Method for detecting wear resistance of bearing

Abstract

The invention belongs to the technical field of bearing wear resistance detection, and particularly relates to a method for detecting the wear resistance of a bearing. Comprising the steps of data acquisition, primary detection, secondary detection and comprehensive evaluation. Real-time temperature and sound data including environment temperature and background noise interference are acquired; in the primary detection, the temperature data is subjected to difference calculation, Kalman filtering and compensation correction, the sound data is subjected to time-spectrum analysis, frequency-band noise reduction and fusion to calculate a primary wear index, if the primary wear index is smaller than a first threshold value, the primary wear index is judged to be qualified, otherwise, the secondary detection is performed; rolling body, inner ring and outer ring slide way abrasion images are collected through secondary detection, and after preprocessing, secondary abrasion indexes are calculated through the double-branch convolutional network and non-abrasion images; and if the comprehensive wear degree is greater than a second threshold value, judging that the comprehensive wear degree is unqualified and giving an alarm, otherwise, judging that the comprehensive wear degree is qualified. The method dynamically eliminates environmental interference, combines multiple parameters and image features, and improves the detection accuracy and reliability.

Description

Technical Field

[0001] The present invention belongs to the technical field of bearing wear resistance detection, and in particular relates to a method for bearing wear resistance detection. Background Art

[0002] Bearings are a crucial component in modern mechanical equipment. Their primary function is to support rotating parts, reduce friction during movement, and ensure rotational accuracy. Bearing wear resistance directly impacts the operational stability, service life, and safety of the equipment. Therefore, accurate wear testing is crucial for ensuring reliable operation. Existing technologies often overlook dynamic characteristics that can interfere with environmental interference and fail to incorporate image features into wear testing, resulting in poor bearing wear testing results. Summary of the Invention

[0003] In view of the technical problems existing in the above background technology, the present invention proposes a method for detecting the wear resistance of bearings.

[0004] In order to achieve the above object, the technical solution adopted by the present invention includes the following steps:

[0005] Data acquisition: Acquiring real-time data during the bearing operation period, including temperature data and sound data. The data may contain interference parameters, including ambient temperature and background noise;

[0006] Primary detection: Pre-process the collected temperature and sound data, process the temperature change data and sound change data during the current detection period, and calculate the primary wear index. If the primary wear index is less than the first threshold, the bearing is judged to have qualified wear resistance. If it is greater than or equal to the first threshold, secondary detection is performed.

[0007] Secondary detection: Collecting bearing surface wear images and preprocessing them; the bearing surface wear images include rolling element images, inner ring slideway images, and outer ring slideway images. The bearing surface wear images are differentiated from pre-set wear-free images to obtain a secondary wear index.

[0008] Comprehensive evaluation: A weighted calculation is performed based on the primary wear index and the secondary wear index to determine the comprehensive wear degree. If the comprehensive wear degree is greater than the second threshold, the bearing wear resistance is judged to be unqualified and an alarm is triggered; otherwise, it is judged to be qualified and the test results are recorded.

[0009] Preferably, the specific implementation of preprocessing the collected temperature data and sound data in the primary detection includes:

[0010] Temperature data preprocessing is used to eliminate ambient temperature interference from the real-time temperature data collected during the operating period;

[0011] Sound data preprocessing is used to eliminate background noise interference from the real-time sound data collected during the runtime period.

[0012] Preferably, the temperature data preprocessing is specifically implemented as follows:

[0013] Collect real-time temperature of bearing outer ring And detect the real-time temperature of the environment , calculate the original difference between the bearing temperature and the ambient temperature, eliminate the common mode interference of the ambient temperature, and the difference sequence is ;

[0014] The smoothed value of the difference sequence is used as the state variable , based on Kalman filter recursion, the filtered smooth difference sequence is obtained: ,in is the optimal estimate at time t after Kalman filtering, is the optimal estimate at time t-1, is the Kalman filter gain;

[0015] When the deviation between the ambient temperature and the rated operating temperature of the bearing exceeds the set threshold, correction is required and the correction coefficient is calculated. First, define the ambient temperature deviation ,in To design the rated operating temperature of the bearing, a quadratic polynomial is used to fit the compensation coefficient , ,in are the constant term coefficient, the linear term coefficient and the quadratic term coefficient respectively;

[0016] Last effective temperature It is the superposition of the filtered difference after compensation and the ambient temperature, and the calculation method is .

