A method for constructing a rolling bearing fault diagnosis model, application thereof, and electronic device

Abstract

This invention belongs to the technical field of fault diagnosis for rotating machinery and discloses a method for constructing a rolling bearing fault diagnosis model, its application, and electronic equipment. The method includes: collecting vibration signals of the rolling bearing under different health conditions; for each health condition, constructing an objective function based on the envelope entropy of the IMF components obtained from VMD decomposition to optimize and obtain optimal VMD parameters; performing vibration signal decomposition based on the optimal VMD parameters; constructing a feature vector based on the energy entropy of the IMF components obtained from the decomposition under the optimal VMD parameters; and obtaining a fault diagnosis model through training and optimization using the feature vector as input and the fault type as output. This invention determines the optimal VMD parameters for different bearing fault types through an optimization algorithm. K , α This enables adaptive and high-fidelity decomposition of vibration signals, ensuring the effective separation of weak fault features and improving diagnostic accuracy.

Description

Technical Field

[0001] This invention belongs to the technical field of fault diagnosis of rotating machinery and equipment, and more specifically, relates to a method for constructing a fault diagnosis model of rolling bearings, its application, and electronic equipment. Background Technology

[0002] As a core component of rotating machinery, the operating condition of rolling bearings directly determines the reliability and safety of the entire equipment. Early bearing failures, such as micro-cracks or pitting in the inner ring, outer ring, or rolling elements, if not diagnosed promptly and accurately, can lead to worsening of the fault, triggering a chain reaction, causing huge economic losses or even safety accidents.

[0003] Currently, vibration signal-based fault diagnosis methods remain the mainstream. However, in engineering practice, fault features are extremely weak and often masked by strong background noise, posing a significant challenge to fault feature extraction and identification. Existing technologies employ advanced signal processing methods such as VMD to decompose the original vibration signal to improve the effectiveness of fault feature extraction, but existing fault diagnosis models still have the following shortcomings: VMD decomposition results are unstable, exhibiting modal aliasing, over-decomposition, or under-decomposition, thereby compromising the integrity of fault features and affecting the model's diagnostic accuracy. Summary of the Invention

[0004] In view of the above-mentioned defects or improvement needs of the existing technology, the present invention provides a method for constructing a rolling bearing fault diagnosis model, its application and electronic equipment, which is used to solve the problem that the existing fault diagnosis model has poor stability of VMD decomposition effect, and will have mode mixing, over-decomposition or under-decomposition, thereby destroying the integrity of fault features and affecting the model diagnosis accuracy.

[0005] To achieve the above objectives, according to a first aspect of the present invention, a method for constructing a rolling bearing fault diagnosis model is provided, comprising: Signal acquisition and preprocessing: Vibration signals of rolling bearings under different health conditions are acquired, and the signals are divided into sliding windows and normalized according to a preset window length to establish a dataset; Parameter-adaptive VMD decomposition: For vibration signals of each health state type, an objective function is constructed based on the envelope entropy of the IMF components obtained by VMD decomposition to optimize and obtain the corresponding optimal VMD parameters, and vibration signal decomposition is performed based on the optimal VMD parameters; Feature vector construction: For each health state type, a feature vector is constructed based on the energy entropy of the IMF components obtained by decomposition under the optimal VMD parameters; Model training and optimization: Based on the dataset, with feature vectors as input and fault type as output, a fault diagnosis model is obtained through training and optimization.

[0006] According to the method for constructing a rolling bearing fault diagnosis model provided by the present invention, the parameter adaptive VMD decomposition specifically includes: For each health state type, a VMD parameter optimization space is constructed, which includes the number of modes and the penalty factor. The search space for the number of modes is defined as the set of integers [2, 12], and the search space for the penalty factor is defined as the continuous interval [500, 5000]. The objective function is to minimize the average sum of the envelope entropies of the vibration signals for each health state type with respect to the decomposed IMF components. Using a Bayesian optimization algorithm, with the objective function as the fitness function, an iterative search is performed in the parameter optimization space to obtain the optimal VMD parameters corresponding to each health state type.

