A bearing surface defect detection system using a model fusion technique

Abstract

The present invention relates to the technical field of surface defect detection, and particularly to a bearing surface defect detection system based on model fusion technology, including a surface defect detection platform, a surface dimension evaluation unit, a model processing unit, an identification output unit, a comparison evaluation unit, and an intelligent management unit; The present invention initially analyzes whether there are defects in the target bearing from the perspective of surface dimensions, and at the same time further differentiates the difference regions from the perspective of simulation, so as to identify and analyze the position of the surface defect region. A surface defect detection model is constructed through the processing of the surface image data set, so as to accurately output the position and defect type of the surface defect of the target bearing. And through the way of information feedback, position comparison visual analysis is carried out on the defect position coordinates to ensure the accuracy and positioning visibility of the identification of the surface defect of the target bearing, and at the same time facilitate the correction of the simulation bearing model or the surface defect detection model with deviations, so as to improve the reliability and effectiveness of the bearing surface defect detection.

Description

Technical Field

[0001] The present invention relates to the technical field of surface defect detection, and particularly to a bearing surface defect detection system based on model fusion technology. Background Art

[0002] With the rapid development of the electronic manufacturing industry, as a basic component of the modern information industry, bearings are widely used in multiple manufacturing fields. Among them, bearings are mainly used to bear the weight of mechanical equipment and provide precise guidance for the rotation of automobile wheels. It bears both axial and radial loads and is a very important component in mechanical equipment. During the manufacturing process of bearings, due to uncertainties in raw materials, production environment, and equipment operation, various surface defects will inevitably occur.

[0003] In order to meet the usage requirements of bearings, bearing manufacturers will conduct strict quality inspections on the produced bearings. However, at present, the bearing quality inspection methods still largely use the method of manual visual inspection to detect surface defects on the bearing surface. This inspection method not only has low inspection efficiency, but also has a high false detection rate, and cannot accurately locate the position of surface defects, and at the same time cannot accurately understand the type of surface defects.

[0004] In view of the above technical defects, a solution is proposed. Summary of the Invention

[0005] The purpose of the present invention is to provide a bearing surface defect detection system based on model fusion technology to solve the above-mentioned technical defects. The present invention initially analyzes whether there are defects in the target bearing from the perspective of surface dimensions, and at the same time further distinguishes the difference regions from the perspective of the simulation model, so as to preliminarily analyze the position of the surface defect region. And a surface defect detection model is constructed through the processing of the surface image data set, so as to accurately output the position and defect type of the surface defect of the target bearing. And through the way of information feedback, a position comparison visual analysis is carried out on the defect position coordinates to ensure the accuracy and positioning visibility of the surface defect recognition of the target bearing, and at the same time facilitate the correction of the simulation bearing model or the surface defect detection model with deviations, so as to improve the reliability and effectiveness of the bearing surface defect detection.

[0006] The purpose of the present invention can be achieved through the following technical solutions: A bearing surface defect detection system based on model fusion technology includes a surface defect detection platform, a surface dimension evaluation unit, a model processing unit, an identification and output unit, a comparison and evaluation unit, and an intelligent management unit.

[0007] The surface defect detection platform is used to retrieve the basic information of the target bearing, and send the basic information to the surface dimension evaluation unit for surface dimension defect processing and simulation comparison analysis. The obtained surface dimension defect values are compared and analyzed to obtain a feedback signal or a dimension defect signal. When a feedback signal is generated, a simulation bearing model of the target bearing is constructed based on the basic information and compared and analyzed to obtain a qualified signal or an unqualified signal;

[0008] The model processing unit is used to perform model processing operations on the retrieved surface image dataset to obtain an adjustment instruction or a compliance instruction. When a compliance instruction is generated, a surface defect detection model is obtained; the recognition output unit is used to perform surface defect recognition and positioning processing on the surface feature image set of the target bearing collected to obtain the output result of the surface defect detection model; the comparison evaluation unit is used to perform position comparison and visual analysis on the defect position coordinates to obtain an accurate signal or a deviation signal.

[0009] Preferably, the surface dimension defect processing and simulation comparison analysis process is as follows:

[0010] The basic information of the target bearing on the detection workbench is obtained through laser scanning technology. The basic information includes the inner diameter and the outer diameter; the standard basic information of the target bearing is obtained. The parameters in the basic information are set as g, where g is a natural number greater than zero. The differences between the parameters in the basic information and the parameters in the corresponding standard basic information are obtained. The number of differences between the parameters in the basic information and the parameters in the corresponding standard basic information that are greater than the preset threshold is set as the surface dimension defect value, and the surface dimension defect value is compared and analyzed to obtain a feedback signal or a dimension defect signal.

