Rolling bearing performance degradation evaluation method based on kernel optimal global locality preserving projection and Sinkhorn distance

Abstract

The invention relates to the technical field of rolling bearing state monitoring, and discloses a rolling bearing performance degradation evaluation method based on kernel optimal global locality preserving projection and Sinkhorn distance, which comprises the following steps: acquiring an original vibration signal of a rolling bearing, performing feature extraction on the original vibration signal, and constructing a high-dimensional feature set; calculating and screening features in the high-dimensional feature set, and constructing a sensitive feature set; performing dimension reduction processing on the sensitive feature set by adopting a kernel optimal global locality preserving projection method to obtain a low-dimensional feature matrix; in the low-dimensional feature matrix, the difference between samples is calculated based on the optimal transmission distance of entropy regularization, and a Sinkhorn distance matrix is constructed; on the basis of the Sinkhorn distance matrix, constructing a degradation index; a health threshold is set, an early degradation point is judged through the degradation index and the health threshold, rolling bearing fault recognition and performance evaluation are achieved in combination with the change trend of the degradation index value, and technical support can be provided for health monitoring and intelligent maintenance of rotary mechanical equipment.

Description

Technical Field

[0001] This invention relates to the field of rolling bearing condition monitoring technology, and in particular to a rolling bearing performance degradation assessment method based on the kernel-optimal global local preservation projection and the Sinkhorn distance. Background Technology

[0002] Rolling bearings are critical components in rotating machinery, and their operating condition directly affects the stability and safety of the equipment. During service, bearings are subjected to various operating conditions such as load impact, vibration, poor lubrication, and high temperatures, making them prone to failures such as fatigue spalling, wear, cracking, and deformation. Once a failure occurs, it not only degrades equipment performance but may also trigger a chain reaction of safety accidents. Therefore, conducting performance degradation assessments and fault early warning systems for rolling bearings has significant theoretical and practical value for achieving intelligent maintenance and reliable operation of equipment.

[0003] Currently, bearing degradation assessment methods mainly include methods based on empirical features to construct health indicators and prediction methods based on data-driven modeling. The former typically uses time-domain, frequency-domain, time-frequency-domain, or entropy-based features to describe the bearing degradation state through statistical regularities, offering good interpretability and physical meaning. However, under complex operating conditions or in the early stages of degradation, these features are easily affected by noise, making it difficult to accurately characterize the degradation trend. The latter, such as neural networks, support vector machines, and deep learning models, can automatically learn the mapping relationship between features and degradation patterns, but they are highly dependent on training samples and have poor model interpretability, making them difficult to promote and apply in practical engineering.

[0004] Furthermore, in feature processing, existing methods often employ traditional dimensionality reduction techniques such as Principal Component Analysis (PCA) or Kernel Principal Component Analysis (KPCA). While these methods can extract global structural information, they tend to overlook local fluctuations in the signal that reflect faults. Local Preserving Projection (LPP), although retaining some local information, cannot effectively reflect global trends. Therefore, it is difficult to construct health indicators that simultaneously possess global robustness and local sensitivity. At the same time, commonly used metrics such as Euclidean distance and Mahalanobis distance are not sensitive enough to the distributional differences in high-dimensional data, and cannot be effectively used to accurately characterize degradation indicators. In recent years, Wasserstein distance and optimal transport theory have been introduced into industrial health assessment to better measure the distributional differences between different states. However, the high computational complexity of Wasserstein distance limits its real-time application. Therefore, Sinkhorn distance with entropy regularization is introduced as an alternative, combining efficiency and discriminative power. Meanwhile, in terms of feature fusion and dimensionality reduction, how to preserve both global and local structural information is also a challenge in constructing degradation indicators. Summary of the Invention

[0005] The purpose of this invention is to provide a rolling bearing performance degradation assessment method that integrates KOGLPP and Sinkhorn distance, which can accurately identify the degradation trend and early failure points of the bearing operating state, and has strong sensitivity, accuracy and practicality.