[0017] Preferably, the sound data preprocessing is specifically implemented as follows:

[0018] The collected sound signals Perform short-time Fourier transform to obtain the time-frequency spectrum , expressed as The time-frequency spectrum value of a time frame at frequency f;

[0019] Calculate the core characteristic frequency based on the bearing speed. First calculate the fundamental frequency. , then divide the frequency band into effective frequency bands , noise frequency band ,in is the maximum frequency of the signal;

[0020] Initialize the noise power spectrum ,in is the initial noise power spectrum, is the total number of frames in the initial stable segment, and the noise power spectrum is updated ,in For the Frame frequency Real-time noise power spectrum at ;

[0021] Frequency band noise suppression, time spectrum Differentiate by frequency band to obtain the denoised time-frequency spectrum value Finally, the denoised time-frequency spectrum Perform inverse short-time Fourier transform to obtain the denoised bearing target sound signal.

[0022] Preferably, the processing of the temperature change data and the sound change data during the current detection period to calculate the primary wear index is specifically implemented as follows:

[0023] First, the pre-processed temperature data of the bearing during the normal operation period is selected, and the normal temperature trend curve is obtained by nonlinear least square fitting. , and calculate the standard deviation of temperature fluctuations during this period As a normal fluctuation threshold;

[0024] Calculate the instantaneous deviation of the temperature data of the current detection period from the normal temperature trend curve ,in The temperature data of the current detection period after preprocessing is used, and the slopes of the current temperature trend curve and the normal temperature trend curve are obtained by linear approximation. ;

[0025] Calculate the initial value of temperature wear ,in, The instantaneous deviation value of the current detection period The average value of

[0026] Then, the correlation coefficient R between the temperature trend and the sound trend is calculated for the pre-processed sound data to obtain the sound adjustment coefficient ,in is the baseline correlation coefficient between temperature and sound trends during normal operation;

[0027] The product of the initial temperature wear value and the sound adjustment coefficient is normalized to obtain the primary wear index.

[0028] Preferably, the image preprocessing in the secondary detection is to perform histogram equalization on the image, thereby changing the histogram distribution of the image and increasing the image contrast.

[0029] Preferably, in the secondary detection, performing differential calculation on the bearing surface wear image and the preset non-wear image to obtain the secondary wear index is specifically implemented as follows:

[0030] Feature differentiation is performed on the pre-processed bearing surface wear image and the preset non-wear image to obtain independent sub-images of the rolling element, inner ring slideway, and outer ring slideway;

[0031] A dual-branch contrast convolutional network is constructed. The worn sub-image and the unworn sub-image of the corresponding area are input into the two branches of the network respectively to extract the wear feature vector and the baseline feature vector.

[0032] Perform differentiation operation on the wear feature vector of each region and the reference feature vector to obtain regional difference features;

[0033] The regional difference features are weightedly fused to obtain the global difference features, which are then input into the fully connected layer for mapping and normalized as the secondary wear index.

[0034] Preferably, the dual-branch contrast convolutional network is constructed, and the worn sub-image and the non-worn sub-image of the corresponding area are input into the two branches of the network respectively, and the wear feature vector and the reference feature vector are extracted as follows:

[0035] First, the sub-images of each region are scaled and uniformly adjusted to the preset size;

[0036] A two-branch contrastive convolutional network was constructed, with each branch consisting of five convolutional layers and two residual connection modules. The first branch inputs the worn sub-image and extracts local wear features using a 3×3 convolution kernel. The second branch inputs the unworn sub-image and extracts baseline features using the same network structure.

[0037] The output features of the two branches are globally averaged and pooled to obtain the wear feature vector and the baseline feature vector with consistent dimensions.

[0038] Preferably, performing a differential operation on the wear feature vector and the reference feature vector of each region to obtain the regional difference feature is specifically implemented as follows:

[0039] For the elements at the same position in the two vectors, calculate the absolute difference value (i.e., the absolute value of the difference between the two element values) and the weighted difference value (i.e., |wear element - reference element| × reference element). Sum the absolute difference value and the weighted difference value in an equally weighted manner to obtain the composite difference value of the element.

[0040] A 3×1 sliding window is used to slide along the vector dimension. The sum and maximum value of the composite difference values ​​within the window are calculated, and the average of the two is taken as the enhanced difference value at the center of the window. The elements at the edge of the window keep the original composite difference value unchanged to obtain the enhanced one-dimensional difference vector.