[0007] According to the method for constructing a rolling bearing fault diagnosis model provided by the present invention, the method further includes the following steps before constructing the feature vector: Key IMF component screening based on EWK: For each health state type, based on the IMF components obtained by decomposition under the optimal VMD parameters, calculate the EWK value of each IMF component, and screen out the top at least one key IMF component based on the EWK value; wherein, the EWK value is the product of the energy weight, kurtosis value and correlation coefficient of the IMF component. Accordingly, the feature vector construction is specifically as follows: feature vectors are constructed based on the energy entropy of key IMF components.

[0008] According to the method for constructing a rolling bearing fault diagnosis model provided by the present invention, at least one IMF component with the largest EWK value is selected as the critical IMF component; the formula for calculating the EWK value is: ; in, η k For the first k Energy weights of each IMF component; K k For the first k Kurtosis values ​​of each IMF component; ρ k For the first k The correlation coefficient between each IMF component and the original vibration signal.

[0009] According to the method for constructing a rolling bearing fault diagnosis model provided by the present invention, the feature vector construction specifically involves: calculating the energy entropy of each key IMF component to construct the feature vector. H 1, H 2, ..., H n The definition of energy entropy is: ; in, n The number of modes; For the first i Energy entropy of a key IMF component; E i For the first i The energy of each key IMF component, i.e., the envelope amplitude; For the first i The energy share of each key IMF component.

[0010] According to the method for constructing a rolling bearing fault diagnosis model provided by the present invention, model training and optimization specifically include: Dataset partitioning: The feature vector set of all samples is proportionally divided into a training set and a test set; GWO-SVM Model Establishment and Optimization: Using the training set, the GWO algorithm is used to simultaneously optimize the penalty factor and kernel parameters of the SVM model to establish a GWO-SVM fault diagnosis model. Model performance validation: Input the feature vectors of the test set samples into the trained GWO-SVM fault diagnosis model to evaluate the model's diagnostic accuracy.

[0011] According to the method for constructing a rolling bearing fault diagnosis model provided by the present invention, the GWO algorithm guides the search with a fitness function, which is the classification accuracy of the SVM model on the test set; the population size of the GWO algorithm is 30, the maximum number of iterations is 200, and the search range of the penalty factor and kernel parameters of the SVM model is [0.01, 100]. The SVM model uses a radial basis function kernel, the expression of which is: ; in, x i and x j For any two input feature vectors, || || represents Euclidean distance; σ These are kernel function parameters used to control the complexity and mapping capability of the model.

[0012] According to a second aspect of the present invention, a method for diagnosing rolling bearing faults is provided, wherein the fault diagnosis model is constructed based on any of the above claims, and the diagnosis method includes: Based on the vibration signal to be diagnosed, a Bayesian optimization algorithm is used to construct an objective optimization function based on the envelope entropy of the IMF components obtained by VMD decomposition. The online VMD parameters are then optimized and obtained within a preset parameter search space. The online VMD parameters are then used to perform VMD decomposition on the vibration signal to be diagnosed. Based on the energy entropy of the IMF components obtained by online VMD parameter decomposition, a feature vector to be diagnosed is constructed. The feature vector to be diagnosed is input into the fault diagnosis model, and the fault category diagnosis result is output.

[0013] According to the rolling bearing fault diagnosis method provided by the present invention, the sum of the envelope entropy of the IMF components obtained by VMD decomposition is designed as the objective optimization function; Before constructing the feature vector to be diagnosed, the following steps are also included: For the IMF components obtained by decomposition under online VMD parameters, the EWK value of each IMF component is calculated, and the top at least one online key IMF component is selected based on the EWK value; wherein, the EWK value is the product of the energy weight, kurtosis value and correlation coefficient of the IMF component. Accordingly, the construction of the feature vector to be diagnosed is specifically as follows: based on the energy entropy of the online key IMF components, the feature vector to be diagnosed is constructed.

[0014] According to a third aspect of the present invention, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the program to implement the steps of the method for constructing a rolling bearing fault diagnosis model as described in any of the preceding claims.

