance. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides a bearing fault identification method and system based on anomaly sample suppression and multi-domain features, which can still stably output bearing fault diagnosis results with high accuracy and low false alarm rate under noise and operating condition fluctuations.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] The first aspect of the present invention provides a bearing fault identification method based on abnormal sample suppression and multi-domain features.

[0009] In one or more embodiments, a bearing fault identification method based on outlier sample suppression and multi-domain features is provided, including:

[0010] Vibration signals of rotating mechanical bearings under different operating conditions are acquired and preprocessed to obtain a sample set;

[0011] Under the constraint that the total removal rate of abnormal samples does not exceed a preset upper limit, samples in the sample set whose fusion abnormal score exceeds the abnormal threshold but does not exceed the upper limit of the category removal rate are removed to obtain a cleaned sample set; wherein, the fusion abnormal score is obtained by weighted fusion of the abnormal scores of the sample based on the isolated forest model and the abnormal scores based on robust statistics.

[0012] From the cleaned sample set, candidate features are extracted in the time domain, frequency domain, and time-frequency domain. For each candidate feature, the class discrimination index, cross-working condition stability index, and redundancy index are calculated. A multi-objective feature scoring function is constructed, and candidate features with scores not lower than the feature screening threshold are grouped into a multi-domain core feature subset.

[0013] The optimization objective is to use a comprehensive fitness function consisting of macro-average F1 score, minimum recall rate of each fault category and model complexity. By iteratively updating particle position and velocity, the hyperparameters and category cost weights corresponding to the optimal particle are obtained, and the fault diagnosis model is trained by combining multi-domain core feature subsets.

[0014] The bearing vibration signal under the diagnostic condition is preprocessed and its features are extracted to obtain multi-domain core features. After being processed by the trained fault diagnosis model, the corresponding fault type is obtained.

[0015] As one implementation method, during the training and optimization process of the fault diagnosis model, the number of trees, maximum depth, feature sampling ratio, minimum number of leaf node samples, and the weight combination of classification cost for each fault category in the random forest are encoded as particles.

[0016] As one implementation method, the threshold for identifying outliers is adaptively determined using the macro-average F1 value as the objective, and its expression is:

[0017] ;

[0018] in, The threshold for identifying abnormal samples; To achieve the optimal removal rate; This represents the macro average F1 value. For each candidate removal rate; This is a set of candidate removal rates for abnormal samples; For quantile functions; Set a preset upper limit for the total removal rate of abnormal samples; It is a function used to find the index of the maximum value.

[0019] As one implementation method, the expression for the upper limit of the category removal rate is:

[0020] ;

[0021] in, Fault Category The maximum removal rate; Fault Category The number of samples; This represents the maximum number of samples across all categories. This is the adjustment coefficient, which is greater than 0 and less than or equal to 1; It is a minimum value function.

[0022] As one implementation method, the expression for the multi-objective feature scoring function is:

[0023] ;

[0024] in, For the first Scores for each candidate feature; These are non-negative weighting coefficients used to balance the discriminative power, operational stability, and redundancy of the features. For the first One candidate feature; For the first The class discriminant index of each candidate feature, For the first Cross-condition stability index of candidate features; For the first Redundancy index of candidate features.

[0025] As one implementation method, the feature selection threshold is adaptively determined based on the statistical distribution of all feature scores, which is expressed as follows:

[0026] ;

[0027] in, The threshold for feature selection; The average score of all candidate features; This is an adjustable coefficient; The standard deviation of the scores for all candidate features.

[0028] As one implementation method, the expression for the fitness function is:

[0029] ;

[0030] in, This is the overall fitness function; The macro-average F1 value for a given particle X; This represents the minimum recall rate among all fault categories. This is a model complexity metric used to measure the number of trees and tree depth. , These are positive weighting coefficients.

[0031] As one implementation method, in the process of iteratively updating particle position and velocity, an inertia weight that decreases linearly with the number of iterations is used. The calculation formula is as follows:

[0032] ;

[0033] in, This represents the current iteration number; This represents the maximum number of iterations. , These are the initial and final inertia weights, respectively.