[0006] To achieve the above objectives, the present invention provides the following solution: A method for evaluating the performance degradation of rolling bearings based on the kernel-optimal global-local-preserving projection and the Sinkhorn distance includes: The original vibration signals of the rolling bearing throughout its entire life cycle are obtained, and features are extracted from the original vibration signals to construct a high-dimensional feature set. The features in the high-dimensional feature set are calculated and filtered to construct a sensitive feature set; The sensitive feature set is reduced in dimensionality using the kernel-optimal global-local-preserving projection method to obtain a low-dimensional feature matrix. In the low-dimensional feature matrix, the difference between each sample is calculated based on the optimal transmission distance with entropy regularization, and a Sinkhorn distance matrix is ​​constructed. A degradation index is constructed based on the Sinkhorn distance matrix; A health threshold is set, and early degradation points are determined by comparing degradation indicators with the health threshold. The changing trend of degradation indicator values ​​is combined to achieve rolling bearing fault identification and performance degradation assessment.

[0007] Furthermore, the feature extraction specifically includes: Multidimensional features in the time domain, frequency domain, and entropy domain are extracted from the original vibration signal. The high-dimensional feature set includes: mean, root mean square value, peak factor, waveform factor, margin factor, weighted frequency amplitude, sample entropy, Shannon entropy, Renyi entropy, and Tsallis entropy.

[0008] Furthermore, the features in the high-dimensional feature set are calculated and filtered to construct a sensitive feature set, specifically as follows: Calculate the monotonicity, correlation, and trend indices of each feature in a high-dimensional feature set; The weighting coefficient of each feature in the above three indicators is calculated by the entropy method to construct a comprehensive selection index; Features with comprehensive selection indicators greater than the selection threshold are selected to construct a sensitive feature set.

[0009] Furthermore, the calculation formulas for the monotonicity, correlation, and trend indicators are as follows: in, As a feature, This represents the difference in feature values ​​between two adjacent samples. For the number of times, The total number of samples, For the sample size, For the first Features in the The value of the sample For the first The mean of the feature across all samples. For the first Features in the The value of the sample For the first Features in all The mean of a sample To indicate the first The feature sequence corresponding to each feature For the first The feature sequence corresponding to each feature.

[0010] Furthermore, the formula for the comprehensive selection index is as follows: in, For importance weights.

[0011] Furthermore, the sensitive feature set is dimensionality reduced using the kernel-optimal global-local-preserving projection method to obtain the low-dimensional feature matrix, specifically: Dimensionality reduction is achieved by constructing a target matrix, specifically as follows: in, The global scatter matrix, is the local divergence matrix, and α is a weighting parameter that adjusts the weights of global and local information. During dimensionality reduction, the target matrix is ​​decomposed into eigenvalues, and the eigenvectors corresponding to the first m largest eigenvalues ​​are selected as projection directions to map the sensitive feature set into a low-dimensional feature matrix.

[0012] Furthermore, the Sinkhorn distance matrix is ​​specifically as follows: in, It is a joint probability transfer matrix. It is the cost matrix between samples. express entropy, This is the regularization parameter.

[0013] Furthermore, the degradation index is constructed based on the Sinkhorn distance matrix as follows: Calculate the Sinkhorn distance between the first component and other components of each sample in the reduced-dimensional space, construct the Sinkhorn distance matrix, and take the maximum value of each row or column in the Sinkhorn distance matrix as the degradation degree of the corresponding sample. Define this value as the degradation index.

[0014] Furthermore, the health threshold is set based on Chebyshev's inequality, and the formula is: in, Represents the probability of any event occurring. This represents a characteristic sequence indicating a healthy operating state. and They represent The mean and standard deviation, To set parameters, , The standard deviation of the characteristic sequence in the degenerate state. This represents the degradation threshold deviation.