[0041] The enhanced one-dimensional difference vector is subjected to feature purification. The median of all elements in the vector is calculated. Elements greater than the median are retained as valid difference features, and elements less than or equal to the median are removed. The retained valid difference features are arranged in order of their original positions to form the regional difference features of the area.

[0042] Compared with the prior art, the advantages and positive effects of the present invention are:

[0043] 1. Temperature data is interpolated to eliminate common-mode interference, combined with Kalman filtering and smoothing, and environmental deviations are corrected using quadratic polynomial compensation coefficients. Sound data is subjected to short-time Fourier transform for frequency band noise reduction, and the noise power spectrum is dynamically updated, solving the problem of existing technologies ignoring the dynamic characteristics of environmental interference.

[0044] 2. Integrate temperature and sound data, fit the normal temperature trend curve, calculate the temperature deviation value and trend slope, introduce the correlation coefficient adjustment between the two, and form a primary wear index, overcoming the limitations of single parameter detection and improving the accuracy of preliminary judgment.

[0045] 3. Collect the bearing surface image, and after histogram equalization, use the dual-branch convolutional network to extract the wear and non-wear features, and obtain the secondary index through composite difference calculation.

[0046] 4. Weighted fusion of primary and secondary indices to obtain a comprehensive wear degree, with dual thresholds for judgment. Primary pass determination improves efficiency, while secondary detection enhances the accuracy of unqualified determination, comprehensively improving detection reliability. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following is a brief introduction to the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0048] Figure 1 The figure is a structural flow diagram of a method for detecting bearing wear resistance. DETAILED DESCRIPTION

[0049] In order to more clearly understand the above-mentioned objects, features and advantages of the present invention, the present invention is further described below in conjunction with the accompanying drawings and embodiments. It should be noted that, in the absence of conflict, the embodiments of the present application and the features therein can be combined with each other.

[0050] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways than those described herein. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0051] Embodiment: Bearings are key components that carry rotating motion in various mechanical equipment. Their wear resistance directly affects the operating stability, service life and safety of the equipment. Most existing bearing wear resistance detection methods rely on a single sensing technology, ignoring the dynamic interference of ambient temperature and background noise, and usually do not combine surface image features, resulting in low detection accuracy, high false alarm rate and missed alarm rate. Therefore, a method for bearing wear resistance detection is proposed, and the specific implementation steps are as follows: Figure 1 shown.

[0052] First, data acquisition is performed to obtain real-time data during the bearing operation period, including temperature data and sound data. The data may contain interference parameters, including ambient temperature and background noise.

[0053] Then, primary detection is performed, the collected temperature data and sound data are preprocessed, the temperature change data and sound change data of the current detection period are processed, and the primary wear index is calculated; if the primary wear index is less than the first threshold, it is determined that the bearing wear resistance is qualified; if it is greater than or equal to the first threshold, secondary detection is performed.

[0054] The specific implementation of preprocessing the collected temperature data and sound data in the primary detection includes: temperature data preprocessing, which is used to eliminate the ambient temperature interference of the real-time temperature data collected during the operating period; sound data preprocessing, which is used to eliminate the background noise interference of the real-time sound data collected during the operating period.

[0055] Specifically, the temperature data preprocessing is implemented as follows: collecting the real-time temperature of the bearing outer ring And detect the real-time temperature of the environment , calculate the original difference between the bearing temperature and the ambient temperature, eliminate the common mode interference of the ambient temperature, and the difference sequence is ; Take the smoothed value of the difference sequence as the state quantity , based on Kalman filter recursion, the filtered smooth difference sequence is obtained: ,in is the optimal estimate at time t after Kalman filtering, is the optimal estimate at time t-1, is the Kalman filter gain; when the deviation between the ambient temperature and the rated operating temperature of the bearing exceeds the set threshold, correction is required and the correction coefficient is calculated; first define the ambient temperature deviation ,in To design the rated operating temperature of the bearing, a quadratic polynomial is used to fit the compensation coefficient , ,in They are the constant term coefficient, the first term coefficient and the second term coefficient respectively; the final effective temperature It is the superposition of the filter difference after compensation and the ambient temperature, and the calculation method is .