[0015] In summary, compared with the prior art, the method for constructing a rolling bearing fault diagnosis model, its application, and electronic equipment provided by this invention offer the following advantages: 1. Improved Feature Extraction Precision: Optimal VMD parameters for different bearing fault types were determined through algorithm optimization. K , α This enables adaptive and high-fidelity decomposition of vibration signals, ensuring the effective separation of weak fault features. 2. Robust Feature Selection: The proposed EWK index integrates three aspects of information: energy, impact characteristics (kurtosis), and waveform similarity (correlation coefficient). It can more comprehensively and stably screen out key fault mode components from the noise background, overcoming the limitations of a single index. 3. Classification performance optimization: GWO is used to efficiently and globally optimize SVM hyperparameters, avoiding the blindness of human selection and the local optima problem of traditional optimization methods, ensuring that the classifier is always in or close to the best performance state; 4. This invention constructs a VMD-EWK-GWO-SVM model, which solves the problems of insufficient extraction of weak fault features under strong noise background, poor robustness of modal component selection, and low efficiency of classifier parameter optimization. While improving diagnostic accuracy, it also enhances the engineering applicability of the method. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the overall process of the rolling bearing fault diagnosis method based on VMD-EWK-GWO-SVM described in this invention.

[0017] Figure 2 This is a time-domain waveform diagram of each IMF component after VMD decomposition of the inner ring fault signal in an embodiment of the present invention.

[0018] Figure 3 This is a frequency domain spectrum diagram of each IMF component after VMD decomposition of the inner ring fault signal in an embodiment of the present invention.

[0019] Figure 4 The diagram shows the confusion matrix results of SVM optimized using PSO.

[0020] Figure 5 The image shows the confusion matrix result of SVM optimized using WOA.

[0021] Figure 6 The diagram shows the confusion matrix results of SVM optimized using the GWO method of this invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0023] Please see Figure 1 This embodiment provides a method for constructing a rolling bearing fault diagnosis model, the method comprising: Signal acquisition and preprocessing: Vibration signals of rolling bearings under different health conditions are acquired, and the signals are divided into sliding windows and normalized according to a preset window length to establish a dataset; for example, the sliding window length is 1024 and the window overlap rate is 0. Parameter-adaptive VMD decomposition: For vibration signals of each health state type, an objective function is constructed based on the envelope entropy of the IMF components obtained by VMD decomposition to optimize and obtain the corresponding optimal VMD parameters, and vibration signal decomposition is performed based on the optimal VMD parameters; Feature vector construction: For each health state type, a feature vector is constructed based on the energy entropy of the IMF components obtained by decomposition under the optimal VMD parameters; Model training and optimization: Based on the dataset, with feature vectors as input and fault type as output, a fault diagnosis model is obtained through training and optimization.

[0024] This embodiment takes into account advanced signal processing methods such as VMD, and its decomposition effect is highly dependent on the number of modes. K and penalty factor α The setting of parameters is crucial; improper parameter selection can lead to modal aliasing, over-decomposition, or under-decomposition, thereby compromising the integrity of fault features. Therefore, this paper proposes optimizing parameters based on the envelope entropy of the IMF components obtained from VMD decomposition. This allows for the selection of better decomposition parameters, improving decomposition performance and stability. Furthermore, it proposes optimizing VMD parameters separately for different health state types. By differentiating and decomposing health state types, the signals of each health state type can be decomposed under optimal parameters. This not only enables targeted signal decomposition for accurate extraction of different types of fault features but also improves the stability of signal decomposition. It helps avoid instability such as modal aliasing, over-decomposition, or under-decomposition caused by inappropriate parameters, thus ensuring the integrity of fault features and improving the diagnostic accuracy of the model.