[0015] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects: In this invention, the KOGLPP method simultaneously preserves the global structure and local change information of sensitive features. Combined with the Sinkhorn distance for accurate measurement of sample distribution differences, the constructed degradation index KOGSD can clearly and stably characterize the entire life-cycle degradation process of bearings from health to failure, avoiding the trend distortion problem caused by ignoring global or local information in traditional methods. This invention effectively amplifies early weak degradation signals by using three indicators—monotonism, correlation, and trend—and sensitive features screened by the entropy method. Furthermore, based on the health threshold set by Chebyshev's inequality, it can accurately capture the critical point of early bearing degradation, thus solving the problem of delayed response to early faults in traditional methods. This invention has high computational efficiency and wide applicability, and can provide accurate and reliable technical support for health monitoring and intelligent maintenance of rotating machinery, effectively reducing the risk of equipment failure cascades and ensuring operational stability and safety. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart of the rolling bearing performance degradation assessment method of the present invention; Figure 2 This is a waveform diagram of the rolling bearing's vibration throughout its entire life cycle, as presented in this invention. Figure 3 This is a diagram showing the feature extraction results of the present invention; Figure 4 This is a diagram showing the feature selection results of the present invention; Figure 5 This is a graph showing the construction results of the degradation index of the present invention; Figure 6 This is a graph showing the bearing degradation performance evaluation results of the present invention; Figure 7 The image shows the spectral analysis results of samples at different degradation stages of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0020] like Figure 1 As shown, this invention provides a method for evaluating the performance degradation of rolling bearings based on the kernel-optimal global-local preserving projection and the Sinkhorn distance, comprising the following steps: Step 1: Feature extraction: Extract time-domain, frequency-domain, and entropy-based multidimensional features from the original vibration signals throughout the entire life cycle of the rolling bearing, and construct a high-dimensional feature set F; Step 2: Feature selection: Calculate the monotonicity (Mon), correlation (Corr), and trend (Tre) indices for each feature in the high-dimensional feature set F, and construct a comprehensive selection index Z by calculating the weighting coefficients using the entropy method. Select features with Z values ​​greater than their mean A to construct a sensitive feature set X. Step 3: Feature Fusion: The sensitive feature set X is reduced in dimensionality using the Kernel Optimal Global Local Preserving Projection (KOGLPP) method to obtain a low-dimensional feature matrix Y. The KOGLPP method constructs a target matrix M containing a global scatter matrix and a local scatter matrix, and introduces a tradeoff parameter α to preserve both global structure and local features. Step 4: Distance Calculation: In the low-dimensional feature matrix Y, the difference between each sample is calculated based on the optimal transmission distance (Sinkhorn distance) with entropy regularization, and the Sinkhorn distance matrix D is constructed; Step 5: Degradation index construction: Define the degradation index KOGSD by taking the maximum value of each row or column in the distance matrix D as the degradation degree of the current sample; Step 6: Fault detection and degradation assessment: Based on the Chebyshev inequality principle, a health threshold HT is set. When the KOGSD value exceeds HT, it is determined to be an early degradation point, thus realizing the identification of rolling bearing faults and the assessment of performance degradation.

[0021] The high-dimensional feature set F in step 1 includes: mean (MV), root mean square (RMS), peak factor (PF), waveform factor (SF), margin factor (CF), weighted frequency amplitude (AnF), sample entropy (SE), Shannon entropy (HS), Renyi entropy (RE), and Tsallis entropy (TE).

[0022] The calculation formulas for Mon, Corr, and Tre in step 2 are as follows: in As a feature, This represents the difference in feature values ​​between two adjacent samples. For the number of times, The total number of samples, For the sample size, For the first Features in the The value of the sample For the first The mean of the feature across all samples. For the first Features in the The value of the sample For the first Features in all The mean of a sample To indicate the first The feature sequence corresponding to each feature For the first The feature sequence corresponding to each feature.

[0023] A comprehensive selection index Z is constructed by calculating weighted coefficients using the entropy method. Features with Z values ​​greater than their mean A are selected to form a sensitive feature set X. The entropy method is then used to calculate the information entropy of each feature under the three indices Mon, Corr, and Tre to determine the importance weight of each index. This leads to a weighted combination of Z: in, Using the entropy method to determine the weights of evaluation indicators is a common objective weighting method that automatically calculates the importance of each indicator based on the differences in information content, avoiding human subjectivity. First, for each feature... , , Normalize the [0,1] interval to eliminate the influence of dimensions. Secondly, for each indicator... Based on its normalized probability distribution Calculate the entropy value The entropy method formula is as follows: in, , The smaller the entropy value, the more concentrated and discriminative the information of the indicator, and the greater its weight should be. Finally, the weights are calculated by calculating the coefficient of variation. Then, calculate the weight of each item: in , .