[0034] A second aspect of the present invention provides a bearing fault identification system based on abnormal sample suppression and multi-domain features.

[0035] In one or more embodiments, a bearing fault identification system based on outlier sample suppression and multi-domain features includes:

[0036] The sample set construction module is used to acquire and preprocess the vibration signals of rotating mechanical bearings under different operating conditions to obtain a sample set.

[0037] The sample set cleaning module is used to remove samples from the sample set whose fusion anomaly score exceeds the anomaly threshold and does not exceed the category removal rate limit, under the constraint that the total removal rate of anomalies does not exceed a preset upper limit, so as to obtain a cleaned sample set; wherein, the fusion anomaly score is obtained by weighted fusion of the anomaly scores of the sample based on the isolated forest model and the anomaly scores based on robust statistics.

[0038] The multi-domain core feature extraction module is used to extract candidate features from the cleaned sample set in the time domain, frequency domain, and time-frequency domain. For each candidate feature, it calculates the class discrimination index, cross-working condition stability index, and redundancy index, constructs a multi-objective feature scoring function, and forms a multi-domain core feature subset with candidate features whose scores are not lower than the feature screening threshold.

[0039] The fault diagnosis model training module is used to train the fault diagnosis model by iteratively updating the particle position and velocity, obtaining the hyperparameters and class cost weights corresponding to the optimal particles, and combining multi-domain core feature subsets with the comprehensive fitness function consisting of macro-average F1 value, minimum recall rate of each fault category and model complexity as the optimization objective.

[0040] The bearing fault diagnosis module is used to preprocess and extract features from the bearing vibration signal under the condition to be diagnosed, obtain multi-domain core features, and then process them through the trained fault diagnosis model to obtain the corresponding fault type.

[0041] A third aspect of the present invention provides an electronic device.

[0042] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps in the bearing fault identification method based on anomaly sample suppression and multi-domain features as described above.

[0043] Compared with the prior art, the beneficial effects of the present invention are:

[0044] This invention provides a bearing fault identification method and system based on anomaly sample suppression and multi-domain features. In the anomaly sample suppression stage, the domain feature construction and screening stage, and the model modeling stage, the samples are adaptively cleaned, adaptively screened, and jointly optimized by multi-objective fitness function and particle swarm algorithm, which improves the overall diagnostic accuracy and significantly improves the identification ability of minority class faults.

[0045] In the anomaly suppression stage, this invention weighted and fused the anomaly scores from the isolated forest model and the robust statistical anomaly scores. While limiting the overall anomaly removal rate, it maximized the adaptive search anomaly threshold using the macro-average F1 score. Simultaneously, it introduced a maximum removal rate formula constrained by fault category, effectively avoiding over-cleaning of minority class samples and improving the targeted nature of the data cleaning process for diagnostic performance. In the multi-domain feature construction and screening stage, it introduced cross-operating condition feature stability and redundancy indices, constructing a multi-objective feature scoring function. Combined with the adaptive feature threshold, this allows for the selection of feature sets with good fault discrimination ability, high operating condition stability, and low redundancy under multiple operating conditions, improving the generalization ability of the diagnostic model under changing operating conditions. In the model modeling stage, it used a multi-objective fitness function composed of macro-average F1, minimum class recall, and model complexity as the optimization objective, employing a particle swarm optimization algorithm for joint optimization. This not only improved the overall diagnostic accuracy but also significantly enhanced the identification ability of minority class faults, while controlling model complexity and computational cost. Attached Figure Description

[0046] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0047] Figure 1 This is a flowchart of the bearing fault identification method based on abnormal sample suppression and multi-domain features according to an embodiment of the present invention;

[0048] Figure 2 This is a graph showing the performance evaluation results of the fault diagnosis model according to an embodiment of the present invention;

[0049] Figure 3 This is an example of a confusion matrix in the training and verification process of the fault diagnosis model according to an embodiment of the present invention;

[0050] Figure 4 This is a schematic diagram of the bearing fault identification system based on abnormal sample suppression and multi-domain features according to an embodiment of the present invention;

[0051] Figure 5 This is a schematic diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0052] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0053] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0054] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0055] Figure 1 A schematic diagram of the bearing fault identification method based on anomaly sample suppression and multi-domain features according to an embodiment of the present invention is provided. Figure 1 The bearing fault identification method based on abnormal sample suppression and multi-domain features in this embodiment may include the following steps S101 to S105.