[0025] In some embodiments, parameter adaptive VMD decomposition specifically includes: For each health state type, construct a system containing the number of modalities. K and penalty factor α The VMD parameter optimization space, where the number of modes K The search space is defined as the set of integers in the range [2, 12], with a penalty factor of . α The search space is defined as a continuous interval of [500, 5000]. The objective function is to minimize the average sum of the envelope entropies of the vibration signals for each health state type with respect to the IMF components obtained from the VMD decomposition; that is, for each VMD decomposition... K Each IMF component is used to calculate the sum of its envelope entropy. Then, by combining the vibration signals of each health state type, the average value of the sum of envelope entropy can be obtained. Minimizing this average value is the optimization objective. A Bayesian optimization algorithm is employed, using the objective function as the fitness function, to iteratively search the parameter optimization space to obtain the optimal VMD parameters corresponding to each health state type. Then, the optimal parameters are used to perform VMD decomposition on the vibration signals of all samples belonging to that state type, yielding... K One IMF component.

[0026] This embodiment considers that current VMD parameter determination largely relies on experience or algorithm optimization. Experience-based trial and error is highly subjective and lacks universality; while some intelligent algorithm optimizations involve high computational costs. Therefore, this embodiment proposes using a Bayesian optimization algorithm within a defined parameter search space, which is computationally fast and efficient. In this embodiment, the search space is set based on the typical physical characteristics of rolling bearing fault vibration signals and the principles of the VMD algorithm. Modal number K The lower bound of 2 ensures that the signal can be decomposed into at least one trend term and one detail component; the upper bound of 12 is based on engineering experience to prevent meaningless over-decomposition of bearing signals with a sampling frequency of 12kHz. Penalty factor. α The lower limit of 500 ensures sufficient bandwidth constraint and prevents excessive modal smoothing; the upper limit of 5000 is because when α When this value is exceeded, further increases do not significantly improve fault feature extraction.

[0027] In some embodiments, the optimization of VMD parameters may further include: firstly, using a Bayesian optimization algorithm to optimize and obtain multiple candidate VMD parameter combinations based on the objective function; then, determining the optimal VMD parameters through quantitative and / or qualitative verification. Quantitative verification involves dividing the dataset into a training set and a test set. Candidate VMD parameter combinations are then obtained by optimizing on the training set, and the objective function value of each candidate parameter combination on the test set is calculated to determine the ranking of the candidate VMD parameter combinations. Qualitative verification involves decomposing the test set samples using the candidate VMD parameter combinations and observing whether the time-domain waveform of the obtained IMF components exhibits periodic impacts, or whether significant spectral lines of theoretical fault characteristic frequencies appear in their envelope spectra. The ranking of the candidate VMD parameter combinations is obtained through qualitative analysis. For a fault-free healthy state type, the combination that does not exhibit periodic impacts and does not have significant spectral lines of theoretical fault characteristic frequencies is preferred; for a faulty healthy state type, the combination that exhibits periodic impacts and has significant spectral lines of theoretical fault characteristic frequencies corresponding to the fault type is preferred. The optimal parameters are selected by combining the results of quantitative and qualitative verification.

[0028] In some embodiments, the following steps are included prior to feature vector construction: Key IMF component selection based on EWK: For each health state type, based on the IMF components obtained under optimal VMD parameters, calculate the EWK value of each IMF component, and select at least one key IMF component based on the EWK value; wherein, the EWK value is the product of the energy weight, kurtosis value, and correlation coefficient of the IMF component; the number of key IMF components can be 3-5, or the IMF component with the smallest envelope entropy can be selected as the key IMF component. Correspondingly, the feature vector construction is specifically as follows: construct feature vectors based on the energy entropy of the key IMF components.

[0029] This embodiment addresses the limitations of existing feature selection methods, which often rely on single metrics and lack robustness. When filtering fault-information components from decomposed modal components, current methods typically employ a single metric, such as correlation coefficient or kurtosis. While correlation coefficients exhibit poor stability in high-noise environments, kurtosis, though sensitive to shocks, fails to comprehensively reflect fault modes that simultaneously possess high energy and specific frequency distribution characteristics, easily leading to the omission of critical components. Therefore, this embodiment proposes a screening method based on an energy-weighted kurtosis comprehensive metric, using EWK as a comprehensive evaluation index. This method can robustly filter out IMF components containing the most significant fault information.