[0024] In step 3, KOGLPP is a dimensionality reduction method that combines the advantages of KPCA and KLPP, aiming to preserve both the global structure and local similarity of the data. The core idea of ​​KOGLPP is to construct a target matrix that combines the strengths of KPCA and KLPP. And introduce trade-off parameters This balances the influence of global and local features to achieve better dimensionality reduction results. The specific steps are as follows: (1) Construct the target matrix When constructing the target matrix, KOGLPP comprehensively considers both the global and local scatter matrices, utilizing parameters. To weigh the two: in, It is the global scatter matrix, used to maintain the global structure of the data, similar to the centralized kernel matrix in KPCA. It primarily reflects the overall trend of data change in the feature space; It is a local scatter matrix, used to preserve the local structure of the data, equivalent to the similarity matrix constructed in KLPP. It primarily reflects the local similarity relationships between data samples; This is a tradeoff parameter used to adjust the weight of global and local information during the dimensionality reduction process. When When the size is large, focus more on the overall structure; when When the size is smaller, more attention is paid to local features.

[0025] (2) Eigenvalue decomposition After constructing the target matrix Then, the optimal projection direction is found by performing eigenvalue decomposition: in: These are eigenvectors, representing the projection direction after dimensionality reduction. It is an eigenvalue that reflects the magnitude of the divergence of the data in that direction.

[0026] Eigenvalue decomposition yields a series of eigenvectors and their corresponding eigenvalues. (Before selection...) The eigenvector corresponding to the largest eigenvalue is used as the optimal projection direction after dimensionality reduction.

[0027] (3) Dimensionality reduction mapping Projecting the original data into a lower-dimensional space ultimately achieves dimensionality reduction mapping: in: This is the data after dimensionality reduction, with dimensions of [dimension number missing]. ; It is the kernel matrix, representing the mapping of the original data in the kernel space; It is the choice of the former A matrix composed of eigenvectors represents the optimal projection direction.

[0028] KOGLPP combines the advantages of KPCA and KLPP, preserving both the global characteristics of the data and capturing its local variations. It is suitable for signals containing both global trends and rich local details. It can better distinguish between healthy and faulty states in a low-dimensional space, making bearing health and fault characteristics easier to identify in the reduced-dimensional space, reflecting the bearing degradation process. Therefore, it is an ideal dimensionality reduction method for constructing degradation indices.

[0029] In step 4, the Sinkhorn distance is defined as: in It is a joint probability matrix. It represents a probability distribution... Transition to another probability distribution All possible "transportation plans" satisfy the edge constraints: in, and These are the probability values ​​for the health and failure distributions. It is a cost matrix. Represents sample points and The distance between them is usually expressed using Euclidean distance. express The entropy. Parameters It is a regularization parameter used to control the strength of the entropy regularization term, thereby helping to improve computational efficiency.

[0030] The entropy regularization term is introduced to encourage a more dispersed transportation plan. A dispersed plan means that all the mass is not concentrated on a few paths, but rather "transported" through as many different paths as possible to improve computational efficiency. Therefore, by introducing the entropy regularization term, the Sinkhorn distance accelerates computation while maintaining the Wasserstein distance's ability to measure distributional differences.

[0031] In step 5, in the dimensionality reduction space Calculate the Sinkhorn distance between samples, and calculate the Sinkhorn distance between the first component and other components of each sample in the matrix. Construct the Sinkhorn distance matrix ,in and Corresponding dimensionality reduction space The Middle The first sample and the first One sample. Then select the distance matrix. The maximum distance between the bearing and its first component is used as the final degradation trend indicator KOGSD. This indicator can be used to monitor the health of the bearing, and a larger value usually indicates that the bearing is more severely degraded.