[0056] The specific implementation process of steps S101 to S105 is as follows:

[0057] Step S101: Obtain vibration signals of rotating mechanical bearings under different operating conditions and perform preprocessing to obtain a sample set.

[0058] In step S101, vibration signals of rotating mechanical bearings under various operating conditions such as rotational speed and load are collected. The vibration signals are preprocessed by detrending, filtering, and normalization, and then segmented according to fixed length and overlap rate to obtain a sample set.

[0059] Step S102: Under the constraint that the total removal rate of abnormal samples does not exceed the preset upper limit, samples in the sample set whose fusion abnormal score exceeds the abnormal threshold and does not exceed the upper limit of the category removal rate are removed to obtain the cleaned sample set; wherein, the fusion abnormal score is obtained by weighted fusion of the abnormal score of the sample based on the isolated forest model and the abnormal score based on robust statistics.

[0060] Specifically, for each sample, both anomaly scores based on the isolated forest model and those based on robust statistics are calculated simultaneously, and a fused anomaly score is constructed. For each sample... Model anomaly scores were obtained based on the isolated forest model. Robust Z-score outlier scores were calculated based on the median and the median absolute deviation. After normalization, the two are weighted and fused according to the following formula:

[0061] ,in For fusion weights, 0 ≤ ≤1.

[0062] Sort all samples by their fusion anomaly scores from largest to smallest, and let the set of candidate anomaly removal rates be denoted as . ,in Set a preset upper limit for the total removal rate of abnormal samples; for each candidate removal rate The corresponding quantile is taken as the candidate anomaly threshold. Delete those that meet the requirements The sample is used to train a random forest classifier with the remaining sample, and the macro-average F1 score is calculated on the validation set. Obtain the optimal removal rate that maximizes the macro average F1 score. and corresponding abnormal thresholds ,in:

[0063] The threshold for identifying outliers is adaptively determined using the macro-average F1 score as the objective, and its expression is as follows:

[0064] ;

[0065] in, For quantile functions; It is a function that takes the parameters of a function.

[0066] In this embodiment of the invention, the expression for the upper limit of the category removal rate is:

[0067] ;

[0068] in, Fault Category The maximum removal rate; Fault Category The number of samples; This represents the maximum number of samples across all categories. This is the adjustment coefficient, which is greater than 0 and less than or equal to 1; This is a minimum value function. Within each category, sort by fusion anomaly score from largest to smallest, and only delete those scores not exceeding a certain threshold. One abnormal sample.

[0069] Regarding outlier suppression, this embodiment weights and fuses the outlier scores of the isolated forest model with the robust statistical outlier scores. Under the condition that the total outlier removal rate is limited, the adaptive search outlier threshold is maximized by the macro-average F1 value. At the same time, a maximum removal rate formula constrained by fault category is introduced to effectively avoid over-cleaning of minority class samples and improve the targeting of the data cleaning process for diagnostic performance.

[0070] Step S103: Extract candidate features from the cleaned sample set in the time domain, frequency domain, and time-frequency domain. Calculate the class discrimination index, cross-working condition stability index, and redundancy index for each candidate feature. Construct a multi-objective feature scoring function and form a multi-domain core feature subset with candidate features whose scores are not lower than the feature screening threshold.

[0071] For example, features extracted in the time domain include, but are not limited to, mean, root mean square, and peak value;

[0072] Features extracted in the frequency domain include, but are not limited to, spectral features;

[0073] Features extracted in the time-frequency domain include, but are not limited to, wavelet energy.

[0074] The fault characteristics corresponding to the candidate features are determined by vibration frequency, which is calculated based on bearing parameters.