[0030] In some embodiments, the first at least one IMF component with the largest EWK value is selected as the key IMF component; the formula for calculating the EWK value is: ; in, η k For the first k Energy weights of each IMF component; K k For the first k Kurtosis values ​​of each IMF component; ρ k For the first k The correlation coefficient between each IMF component and the original vibration signal. μ IMF component signal x The mean, σ For signal x The standard deviation. Subscript k Indicates the first k One IMF component; E It represents energy.

[0031] In some embodiments, feature vector construction specifically involves calculating the energy entropy of each key IMF component to construct the feature vector. H 1, H 2, ..., H n The definition of energy entropy is: ; in, n The number of modes; For the first i Energy entropy of a key IMF component; E i For the first i The energy of each key IMF component, i.e., the envelope amplitude; For the first i The energy share of each key IMF component.

[0032] In some embodiments, model training and optimization specifically include: Dataset partitioning: The feature vector set of all samples is proportionally divided into a training set and a test set; GWO-SVM Model Establishment and Optimization: Using the training set, the GWO algorithm is used to simultaneously optimize the penalty factor and kernel parameters of the SVM model to establish a GWO-SVM fault diagnosis model. Model performance validation: Input the feature vectors of the test set samples into the trained GWO-SVM fault diagnosis model to evaluate the model's diagnostic accuracy.

[0033] This embodiment addresses the low efficiency of existing classifier parameter optimization: SVM is widely used in fault classification, but its performance is highly dependent on the settings of the penalty factor C and the kernel function parameter σ. Traditional methods such as grid search are inefficient and prone to getting trapped in local optima, failing to guarantee that the classifier is always in an optimal performance state, thus limiting further improvements in diagnostic accuracy. Therefore, this embodiment proposes using GWO to efficiently and globally optimize SVM hyperparameters, avoiding the blindness of manual selection and the local optima problem of traditional optimization methods, ensuring that the classifier is always in or close to its optimal performance state.

[0034] In some embodiments, the GWO algorithm is guided by a fitness function, which is the classification accuracy of the SVM model on the test set; the population size of the GWO algorithm is 30, the maximum number of iterations is 200, and the search range of the penalty factor and kernel parameters of the SVM model is [0.01, 100]. For high-dimensional fault features, the SVM model uses a radial basis function kernel, the expression of which is: ; in, x i and x j For any two input feature vectors, || || represents Euclidean distance; σ These are kernel function parameters used to control the complexity and mapping capability of the model. C This is a penalty factor for synchronous optimization, used to control the trade-off between the tolerance for classification errors and model complexity.

[0035] The process of optimizing SVM parameters using the GWO algorithm includes: Step 1: Initialize the gray wolf population. The position vector of each individual gray wolf represents a set of SVM parameters. C , σ ); Step 2: Calculate the fitness value of each individual, which is the classification accuracy on the validation set after training the SVM with the parameters it represents; Step 3: Determine the population's fitness values. α , β , δ Wolf; Step 4: Update based on the positions of α, β, and δ wolves ω The wolf's location; Step 5: Iterate through steps 2 to 4 until the maximum number of iterations is reached, and output the optimal result. C and σ .

[0036] In some embodiments, a rolling bearing fault diagnosis method is also provided, which is based on the fault diagnosis model constructed according to any of the above, and the diagnosis method includes: Based on the vibration signal to be diagnosed, a Bayesian optimization algorithm is used to construct an objective optimization function based on the envelope entropy of the IMF components obtained by VMD decomposition. The online VMD parameters are then optimized and obtained within a preset parameter search space. The online VMD parameters are then used to perform VMD decomposition on the vibration signal to be diagnosed. Based on the energy entropy of the IMF components obtained by online VMD parameter decomposition, a feature vector to be diagnosed is constructed. The feature vector to be diagnosed is input into the fault diagnosis model, and the fault category diagnosis result is output.