[0056] The specific implementation of the sound data preprocessing is: Perform short-time Fourier transform to obtain the time-frequency spectrum , expressed as The time-frequency spectrum value of a time frame at frequency f; the core characteristic frequency is calculated based on the bearing speed, and the fundamental frequency is first calculated. , then divide the frequency band into effective frequency bands , noise frequency band ,in is the maximum frequency of the signal; initialize the noise power spectrum ,in is the initial noise power spectrum, is the total number of frames in the initial stable segment, and the noise power spectrum is updated ,in For the Frame frequency Real-time noise power spectrum at ;

[0057] Frequency band suppression of noise, time spectrum Differentiate by frequency band to obtain the denoised time-frequency spectrum value Finally, the denoised time-frequency spectrum Perform inverse short-time Fourier transform to obtain the denoised bearing target sound signal.

[0058] After preprocessing the temperature and sound data, it is necessary to process the temperature change data and sound change data of the current detection period to calculate the primary wear index. Specifically, first select the preprocessed temperature data of the historical normal operation period of the bearing, and use nonlinear least squares fitting to obtain the normal temperature trend curve. , and calculate the standard deviation of temperature fluctuations during this period As the normal fluctuation threshold; for the temperature data of the current detection period, calculate its instantaneous deviation from the normal temperature trend curve ,in The temperature data of the current detection period after preprocessing is used, and the slopes of the current temperature trend curve and the normal temperature trend curve are obtained by linear approximation. ;

[0059] Calculate the initial value of temperature wear ,in, The instantaneous deviation value of the current detection period Then, the trend slope of the curve of the current detection period is calculated for the pre-processed sound data. , calculate the correlation coefficient R between temperature trend and sound trend, and get the sound adjustment coefficient ,in is the baseline correlation coefficient between temperature and sound trends during normal operation; the product of the initial temperature wear value and the sound adjustment coefficient is normalized to obtain the primary wear index.

[0060] If the primary wear index is less than the first threshold, the bearing wear resistance is determined to be qualified; if it is greater than or equal to the first threshold, a secondary test is performed.

[0061] Secondary detection: Images of bearing surface wear are collected and preprocessed. These images include images of rolling elements, inner raceway, and outer raceway. The secondary wear index is calculated by differentiating the wear images from pre-set, wear-free images. Image preprocessing involves performing histogram equalization on the images, altering their distribution and increasing contrast.

[0062] The secondary wear index obtained by performing differential calculation on the bearing surface wear image and the preset wear-free image is to perform feature differentiation on the preprocessed bearing surface wear image and the preset wear-free image to obtain independent sub-images of the rolling element, the inner ring slideway, and the outer ring slideway; construct a dual-branch contrast convolutional network, input the wear sub-image and the wear-free sub-image of the corresponding area into the two branches of the network respectively, and extract the wear feature vector and the reference feature vector; perform differential operation on the wear feature vector and the reference feature vector of each area to obtain regional difference features; perform weighted fusion on the regional difference features to obtain global difference features, and input the global difference features into the fully connected layer for mapping and normalization as the secondary wear index.

[0063] Specifically, the preprocessed bearing surface wear image and the pre-set wear-free image are subjected to feature differentiation to obtain independent sub-images of the rolling element, inner raceway, and outer raceway. To efficiently compare and analyze the worn and non-worn areas, a two-branch comparative convolutional neural network is constructed. The two branches of the network are fed with a sub-image of the worn area and a non-worn sub-image of the corresponding area, respectively. Each branch uses the same convolutional and pooling layer structure to scale-normalize the sub-images of each area to a uniformly preset size. A two-branch comparative convolutional network is constructed, each consisting of five convolutional layers and two residual connection modules. The first branch takes the worn sub-image as input and extracts local wear features using a 3×3 convolution kernel. The second branch takes the non-worn sub-image and uses the same network structure to extract baseline features. The output features of both branches are globally average-pooled to obtain wear feature vectors and baseline feature vectors of consistent dimensions. These features are extracted through convolutional layers to extract local texture and shape information. Pooling layers perform spatial dimensionality reduction to reduce computational complexity while retaining important high-level features. The output of each branch is the feature vector for that area. For elements at the same position in the two vectors, the absolute difference value (i.e., the absolute value of the difference between the two elements) and the weighted difference value (|wear element - reference element| × reference element) are calculated respectively. The absolute difference value and the weighted difference value are summed in an equally weighted manner to obtain the composite difference value of the element. A 3×1 sliding window is used to slide along the vector dimension, and the sum and maximum value of the composite difference values ​​within the window are calculated. The average of the two is taken as the enhanced difference value at the center of the window. The original composite difference value of the elements at the edge of the window remains unchanged to obtain the enhanced one-dimensional difference vector. The enhanced one-dimensional difference vector is subjected to feature extraction, and the median of all elements in the vector is calculated. Elements greater than the median are retained as valid difference features, and elements less than or equal to the median are removed. The retained valid difference features are arranged in order of their original positions to form the regional difference features of the area. Finally, after calculating the difference features of each region, the difference features of each region are merged using a weighted fusion method to comprehensively evaluate the overall wear condition of the bearing. The weighted fusion method takes into account that different regions may have different importance in the wear process. Therefore, it is necessary to weight the difference features of each region according to the preset weight coefficient, sum up the weighted difference features of all regions to obtain the global difference features, and input the global difference features into the fully connected layer for mapping and normalization as the secondary wear index.