[0075] In step S103, the expression for the multi-target feature scoring function is:

[0076] ;

[0077] in, For the first Scores for each candidate feature; These are non-negative weighting coefficients used to balance the discriminative power, operational stability, and redundancy of the features. For the first One candidate feature; For the first The class discriminant index of each candidate feature, For the first Cross-condition stability index of candidate features; For the first Redundancy index of candidate features.

[0078] The feature selection threshold is adaptively determined based on the statistical distribution of all feature scores, and is expressed as follows:

[0079] ;

[0080] in, The threshold for feature selection; The average score of all candidate features; This is an adjustable coefficient; The standard deviation of the scores for all candidate features.

[0081] Preferred 0≤ ≤2, will satisfy The candidate features are used as a subset of the multi-domain core features.

[0082] In terms of multi-domain feature construction and selection, this embodiment introduces cross-operating condition feature stability index and redundancy index, constructs a multi-objective feature scoring function, and combines the adaptive threshold design of the scoring distribution to select a feature set with good fault differentiation ability, high operating condition stability and low redundancy under multiple operating conditions, thereby improving the generalization ability of the diagnostic model under changing operating conditions.

[0083] Step S104: Using the comprehensive fitness function composed of macro-average F1 value, minimum recall rate of each fault category and model complexity as the optimization objective, the hyperparameters and category cost weights corresponding to the optimal particles are obtained by iteratively updating particle positions and velocities, and the fault diagnosis model is trained by combining multi-domain core feature subsets.

[0084] In this embodiment of the invention, the fault diagnosis model selected is: Particle Swarm Optimization (PSO)-Random Forest model.

[0085] In the training and optimization process of the fault diagnosis model, the number of trees, maximum depth, feature sampling ratio, minimum number of leaf node samples, and the weight combination of classification cost for each fault category of the random forest are encoded as particles.

[0086] Particle coding includes:

[0087] ;

[0088] in, The number of trees in the random forest. The maximum depth of the tree. The feature sampling ratio, The minimum number of samples for a leaf node. The classification values ​​are weighted according to normal condition, rolling element failure, inner ring failure, and outer ring failure.

[0089] The expression for the overall fitness function is:

[0090] ;

[0091] in, This is the overall fitness function; The macro-average F1 value for a given particle X; This represents the minimum recall rate among all fault categories. This is a model complexity metric used to measure the number of trees and tree depth. , These are positive weighting coefficients.

[0092] During the iterative update of particle position and velocity, an inertial weight that decreases linearly with the number of iterations is used. The calculation formula is as follows:

[0093] ;

[0094] in, This represents the current iteration number; This represents the maximum number of iterations. , These are the initial and final inertia weights, respectively.

[0095] Random forest models are widely used in bearing fault diagnosis due to their strong nonlinear mapping capabilities and resistance to overfitting. Existing technologies employ particle swarm optimization (PSO) to optimize the hyperparameters of random forests, but most only perform single-objective optimization on a small number of structural parameters such as the number of trees and maximum depth. They typically focus on classification accuracy or a single F1 score, without explicitly considering class imbalance or introducing minority class recall and model complexity constraints during optimization. This can easily lead to insufficient identification of a few key fault categories or excessively large model parameters resulting in increased computational overhead. This invention, based on actual engineering conditions and data characteristics, constructs a bearing fault diagnosis method that integrates outlier suppression, adaptive multi-domain feature selection, and multi-objective PSO-random forest modeling, improving diagnostic accuracy and robustness.

[0096] In terms of model building, this embodiment encodes the random forest structure parameters and the classification cost weights of each fault category into particles, and uses a multi-objective fitness function composed of macro-average F1, minimum class recall and model complexity as the optimization objective. It uses particle swarm optimization to jointly optimize the model, which not only improves the overall diagnostic accuracy, but also significantly improves the ability to identify minority class faults, while controlling the model complexity and computational cost.