[0032] In step 6, the Chebyshev inequality principle is used to determine the health threshold. The relevant formulas are as follows. in, Represents the probability of any event occurring. This represents a characteristic sequence indicating a healthy operating state. and They represent The mean and standard deviation, To set the parameters, relevant literature was reviewed, and the following was selected: It indicates the health status of the bearing. Eigenvalues ​​in The probability of occurrence is greater than 96%. Since degradation follows a monotonically changing trend, only the upper threshold is considered, i.e., when it exceeds... ( The point at which the KOGSD value is determined is the early failure point, and then the degradation performance of the bearing is evaluated based on the changing trend of the KOGSD value.

[0033] Example 1 This embodiment utilizes an experimental platform for rolling bearing lifecycle data provided by the IMS Center at the University of Cincinnati. The platform's AC motor drives the shaft at a constant speed of 2000 r / min. Four Rexnord ZA-2115 bearings are mounted on the main shaft, each subjected to a radial load of 2700 kg. Vibration sensors (PCB353B33) are installed in both the horizontal and vertical directions of each bearing to collect vibration signals. A 6062E data acquisition card is used for data acquisition at a sampling frequency of 20 kHz, a sampling time of 1.024 s, and a sampling interval of 10 min. Using the full lifecycle data of bearing 1's outer ring from normal operation to failure, a total of 982 samples were collected. Based on the bearing parameters, the theoretical failure frequency of the bearing's outer ring can be obtained. . Figure 2 It is a time-domain waveform diagram of its full-life-cycle vibration data. The rolling bearing performance degradation assessment method described in this embodiment includes the following steps: Step 1: Feature Extraction: Extract the mean (MV), root mean square (RMS), peak factor (PF), waveform factor (SF), margin factor (CF), weighted frequency amplitude (AnF), sample entropy (SE), Shannon entropy (HS), Renyi entropy (RE), and Tsallis entropy (TE) from the raw vibration signal of the rolling bearing throughout its entire life cycle, and construct a high-dimensional feature set F, such as... Figure 3 The image shows the normalized feature map of the constructed multidimensional feature set F. The image reflects the complete operating state of the bearing represented by 17 different features, but not all features can well reflect the degradation process of the bearing.

[0034] Step 2: Feature Selection: Calculate the monotonicity (Mon), correlation (Corr), and trend (Tre) indices for each feature in the high-dimensional feature set F. Construct a comprehensive selection index Z by calculating weighted coefficients using the entropy method. Select features with Z values ​​greater than their mean A to construct a sensitive feature set X. The calculation formulas for Mon, Corr, and Tre are as follows: in, As a feature, This represents the difference in feature values ​​between two adjacent samples. frequency, The total number of samples, For the sample size, For the first Features in the The value of the sample For the first The mean of the feature across all samples. For the first Features in the The value of the sample For the first Features in all The mean of a sample To indicate the first The feature sequence corresponding to each feature For the first The feature sequence corresponding to each feature.

[0035] A comprehensive selection index Z is constructed by calculating weighted coefficients using the entropy method. Features with Z values ​​greater than their mean A are selected to form a sensitive feature set X. The entropy method is then used to calculate the information entropy of each feature under the three indices Mon, Corr, and Tre to determine the importance weight of each index. This leads to a weighted combination of Z: in, The entropy method is used to determine the weights of the evaluation indicators. First, for each feature... , , Normalize the [0,1] interval to eliminate the influence of dimensions. Secondly, for each indicator... Based on its normalized probability distribution Calculate the entropy value The entropy method formula is as follows: in, , The smaller the entropy value, the more concentrated and discriminative the information of the indicator, and the greater its weight should be. Finally, the weights are calculated by calculating the coefficient of variation. Then, calculate the weight of each item: in , .