[0064] Finally, a comprehensive evaluation is performed, and a weighted calculation is performed based on the primary wear index and the secondary wear index to determine the comprehensive wear degree. If the comprehensive wear degree is greater than the second threshold, the bearing wear resistance is judged to be unqualified and an alarm is triggered; otherwise, it is judged to be qualified and the test results are recorded.

[0065] The above description is merely a preferred embodiment of the present invention and does not constitute any other form of limitation to the present invention. Any person skilled in the art may utilize the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes for application in other fields. However, any simple modification, equivalent change, and modification of the above embodiments made in accordance with the technical essence of the present invention without departing from the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A method for detecting bearing wear resistance, characterized in that: The following steps are involved: Data acquisition: Acquiring real-time data during the bearing operation period, including temperature data and sound data. The data may contain interference parameters, including ambient temperature and background noise; Primary detection: Pre-process the collected temperature and sound data, process the temperature change data and sound change data during the current detection period, and calculate the primary wear index. If the primary wear index is less than the first threshold, the bearing is judged to have qualified wear resistance. If it is greater than or equal to the first threshold, secondary detection is performed. Secondary detection: Collecting bearing surface wear images and preprocessing them; the bearing surface wear images include rolling element images, inner ring slideway images, and outer ring slideway images. The bearing surface wear images are differentiated from pre-set wear-free images to obtain a secondary wear index. Comprehensive evaluation: A weighted calculation is performed based on the primary wear index and the secondary wear index to determine the comprehensive wear degree. If the comprehensive wear degree is greater than the second threshold, the bearing wear resistance is judged to be unqualified and an alarm is triggered; otherwise, it is judged to be qualified and the test results are recorded.

2. A method for testing bearing wear resistance according to claim 1, characterized in that: The specific implementation of preprocessing the collected temperature data and sound data in the primary detection includes: Temperature data preprocessing is used to eliminate ambient temperature interference from the real-time temperature data collected during the operating period; Sound data preprocessing is used to eliminate background noise interference from the real-time sound data collected during the runtime period.

3. A method for testing bearing wear resistance according to claim 2, characterized in that: The specific implementation of the temperature data preprocessing is: Collect real-time temperature of bearing outer ring And detect the real-time temperature of the environment , calculate the original difference between the bearing temperature and the ambient temperature, eliminate the common mode interference of the ambient temperature, and the difference sequence is ; The smoothed value of the difference sequence is used as the state variable , based on Kalman filter recursion, the filtered smooth difference sequence is obtained: ,in is the optimal estimate at time t after Kalman filtering, is the optimal estimate at time t-1, is the Kalman filter gain; When the deviation between the ambient temperature and the rated operating temperature of the bearing exceeds the set threshold, correction is required and the correction coefficient is calculated. First, define the ambient temperature deviation ,in To design the rated operating temperature of the bearing, a quadratic polynomial is used to fit the compensation coefficient , ,in are the constant term coefficient, the linear term coefficient and the quadratic term coefficient respectively; Last effective temperature It is the superposition of the filter difference after compensation and the ambient temperature, and the calculation method is .