[0097] Figure 2 This is a graph showing the performance evaluation results of the fault diagnosis model according to an embodiment of the present invention, wherein, Figure 2 Figure (a) in the figure shows the distribution of prediction confidence. Figure 2 Figure (b) shows the prediction accuracy for each category. Figure 2 Figure (c) in the figure shows the confidence comparison between correct and misclassification. Figure 2 Figure (d) in the figure shows the predicted probability distribution of each category, proving that the random forest diagnostic model not only performs well overall in the validation / test, but also outputs high confidence, clearly distinguishes between categories, and often has low confidence in erroneous samples. Therefore, the model results have engineering usability and controllability.

[0098] Figure 3 This is an example of a confusion matrix in the training and verification process of the fault diagnosis model according to an embodiment of the present invention. Figure 3 In this study, the model's ability to identify four states / faults was demonstrated by using a "class-by-class, error-by-error" approach: most samples fell on the diagonal (correctly judged), the normal class had zero misclassifications, and a small number of errors were concentrated in specific similar fault pairs, thus demonstrating that the model is generally reliable and the errors are explainable and can be improved in a targeted manner.

[0099] Step S105: Preprocess and extract features from the bearing vibration signal under the working condition to be diagnosed to obtain multi-domain core features. After processing by the trained fault diagnosis model, the corresponding fault type is obtained.

[0100] The bearing fault identification method based on abnormal sample suppression and multi-domain features in this invention is easy to implement on existing vibration monitoring platforms and can be widely applied to bearing condition monitoring and fault diagnosis in high-speed trains, motor drive systems and wind power generation.

[0101] like Figure 4 As shown, the bearing fault identification system based on abnormal sample suppression and multi-domain features provided in this embodiment of the invention can be implemented in software. The bearing fault identification system based on abnormal sample suppression and multi-domain features includes the following software modules: sample set construction module 401, sample set cleaning module 402, multi-domain core feature extraction module 403, fault diagnosis model training module 404, and bearing fault diagnosis module 405.

[0102] The functions of each software module in the bearing fault identification system based on outlier suppression and multi-domain features are described below:

[0103] The sample set construction module 401 is used to acquire vibration signals of rotating mechanical bearings under different working conditions and preprocess them to obtain a sample set.

[0104] The sample set cleaning module 402 is used to remove samples from the sample set whose fusion anomaly score exceeds the anomaly threshold and does not exceed the category removal rate limit, under the constraint that the total removal rate of abnormal samples does not exceed a preset upper limit, so as to obtain a cleaned sample set; wherein, the fusion anomaly score is obtained by weighted fusion of the anomaly score of the sample based on the isolated forest model and the anomaly score based on robust statistics.

[0105] The multi-domain core feature extraction module 403 is used to extract candidate features from the cleaned sample set in the time domain, frequency domain, and time-frequency domain. For each candidate feature, it calculates the class discrimination index, cross-working condition stability index, and redundancy index, constructs a multi-objective feature scoring function, and forms a multi-domain core feature subset with candidate features whose scores are not lower than the feature screening threshold.

[0106] The fault diagnosis model training module 404 is used to train the fault diagnosis model by iteratively updating the particle position and velocity, obtaining the hyperparameters and class cost weights corresponding to the optimal particle, and combining the multi-domain core feature subsets with the macro-average F1 value, the minimum recall rate of each fault category and the model complexity as the optimization objective.

[0107] The bearing fault diagnosis module 405 is used to preprocess and extract features from the bearing vibration signal under the working condition to be diagnosed, obtain multi-domain core features, and then process them through the trained fault diagnosis model to obtain the corresponding fault type.

[0108] It should be noted that each module in the bearing fault identification system based on anomaly sample suppression and multi-domain features in this embodiment corresponds one-to-one with each step in the bearing fault identification method based on anomaly sample suppression and multi-domain features in the above embodiment, and their specific implementation processes are the same, so they will not be repeated here.

[0109] The structure of the electronic device according to an embodiment of the present invention will be described in detail below. Figure 5 This is a schematic diagram of the composition structure of an electronic device provided in an embodiment of the present invention. It can be understood that... Figure 5 The diagram shows only an exemplary structure of the electronic device, not the entire structure. Some or all of the structures shown may be implemented as needed.