[0037] In this embodiment, during online diagnosis, for samples with unknown states, a Bayesian optimization algorithm is used to determine the VMD parameters in real time and decompose them to obtain a series of IMFs. These IMFs are then filtered using EWK (Extended Wake-Kill) to construct feature vectors for fault diagnosis. Online optimization of VMD parameters can be performed during each diagnosis, or the frequency of online optimization can be set to once every preset time interval; the specific frequency is not limited. Online VMD parameter optimization further ensures the decomposition effect, which is beneficial for ensuring the effectiveness of fault feature extraction and the accuracy of model diagnosis.

[0038] In some embodiments, the objective optimization function is to minimize the sum of the envelope entropies of the IMF components obtained from VMD decomposition. Before constructing the feature vector to be diagnosed, the following steps are also included: For the IMF components obtained by decomposition under online VMD parameters, the EWK value of each IMF component is calculated, and the top at least one online key IMF component is selected based on the EWK value; wherein, the EWK value is the product of the energy weight, kurtosis value and correlation coefficient of the IMF component. Accordingly, the construction of the feature vector to be diagnosed is specifically as follows: based on the energy entropy of the online key IMF components, the feature vector to be diagnosed is constructed.

[0039] In some embodiments, an electronic device is also provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the program to implement the steps of the method for constructing the rolling bearing fault diagnosis model described in any of the preceding embodiments.

[0040] In some embodiments, considering the challenges of collaboratively addressing the three stages of feature extraction, feature selection, and classifier parameter optimization in existing technologies to form an adaptive, high-precision, and robust fault diagnosis process, this is a pressing technical problem in the field. This embodiment proposes a rolling bearing fault diagnosis method based on Variational Mode Decomposition (VMD), Energy Weighted Kurtosis (EWK), and Grey Wolf Optimizer (GWO) optimization of Support Vector Machine (SVM), specifically a VMD-EWK-GWO-SVM-based rolling bearing fault diagnosis method. This method determines the optimal VMD parameters through Bayesian optimization, proposes a comprehensive feature selection index, and utilizes an efficient optimization algorithm to fine-tune the classifier parameters, ultimately achieving accurate diagnosis of early bearing faults even in strong noise environments.

[0041] Specifically, a rolling bearing fault diagnosis method based on VMD-EWK-GWO-SVM is presented. The model parameters are trained using a bearing vibration data training set. Data acquisition and preprocessing include: using drive-end bearing vibration signals publicly available from the Bearing Data Center of Case Western Reserve University (CWRU). Data was selected at a motor speed of 1750 r / min and a sampling frequency of 12 kHz. Data categories included normal state, inner race fault, outer race fault, and rolling element fault. For each state, 100 samples were extracted, each with 1024 data points. To facilitate model training, the sample data were normalized, with amplitudes normalized to the [-1, 1] interval.

[0042] First, variational mode decomposition (VMD) is used to adaptively decompose the bearing vibration signal, yielding a series of intrinsic mode functions (IMFs). Then, an energy-weighted kurtosis (EWK) index is proposed to robustly screen key IMF components containing significant fault characteristics. Based on the screened IMF components, their energy entropy is calculated to construct feature vectors. Finally, the Grey Wolf Optimizer (GWO) algorithm is used to apply a penalty factor to the Support Vector Machine (SVM). C and kernel function parameters σ Synchronous global optimization is performed to establish a high-precision fault classification model. To verify the effectiveness of the method, a widely used public bearing dataset (such as the Case Western Reserve University bearing dataset) is used for demonstration. It should be noted that the method in this embodiment is applicable to vibration signal analysis of various types of rolling bearings and is not limited to a specific dataset.

[0043] Data preparation: Vibration signals of the motor drive bearing at a speed of 1750 r / min and a sampling frequency of 12 kHz were selected from the CWRU database. Four health states were selected: normal, inner race fault, outer race fault, and rolling element fault. For each state, 100 consecutive samples were extracted, each containing 1024 data points. Subsequently, all sample data were normalized, and their amplitudes were mapped to the interval [-1, 1].