4. A method for testing bearing wear resistance according to claim 2, characterized in that: The specific implementation of the sound data preprocessing is: The collected sound signals Perform short-time Fourier transform to obtain the time-frequency spectrum , expressed as The time-frequency spectrum value of a time frame at frequency f; Calculate the core characteristic frequency based on the bearing speed. First calculate the fundamental frequency. , then divide the frequency band into effective frequency bands , noise frequency band ,in is the maximum frequency of the signal; Initialize the noise power spectrum ,in is the initial noise power spectrum, is the total number of frames in the initial stable segment, and the noise power spectrum is updated ,in For the Frame frequency Real-time noise power spectrum at ; Frequency band suppression of noise, time spectrum Differentiate by frequency band to obtain the denoised time-frequency spectrum value Finally, the denoised time-frequency spectrum Perform inverse short-time Fourier transform to obtain the denoised bearing target sound signal.

5. The method for testing bearing wear resistance according to claim 1, characterized in that: The specific implementation of processing the temperature change data and sound change data of the current detection period to calculate the primary wear index is as follows: First, the pre-processed temperature data of the bearing during the normal operation period is selected, and the normal temperature trend curve is obtained by nonlinear least square fitting. , and calculate the standard deviation of temperature fluctuations during this period As a normal fluctuation threshold; Calculate the instantaneous deviation of the temperature data of the current detection period from the normal temperature trend curve ,in The temperature data of the current detection period after preprocessing is used, and the slopes of the current temperature trend curve and the normal temperature trend curve are obtained by linear approximation. ; Calculate the initial value of temperature wear ,in, The instantaneous deviation value of the current detection period The average value of Then, the trend slope of the current detection period curve is calculated for the pre-processed sound data. , calculate the correlation coefficient R between temperature trend and sound trend, and get the sound adjustment coefficient ,in is the baseline correlation coefficient between temperature and sound trends during normal operation; The product of the initial temperature wear value and the sound adjustment coefficient is normalized to obtain the primary wear index.

6. The method for testing bearing wear resistance according to claim 1, characterized in that: The image preprocessing in the secondary detection is to perform histogram equalization on the image, thereby changing the histogram distribution of the image and increasing the image contrast.

7. The method for testing bearing wear resistance according to claim 1, characterized in that: The specific implementation of performing differential calculation on the bearing surface wear image and the preset non-wear image in the secondary detection to obtain the secondary wear index is as follows: Feature differentiation is performed on the pre-processed bearing surface wear image and the preset non-wear image to obtain independent sub-images of the rolling element, inner ring slideway, and outer ring slideway; A dual-branch contrast convolutional network is constructed. The worn sub-image and the unworn sub-image of the corresponding area are input into the two branches of the network respectively to extract the wear feature vector and the baseline feature vector. Perform differentiation operation on the wear feature vector of each region and the reference feature vector to obtain regional difference features; The regional difference features are weightedly fused to obtain the global difference features, which are then input into the fully connected layer for mapping and normalized as the secondary wear index.

8. A method for testing bearing wear resistance according to claim 7, characterized in that: The specific implementation of constructing a dual-branch contrast convolutional network, inputting the worn sub-image and the non-worn sub-image of the corresponding area into the two branches of the network respectively, and extracting the wear feature vector and the reference feature vector is as follows: First, the sub-images of each region are scaled and uniformly adjusted to the preset size; A two-branch contrastive convolutional network was constructed, with each branch consisting of five convolutional layers and two residual connection modules. The first branch inputs the worn sub-image and extracts local wear features using a 3×3 convolution kernel. The second branch inputs the unworn sub-image and extracts baseline features using the same network structure. The output features of the two branches are globally averaged and pooled to obtain the wear feature vector and the baseline feature vector with the same dimension.

9. The method for testing bearing wear resistance according to claim 7, characterized in that: The specific implementation of performing differentiation operation on the wear feature vector and the reference feature vector of each region to obtain the regional difference feature is as follows: For the elements at the same position in the two vectors, calculate the absolute difference value (i.e., the absolute value of the difference between the two element values) and the weighted difference value (i.e., |wear element - reference element| × reference element). Sum the absolute difference value and the weighted difference value in an equally weighted manner to obtain the composite difference value of the element. A 3×1 sliding window is used to slide along the vector dimension. The sum and maximum value of the composite difference values ​​within the window are calculated, and the average of the two is taken as the enhanced difference value at the center of the window. The elements at the edge of the window keep the original composite difference value unchanged to obtain the enhanced one-dimensional difference vector. The enhanced one-dimensional difference vector is subjected to feature purification. The median of all elements in the vector is calculated. Elements greater than the median are retained as valid difference features, and elements less than or equal to the median are removed. The retained valid difference features are arranged in order of their original positions to form the regional difference features of the area.

Images