[0036] like Figure 4 As shown in the figure, the monotonicity, correlation, and trend analysis of different features can be quantitatively observed. Some features exhibit good monotonicity, correlation, and trend, while others fail to reflect these characteristics. The feature plots also reflect this situation. Then, a weighted comprehensive selection index is calculated. and screening thresholds Select comprehensive selection indicators ( A sensitive degradation feature set is constructed from eight features (=0.5824). It includes the following features . Step 3: Feature Fusion: Combine the sensitive degradation feature set First, the sensitive indicator set Normalization Using radial basis kernel function and Construct the global scatter matrix and local divergence matrix and kernel matrix Then according to Construct the target matrix ,in Next, the target matrix Perform eigenvalue decomposition, select the eigenvectors corresponding to the top three largest eigenvalues ​​as the optimal projection directions, and then... The final low-dimensional matrix is ​​obtained by performing a mapping. .

[0037] Step 4: Distance Calculation: In the low-dimensional feature matrix Y, the difference between each sample is calculated based on the optimal transmission distance (Sinkhorn distance) with entropy regularization, and the Sinkhorn distance matrix D is constructed; Step 5: Degradation Index Construction: Using the maximum value of each row or column in the distance matrix D as the degradation degree of the current sample, define the degradation index KOGSD, such as... Figure 5 As shown in the figure, the degradation index KOGSD can effectively reflect the overall operating status of the bearing. After feature fusion through KOGLPP, it not only retains the global features of the data but also reflects the local changes in the data, and is also relatively sensitive to early degradation.

[0038] Step 6: Fault Detection and Degradation Assessment: Set the health threshold HT based on the Chebyshev inequality principle. The specific formula is as follows: in, Represents the probability of any event occurring. This represents a characteristic sequence indicating a healthy operating state. and They represent The mean and standard deviation, To set parameters, take The KOGSD characteristic value, which represents the health status of the bearing, is... The probability of occurrence is greater than 96%. Since degradation follows a monotonically changing trend, only the upper threshold is considered, i.e., when it exceeds... ( The point at which the KOGSD value is determined is the early failure point, and then the degradation performance of the bearing is evaluated based on the changing trend of the KOGSD value.

[0039] The health threshold is determined based on the Chebyshev inequality principle above. Points greater than this threshold are selected as early degradation points, such as... Figure 6 The figure shows a significant abrupt change in the KOGSD amplitude at sample 535, indicating an early failure point, consistent with numerous studies. The graph reveals that from sample 700 onwards, the bearing enters a moderate degradation stage, with a significant and sharp increase in the KOGSD value. During this stage, the amplitude initially rises and then falls, indicating the bearing is undergoing a "wear-smoothing" process, a normal phenomenon, and further suggesting that wear is gradually increasing. From sample 800 onwards, the KOGSD value gradually increases again, indicating the bearing has entered a severe degradation stage, with multiple "wear-smoothing" processes, and the wear has spread to multiple areas. From sample 900 onwards, the KOGSD amplitude surges dramatically, with each increase being substantial, indicating the bearing has entered the failure stage until complete failure.

[0040] To verify the accuracy of the rolling bearing degradation assessment, spectral analysis was performed on samples 200, 532, 533, 700, and 956, respectively. Figure 7 As shown, according to the spectrum analysis, no fault frequency was found at the 200th sample, indicating that the bearing was in a healthy state at this time; a weak outer ring fault frequency of 263Hz was found at the 532nd sample, and a significant outer ring fault frequency was found at the 533rd sample, indicating that the 533rd sample was indeed the early degradation point of the bearing; the 2nd to 5th harmonics of the fault frequency were clearly found at the 700th sample, indicating that the bearing damage had deepened and entered the moderate degradation stage; from the 800th sample, the amplitude of the 2nd to 5th harmonics of the fault frequency further increased, indicating that the bearing had entered the severe degradation stage; from Figure 7 The 922 samples show that in addition to the obvious 2nd to 5th harmonics of the fault frequency, there are many other frequency components with high amplitude, indicating that the bearing has been severely worn and has entered the failure stage. Thus, the bearing performance degradation assessment was completed through spectrum analysis.