[0044] VMD decomposition: In the parameter search space ( K ∈[2,12], integer; α Within the range [500, 5000], a Bayesian optimization algorithm was used to optimize the VMD parameters, determining the optimal VMD parameter combination for different health states. Based on the optimal parameters, VMD decomposition was performed on each sample. Taking the inner ring fault sample as an example, five IMF components were obtained. The time-domain waveform and frequency-domain spectrum after decomposition are shown below. Figure 2 and Figure 3 As shown.

[0045] Feature selection and construction: The EWK value of each IMF component after VMD decomposition of each sample is calculated. The top 5 IMF components with the largest EWK values ​​are selected, and their energy entropy is calculated to construct a 5-dimensional feature vector [H1, H2, H3, H4, H5]. Feature vectors are constructed for all four states and assigned class labels (1, 2, 3, 4). Finally, a feature dataset containing 400 samples (4 classes × 100) is obtained, as shown in Table 1.

[0046] Table 1 Feature Vectors

[0047] GWO-SVM Model Training and Testing: The feature dataset of the aforementioned 400 samples was randomly divided into a training set (320 samples) and a test set (80 samples). Using the training set data, the GWO algorithm was employed to optimize the SVM parameters, with a population size of 30 and a maximum of 200 iterations. C and σ The search range is [0.01, 100]. After the optimization process converges, the optimal parameters are obtained. C = 73.2021, σ = 29.2215. Use this parameter to train the final SVM model.

[0048] Diagnostic results and analysis: The test set is input into the trained GWO-SVM model for diagnostics. The resulting confusion matrix is ​​as follows: Figure 6 As shown, all 80 test samples were correctly classified, achieving 100% accuracy. In comparison, under the same data and features, SVM optimized using Particle Swarm Optimization (PSO) and Whale Optimization Algorithm (WOA) both achieved an accuracy of 98.75% (see [link to relevant documentation]). Figure 4 and Figure 5 Each of the above methods had one false positive, and the performance comparison is shown in Table 2 below. This indicates that the GWO algorithm used in this invention has a greater advantage in SVM parameter optimization, thereby achieving higher diagnostic accuracy.

[0049] Table 2 Performance Comparison of Different Fault Diagnosis Methods

[0050] Online diagnosis: For a new vibration signal sample in an unknown state, real-time Bayesian optimization is performed to determine its VMD parameters as [7, 1469.4]. VMD decomposition yields 7 IMF components. Five key IMF components are selected, energy entropy features are extracted, feature vectors are constructed, and input into the pre-trained GWO-SVM model. The output fault category label "4" represents a rolling element fault.

[0051] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements 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 method for constructing a rolling bearing fault diagnosis model, characterized in that, include: Signal acquisition and preprocessing: Vibration signals of rolling bearings under different health conditions are acquired, and the signals are divided into sliding windows and normalized according to a preset window length to establish a dataset; Parameter-adaptive VMD decomposition: For vibration signals of each health state type, an objective function is constructed based on the envelope entropy of the IMF components obtained by VMD decomposition to optimize and obtain the corresponding optimal VMD parameters, and vibration signal decomposition is performed based on the optimal VMD parameters; Feature vector construction: For each health state type, a feature vector is constructed based on the energy entropy of the IMF components obtained by decomposition under the optimal VMD parameters; Model training and optimization: Based on the dataset, with feature vectors as input and fault type as output, a fault diagnosis model is obtained through training and optimization.

2. The method for constructing a rolling bearing fault diagnosis model as described in claim 1, characterized in that, The parameter adaptive VMD decomposition specifically includes: For each health state type, a VMD parameter optimization space is constructed, which includes the number of modes and the penalty factor. The search space for the number of modes is defined as the set of integers [2, 12], and the search space for the penalty factor is defined as the continuous interval [500, 5000]. The objective function is to minimize the average sum of the envelope entropies of the vibration signals for each health state type with respect to the decomposed IMF components. Using a Bayesian optimization algorithm, with the objective function as the fitness function, an iterative search is performed in the parameter optimization space to obtain the optimal VMD parameters corresponding to each health state type.