[0110] The electronic device provided in this embodiment of the invention includes: at least one processor 501, a memory 502, a user interface 503, and at least one network interface 504. The various components in the bearing fault identification system based on anomaly sample suppression and multi-domain features are coupled together via a bus system 505. It can be understood that the bus system 505 is used to realize the connection and communication between these components. In addition to a data bus, the bus system 505 also includes a power bus, a control bus, and a status signal bus. However, for clarity, in... Figure 5 The general designated all buses as Bus System 505.

[0111] The user interface 503 may include a monitor, keyboard, mouse, trackball, click wheel, buttons, touchpad, or touch screen.

[0112] It is understood that memory 502 can be volatile memory or non-volatile memory, or both. In this embodiment of the invention, memory 502 is capable of storing data to support the operation of the terminal. Examples of this data include any computer programs used to operate on the terminal, such as operating systems and applications. The operating system includes various system programs, such as framework layers, core library layers, driver layers, etc., used to implement various basic services and handle hardware-based tasks. Applications can include various applications.

[0113] In some embodiments, the bearing fault identification system based on anomaly sample suppression and multi-domain features provided in this invention can be implemented using a combination of hardware and software. As an example, the bearing fault identification system based on anomaly sample suppression and multi-domain features provided in this invention can be a processor in the form of a hardware decoding processor, which is programmed to execute the bearing fault identification method based on anomaly sample suppression and multi-domain features provided in this invention. For example, the processor in the form of a hardware decoding processor can employ one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), or other electronic components.

[0114] As an example, processor 501 can be an integrated circuit chip with signal processing capabilities, such as a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc., wherein the general-purpose processor can be a microprocessor or any conventional processor, etc.

[0115] As an example of the hardware implementation of the bearing fault identification system based on anomaly sample suppression and multi-domain features provided in this embodiment of the invention, the device provided in this embodiment of the invention can be directly executed by a processor 501 in the form of a hardware decoding processor. For example, it can be executed by one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), or other electronic components to implement the bearing fault identification method based on anomaly sample suppression and multi-domain features provided in this embodiment of the invention.

[0116] The memory 502 in this embodiment of the invention is used to store various types of data to support the operation of the bearing fault identification system based on outlier suppression and multi-domain features, or to store data for execution. Figure 1The program code for the method shown. Examples of this data include: any executable instructions for operation on a bearing fault identification system based on outlier suppression and multi-domain features, such as executable instructions that can be included in the executable instructions to implement the bearing fault identification method based on outlier suppression and multi-domain features of the embodiments of the present invention.

[0117] Specifically, according to embodiments of this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program including functions for executing... Figure 1 The program code for the method shown. In such an embodiment, the computer program can be downloaded and installed from a network via a communication component, and / or installed from a removable medium. When the computer program is executed by the central processing unit, it performs the various functions defined in the apparatus of this application.

[0118] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0119] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A bearing fault identification method based on outlier suppression and multi-domain features, characterized in that, include: Vibration signals of rotating mechanical bearings under different operating conditions are acquired and preprocessed to obtain a sample set; Under the constraint that the total removal rate of abnormal samples does not exceed a preset upper limit, samples in the sample set whose fusion abnormal score exceeds the abnormal threshold but does not exceed the upper limit of the category removal rate are removed to obtain a cleaned sample set; wherein, the fusion abnormal score is obtained by weighted fusion of the abnormal scores of the sample based on the isolated forest model and the abnormal scores based on robust statistics. The threshold for identifying outliers is adaptively determined using the macro-average F1 score as the objective, and its expression is as follows: ； in, The threshold for identifying abnormal samples; To achieve the optimal removal rate; This represents the macro average F1 value. For each candidate removal rate; This is a set of candidate removal rates for abnormal samples; For quantile functions; Set a preset upper limit for the total removal rate of abnormal samples; It is a function used to find the index of the maximum value; The expression for the upper limit of category removal rate is: ； in, Fault Category The maximum removal rate; Fault Category The number of samples; This represents the maximum number of samples across all categories. This is the adjustment coefficient, which is greater than 0 and less than or equal to 1; It is a minimum value function; From the cleaned sample set, candidate features are extracted in the time domain, frequency domain, and time-frequency domain. For each candidate feature, the class discrimination index, cross-working condition stability index, and redundancy index are calculated. A multi-objective feature scoring function is constructed, and candidate features with scores not lower than the feature screening threshold are grouped into a multi-domain core feature subset. The optimization objective is to use a comprehensive fitness function consisting of macro-average F1 score, minimum recall rate of each fault category and model complexity. By iteratively updating particle position and velocity, the hyperparameters and category cost weights corresponding to the optimal particle are obtained, and the fault diagnosis model is trained by combining multi-domain core feature subsets. The expression for the overall fitness function is: ； in, This is the overall fitness function; The macro-average F1 value for a given particle X; This represents the minimum recall rate among all fault categories. This is a model complexity metric used to measure the number of trees and tree depth. , These are positive weighting coefficients; In the training and optimization process of the fault diagnosis model, the number of trees, maximum depth, feature sampling ratio, minimum number of leaf node samples, and the weight combination of classification cost of each fault category in the random forest are encoded as particles. The bearing vibration signal under the diagnostic condition is preprocessed and its features are extracted to obtain multi-domain core features. After being processed by the trained fault diagnosis model, the corresponding fault type is obtained.