[0041] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0042] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for evaluating the performance degradation of rolling bearings based on the kernel-optimal global-local preserving projection and the Sinkhorn distance, characterized in that, include: The original vibration signals of the rolling bearing throughout its entire life cycle are obtained, and features are extracted from the original vibration signals to construct a high-dimensional feature set. The features in the high-dimensional feature set are calculated and filtered to construct a sensitive feature set; The sensitive feature set is reduced in dimensionality using the kernel-optimal global-local-preserving projection method to obtain a low-dimensional feature matrix. In the low-dimensional feature matrix, the difference between each sample is calculated based on the optimal transmission distance with entropy regularization, and a Sinkhorn distance matrix is ​​constructed. A degradation index is constructed based on the Sinkhorn distance matrix; A health threshold is set, and early degradation points are determined by comparing degradation indicators with the health threshold. The changing trend of degradation indicator values ​​is combined to achieve rolling bearing fault identification and performance degradation assessment.

2. The method for evaluating the performance degradation of rolling bearings according to claim 1, characterized in that, The feature extraction specifically involves: Multidimensional features in the time domain, frequency domain, and entropy domain are extracted from the original vibration signal. The high-dimensional feature set includes: mean, root mean square value, peak factor, waveform factor, margin factor, weighted frequency amplitude, sample entropy, Shannon entropy, Renyi entropy, and Tsallis entropy.

3. The method for evaluating the performance degradation of rolling bearings according to claim 1, characterized in that, The process of calculating and filtering features in the high-dimensional feature set to construct a sensitive feature set is as follows: Calculate the monotonicity, correlation, and trend indices of each feature in a high-dimensional feature set; The weighting coefficient of each feature in the above three indicators is calculated by the entropy method to construct a comprehensive selection index; Features with comprehensive selection indicators greater than the selection threshold are selected to construct a sensitive feature set.

4. The method for evaluating the performance degradation of rolling bearings according to claim 3, characterized in that, The calculation formulas for the monotonicity, correlation, and trend indicators are as follows: in, As a feature, This represents the difference in feature values ​​between two adjacent samples. For the number of times, The total number of samples, For the sample size, For the first Features in the The value of the sample For the first The mean of the feature across all samples. For the first Features in the The value of the sample For the first Features in all The mean of a sample To indicate the first The feature sequence corresponding to each feature For the first The feature sequence corresponding to each feature.

5. The method for evaluating the performance degradation of rolling bearings according to claim 3, characterized in that, The formula for the comprehensive selection index is as follows: in, For importance weights.

6. The method for evaluating the performance degradation of rolling bearings according to claim 1, characterized in that, The sensitive feature set is dimensionality reduced using the kernel-optimal global-local-preserving projection method to obtain the low-dimensional feature matrix, specifically: Dimensionality reduction is achieved by constructing a target matrix, specifically as follows: in, The global scatter matrix, is the local divergence matrix, and α is a weighting parameter that adjusts the weights of global and local information. During dimensionality reduction, the target matrix is ​​decomposed into eigenvalues, and the eigenvectors corresponding to the first m largest eigenvalues ​​are selected as projection directions to map the sensitive feature set into a low-dimensional feature matrix.

7. The method for evaluating the performance degradation of rolling bearings according to claim 1, characterized in that, The Sinkhorn distance matrix is ​​specifically as follows: in, It is a joint probability transfer matrix. It is the cost matrix between samples. express entropy, This is the regularization parameter.

8. The method for evaluating the performance degradation of rolling bearings according to claim 1, characterized in that, The degradation index is constructed based on the Sinkhorn distance matrix as follows: Calculate the Sinkhorn distance between the first component and other components of each sample in the reduced-dimensional space, construct the Sinkhorn distance matrix, and take the maximum value of each row or column in the Sinkhorn distance matrix as the degradation degree of the corresponding sample. Define this value as the degradation index.

9. The method for evaluating the performance degradation of rolling bearings according to claim 1, characterized in that, The health threshold is set based on Chebyshev's inequality, and the formula is as follows: in, Represents the probability of any event occurring. This represents a characteristic sequence indicating a healthy operating state. and They represent The mean and standard deviation, To set parameters, , The standard deviation of the feature sequence in the degenerate state. This represents the degradation threshold deviation.

Images