3. The method for constructing a rolling bearing fault diagnosis model as described in claim 1, characterized in that, Before feature vector construction, the following also applies: Key IMF component screening based on EWK: For each health state type, based on the IMF components obtained by decomposition under the optimal VMD parameters, calculate the EWK value of each IMF component, and screen out the top at least one key IMF component based on the EWK value; wherein, the EWK value is the product of the energy weight, kurtosis value and correlation coefficient of the IMF component. Accordingly, the feature vector construction is specifically as follows: feature vectors are constructed based on the energy entropy of key IMF components.

4. The method for constructing a rolling bearing fault diagnosis model as described in claim 3, characterized in that, The first at least one IMF component with the largest EWK value is selected as the key IMF component; the formula for calculating the EWK value is: ； in, η k For the first k Energy weights of each IMF component; K k For the first k Kurtosis values ​​of each IMF component; ρ k For the first k The correlation coefficient between each IMF component and the original vibration signal.

5. The method for constructing a rolling bearing fault diagnosis model as described in claim 3, characterized in that, The feature vector construction specifically involves calculating the energy entropy of each key IMF component to construct the feature vector. H 1, H 2, ..., H n The energy entropy value is defined as follows: ； in, n The number of modes; For the first i Energy entropy of a key IMF component; E i For the first i The energy of each key IMF component, i.e., the envelope amplitude; For the first i The energy share of each key IMF component.

6. The method for constructing a rolling bearing fault diagnosis model as described in claim 1, characterized in that, Model training and optimization specifically include: Dataset partitioning: The feature vector set of all samples is proportionally divided into a training set and a test set; GWO-SVM Model Establishment and Optimization: Using the training set, the GWO algorithm is used to simultaneously optimize the penalty factor and kernel parameters of the SVM model to establish a GWO-SVM fault diagnosis model. Model performance validation: Input the feature vectors of the test set samples into the trained GWO-SVM fault diagnosis model to evaluate the model's diagnostic accuracy.

7. The method for constructing a rolling bearing fault diagnosis model as described in claim 6, characterized in that, The GWO algorithm uses a fitness function to guide the search, where the fitness function is the classification accuracy of the SVM model on the test set; the population size of the GWO algorithm is 30, the maximum number of iterations is 200, and the search range of the penalty factor and kernel parameters of the SVM model is [0.01, 100]. The SVM model uses a radial basis function kernel, the expression of which is: ； in, x i and x j For any two input feature vectors, || || represents Euclidean distance; σ These are kernel function parameters used to control the complexity and mapping capability of the model.

8. A method for diagnosing rolling bearing faults, characterized in that, Based on the fault diagnosis model constructed according to any one of claims 1-7 above, the diagnosis method includes: Based on the vibration signal to be diagnosed, a Bayesian optimization algorithm is used to construct an objective optimization function based on the envelope entropy of the IMF components obtained by VMD decomposition. The online VMD parameters are then optimized and obtained within a preset parameter search space. The online VMD parameters are then used to perform VMD decomposition on the vibration signal to be diagnosed. Based on the energy entropy of the IMF components obtained by online VMD parameter decomposition, a feature vector to be diagnosed is constructed. The feature vector to be diagnosed is input into the fault diagnosis model, and the fault category diagnosis result is output.

9. The rolling bearing fault diagnosis method as described in claim 8, characterized in that, The objective optimization function is to minimize the sum of the envelope entropy of the IMF components obtained from VMD decomposition. Before constructing the feature vector to be diagnosed, the following steps are also included: For the IMF components obtained by decomposition under online VMD parameters, the EWK value of each IMF component is calculated, and the top at least one online key IMF component is selected based on the EWK value; wherein, the EWK value is the product of the energy weight, kurtosis value and correlation coefficient of the IMF component. Accordingly, the construction of the feature vector to be diagnosed is specifically as follows: based on the energy entropy of the online key IMF components, the feature vector to be diagnosed is constructed.

10. An electronic device, characterized in that, The system includes a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the program to implement the steps of the method for constructing a rolling bearing fault diagnosis model as described in any one of claims 1 to 7.

Images