2. The bearing fault identification method based on outlier suppression and multi-domain features as described in claim 1, characterized in that, The expression for the multi-objective feature scoring function is: ； in, For the first Scores for each candidate feature; These are non-negative weighting coefficients used to balance the discriminative power, operational stability, and redundancy of the features. For the first One candidate feature; For the first The class discriminant index of each candidate feature, For the first Cross-condition stability index of candidate features; For the first Redundancy index of candidate features.

3. The bearing fault identification method based on outlier suppression and multi-domain features as described in claim 1, characterized in that, The feature selection threshold is adaptively determined based on the statistical distribution of all feature scores, and is expressed as follows: ； in, The threshold for feature selection; The average score of all candidate features; This is an adjustable coefficient; The standard deviation of the scores for all candidate features.

4. The bearing fault identification method based on outlier suppression and multi-domain features as described in claim 1, characterized in that, During the iterative update of particle position and velocity, an inertial weight that decreases linearly with the number of iterations is used. The calculation formula is as follows: ； in, This represents the current iteration number; This represents the maximum number of iterations. , These are the initial and final inertia weights, respectively.

5. A bearing fault identification system based on outlier suppression and multi-domain features, characterized in that, The bearing fault identification method based on outlier suppression and multi-domain features as described in any one of claims 1-4 includes: The sample set construction module is used to acquire and preprocess the vibration signals of rotating mechanical bearings under different operating conditions to obtain a sample set. The sample set cleaning module is used to remove samples from the sample set whose fusion anomaly score exceeds the anomaly threshold and does not exceed the category removal rate limit, under the constraint that the total removal rate of anomalies does not exceed a preset upper limit, so as to obtain a cleaned sample set; wherein, the fusion anomaly score is obtained by weighted fusion of the anomaly scores of the sample based on the isolated forest model and the anomaly scores based on robust statistics. The multi-domain core feature extraction module is used to extract candidate features from the cleaned sample set in the time domain, frequency domain, and time-frequency domain. For each candidate feature, it calculates the class discrimination index, cross-working condition stability index, and redundancy index, constructs a multi-objective feature scoring function, and forms a multi-domain core feature subset with candidate features whose scores are not lower than the feature screening threshold. The fault diagnosis model training module is used to train the fault diagnosis model by iteratively updating the particle position and velocity, obtaining the hyperparameters and class cost weights corresponding to the optimal particles, and combining multi-domain core feature subsets with the comprehensive fitness function consisting of macro-average F1 value, minimum recall rate of each fault category and model complexity as the optimization objective. The bearing fault diagnosis module is used to preprocess and extract features from the bearing vibration signal under the condition to be diagnosed, obtain multi-domain core features, and then process them through the trained fault diagnosis model to obtain the corresponding fault type.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps in the bearing fault identification method based on anomaly sample suppression and multi-domain features as described in any one of claims 1-4.

Images