[0011] Preferably, when a feedback signal is generated, a simulation bearing model of the target bearing is constructed based on the basic information. At the same time, the standard simulation bearing model of the target bearing is obtained. The difference analysis is performed between the simulation bearing model and the standard simulation bearing model to obtain the difference value between the simulation bearing model and the standard simulation bearing model. The difference value between the simulation bearing model and the standard simulation bearing model is judged and processed to obtain a qualified signal or an unqualified signal. When an unqualified signal is generated, the corresponding difference area between the simulation bearing model and the standard simulation bearing model is obtained and set as the model difference area.

[0012] Preferably, the model processing operation process is as follows:

[0013] The surface image dataset including normal and defective ones is obtained. The surface image dataset includes the end face image dataset, the outer circle face image dataset, and the inner circle face image dataset; the end face image dataset, the outer circle face image dataset, and the inner circle face image dataset are respectively divided into three parts, namely the training set, the test set, and the validation set;

[0014] Input each training set into the corresponding set algorithm model for training until the samples in the training set are exhausted, obtaining each pre-trained set algorithm model. Further, perform verification processing on each verification set and the corresponding set algorithm models to obtain the output results of each set algorithm model. The output results include overfitting, underfitting, and qualified fitting. If the output result is overfitting or underfitting, generate an improvement signal. If the output result is qualified fitting, generate an accurate signal.

[0015] Preferably, when generating an accurate signal, use the test set to test the set algorithm model corresponding to the accurate signal, obtain the ratio between the total number of errors and the total number of tests in the output result of the test of the set algorithm model corresponding to the accurate signal, and perform discrimination processing on the ratio between the total number of errors and the total number of tests in the output result to obtain an adjustment instruction or a compliance instruction. When generating a compliance instruction, set the set algorithm model corresponding to the compliance instruction as the end face defect detection model, the outer cylindrical surface defect detection model, and the inner cylindrical surface defect detection model, and fuse the end face defect detection model, the outer cylindrical surface defect detection model, and the inner cylindrical surface defect detection model to obtain a surface defect detection model.

[0016] Preferably, the surface defect recognition and positioning process is as follows:

[0017] Take the midpoint connection of the long sides of the detection workbench as the Y-axis, and the starting point of the midpoint connection of the long sides as the negative direction of the Y-axis. Take the midpoint connection of the short sides of the detection workbench as the X-axis, and the starting point of the midpoint connection of the short sides as the negative direction of the X-axis. Set the intersection point of the Y-axis and the X-axis as the origin, and take the ray perpendicular to the detection workbench surface starting from the origin as the Z-axis, thereby establishing a space coordinate system;

[0018] Obtain the surface feature image set of the target bearing located on the detection workbench through a high-definition camera. The surface feature image set includes end face feature images, outer cylindrical surface feature images, and inner cylindrical surface feature images, and perform preprocessing and grayscale processing on the surface feature image set. The preprocessing includes filtering and cropping. Input the preprocessed and grayscale processed surface feature image set into the surface defect detection model;

[0019] The surface defect detection model recognizes and matches the preprocessed and grayscale processed surface feature image set, and inputs the surface feature image set into the end face defect detection model, the outer cylindrical surface defect detection model, and the inner cylindrical surface defect detection model respectively for defect detection to obtain the output result of the surface defect detection model.

[0020] Preferably, the position comparison visual analysis process is as follows:

[0021] Obtain the defect position coordinates in the output result of the surface defect detection model, obtain the positions of the corresponding points on the simulated bearing model based on the defect position coordinates, and compare and analyze the positions of the corresponding points on the simulated bearing model with the model difference area: If the position of the corresponding point on the simulated bearing model belongs to the model difference area, generate an accurate signal; if the position of the corresponding point on the simulated bearing model does not belong to the model difference area, generate a deviation signal.

[0022] The beneficial effects of the present invention are as follows:

[0023] (1) The present invention initially analyzes whether the target bearing has defects from the perspective of surface dimensions, and at the same time further distinguishes the difference area from the perspective of the simulation model to initially analyze the position of the surface defect area. And a surface defect detection model is constructed through the processing of the surface image dataset to accurately output the position and defect type of the surface defect of the target bearing;

[0024] (2) Through the method of information feedback, a visual analysis of the position comparison of the defect position coordinates is carried out to ensure the accuracy and positioning visibility of the surface defect recognition of the target bearing. At the same time, it is convenient to correct the simulated bearing model or the surface defect detection model with deviations to improve the reliability and effectiveness of the bearing surface defect detection. Brief Description of the Drawings

[0025] The following further describes the present invention with reference to the accompanying drawings;

[0026] Figure 1 is the system flow block diagram of the present invention;

[0027] Figure 2 is the partial analysis reference diagram of the second embodiment of the present invention. Specific Embodiments

[0028] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0029] Embodiment 1:

[0030] Please refer to Figures 1 to 2As shown in the figure, the present invention is a bearing surface defect detection system based on model fusion technology, which includes a surface defect detection platform, a surface dimension evaluation unit, a model processing unit, an identification output unit, a comparison evaluation unit, and an intelligent management unit. The surface defect detection platform is unidirectionally communicatively connected to both the surface dimension evaluation unit and the model processing unit. The surface dimension evaluation unit is unidirectionally communicatively connected to the comparison evaluation unit. The model processing unit is unidirectionally communicatively connected to the identification output unit. The identification output unit is unidirectionally communicatively connected to both the comparison evaluation unit and the intelligent management unit. The comparison evaluation unit is unidirectionally communicatively connected to the intelligent management unit;

[0031] The surface defect detection platform is used to retrieve the basic information of the target bearing and send the basic information to the surface dimension evaluation unit for surface dimension defect processing and simulation comparison analysis, so as to preliminarily analyze whether there are defects in the target bearing from the perspective of surface dimensions, and at the same time distinguish the difference regions from the perspective of the simulation model, so as to preliminarily analyze the location of the surface defect region. The specific surface dimension defect processing and simulation comparison analysis process is as follows:

[0032] The basic information of the target bearing on the detection workbench is obtained through laser scanning technology. The basic information includes inner diameter, outer diameter, etc.;

[0033] At the same time, the standard basic information of the target bearing is obtained. The parameters in the basic information are set as g, where g is a natural number greater than zero. For example, when g = 1, it represents the inner diameter; when g = 2, it represents the outer diameter, and so on. The difference between each parameter in the basic information and the corresponding parameter in the standard basic information is obtained, and the number of differences between each parameter in the basic information and the corresponding parameter in the standard basic information that is greater than the preset threshold is set as the surface dimension defect value, and the surface dimension defect value is compared and analyzed:

[0034] If the surface dimension defect value is equal to zero, a feedback signal is generated;

[0035] If the surface dimension defect value is not equal to zero, a dimension defect signal is generated, and the dimension defect signal is sent to the intelligent management unit. After receiving the dimension defect signal, the intelligent management unit immediately displays the preset warning text corresponding to the dimension defect signal;

[0036] When a feedback signal is generated, a simulation bearing model of the target bearing is constructed based on the basic information. At the same time, the standard simulation bearing model of the target bearing is obtained, and the difference analysis is carried out between the simulation bearing model and the standard simulation bearing model to obtain the difference value between the simulation bearing model and the standard simulation bearing model, and the difference value between the simulation bearing model and the standard simulation bearing model is judged and processed:

[0037] If the difference value between the simulated bearing model and the standard simulated bearing model is less than the preset threshold value, a qualified signal is generated. When the qualified signal is generated, the preset warning text corresponding to the qualified signal is immediately displayed;

[0038] If the difference value between the simulated bearing model and the standard simulated bearing model is greater than or equal to the preset threshold value, an unqualified signal is generated. When the unqualified signal is generated, the corresponding difference region between the simulated bearing model and the standard simulated bearing model is obtained and set as the model difference region.

[0039] Embodiment 2:

[0040] The model processing unit is used to retrieve the surface image data set and perform model processing operations on the surface image data set at the same time, so as to accurately obtain the surface defect detection model, and at the same time process the defects in the model processing to improve the generalization ability of the model. The specific model processing operation process is as follows:

[0041] Construction of the end face defect detection model, the outer circle face defect detection model and the inner circle face defect detection model:

[0042] The surface image data set including normal and defective ones is obtained. The surface image data set includes the end face image data set, the outer circle face image data set and the inner circle face image data set;

[0043] The end face image data set, the outer circle face image data set and the inner circle face image data set are respectively divided into three parts, namely the training set, the test set and the validation set;

[0044] Each training set is input into the corresponding set algorithm model for training until the samples in the training set are exhausted, and each pre-trained set algorithm model is obtained. Further, each validation set is used to perform validation processing with the corresponding set algorithm model, and the output results of each set algorithm model are obtained. The output results include overfitting, underfitting and qualified fitting. If the output result is overfitting or underfitting, an improvement signal is generated. If the output result is qualified fitting, an accurate signal is generated;

[0045] When the improvement signal is generated, the hyperparameters of the set algorithm model corresponding to the improvement signal are adjusted. The hyperparameters include the learning rate, the regularization coefficient, etc.;

[0046] When the accurate signal is generated, the accurate signal corresponding set algorithm model is tested through the test set, and the ratio between the total number of errors and the total number of tests in the output result of the accurate signal corresponding set algorithm model test is obtained, and the ratio between the total number of errors and the total number of tests in the output result is discriminated:

[0047] If the ratio between the total number of errors and the total number of tests in the output result is greater than the preset threshold, an adjustment instruction is generated and sent to the intelligent management unit. After receiving the adjustment instruction, the intelligent management unit immediately displays the preset warning text corresponding to the adjustment instruction to adjust the set algorithm model corresponding to the adjustment instruction, so as to improve the generalization ability of the set algorithm model;

[0048] If the ratio between the total number of errors and the total number of tests in the output result is less than or equal to the preset threshold, when generating a passing instruction, the set algorithm model corresponding to the passing instruction is set as the end face defect detection model, the outer cylindrical surface defect detection model, and the inner cylindrical surface defect detection model, and the end face defect detection model, the outer cylindrical surface defect detection model, and the inner cylindrical surface defect detection model are fused to obtain a surface defect detection model;

[0049] The recognition output unit is used to perform surface defect recognition and positioning processing on the surface feature image set of the collected target bearing, so as to accurately identify the surface defect position and defect type of the target bearing. The specific surface defect recognition and positioning processing process is as follows:

[0050] Taking the midpoint connection line of the long side of the detection workbench as the Y-axis, and the starting point of the midpoint connection line of the long side as the negative direction of the Y-axis, taking the midpoint connection line of the short side of the detection workbench as the X-axis, and the starting point of the midpoint connection line of the short side as the negative direction of the X-axis, setting the intersection point of the Y-axis and the X-axis as the origin, and taking the ray perpendicular to the detection workbench surface starting from the origin as the Z-axis, thereby establishing a space coordinate system. It should be noted that the detection workbench is rectangular;

[0051] The surface feature image set of the target bearing located on the detection workbench is obtained through a high-definition camera. The surface feature image set includes an end face feature image, an outer cylindrical surface feature image, and an inner cylindrical surface feature image, and the surface feature image set is preprocessed and grayscale processed. The preprocessing includes filtering, cropping, etc. The preprocessed and grayscale processed surface feature image set is input into the surface defect detection model;

[0052] The surface defect detection model recognizes and matches the preprocessed and grayscale processed surface feature image set, inputs the surface feature image set into the end face defect detection model, the outer cylindrical surface defect detection model, and the inner cylindrical surface defect detection model for defect detection respectively, obtains the output result of the surface defect detection model, and sends the output result of the surface defect detection model to the intelligent management unit. After receiving the output result number, the intelligent management unit immediately displays the preset warning text corresponding to the output result;

[0053] The output result of the end face defect detection model is the defect position coordinates and their corresponding defect categories or normal signals in the upper end face feature image of the target bearing;

[0054] The output result of the outer cylindrical surface defect detection model is the defect position coordinates and their corresponding defect categories or normal signals in the feature image of the outer cylindrical surface of the target bearing;

[0055] The output result of the inner cylindrical surface defect detection model is the defect position coordinates and their corresponding defect categories or normal signals in the feature image of the inner cylindrical surface of the target bearing;

[0056] Flip the target bearing on the detection workbench to obtain the end face feature image of the flipped target bearing. Input the end face feature image into the surface defect detection model for recognition and matching, and input the surface feature image set into the end face defect detection model for defect detection;

[0057] In the embodiment of the present invention, the defect types include bruises, scratches, potholes, etc.;

[0058] The comparison and evaluation unit is used for visual analysis of position comparison of defect position coordinates to ensure the accuracy and positioning visibility of surface defect recognition of the target bearing. At the same time, it is convenient to correct the simulation bearing model or surface defect detection model with deviations to improve the reliability and effectiveness of bearing surface defect detection. The specific position comparison visual analysis process is as follows:

[0059] Obtain the defect position coordinates in the output result of the surface defect detection model, obtain the positions of the corresponding points on the simulation bearing model based on the defect position coordinates, and compare and analyze the positions of the corresponding points on the simulation bearing model with the model difference area:

[0060] If the position of the corresponding point on the simulation bearing model belongs to the model difference area, a precise signal is generated;

[0061] If the position of the corresponding point on the simulation bearing model does not belong to the model difference area, a deviation signal is generated. Send the precise signal or deviation signal to the intelligent management unit. When the intelligent management unit receives the precise signal or deviation signal, it immediately displays the preset warning text corresponding to the precise signal or deviation signal, that is, correct the simulation bearing model or surface defect detection model with deviations based on the feedback situation to improve the reliability and effectiveness of bearing surface defect detection;

[0062] In summary, the present invention initially analyzes whether there are defects in the target bearing from the perspective of surface dimensions, and further distinguishes the difference area from the perspective of the simulation model to initially analyze the position of the surface defect area. By processing the surface image data set, a surface defect detection model is constructed to accurately output the position and defect type of the surface defect of the target bearing. Through the method of information feedback, visual analysis of position comparison of defect position coordinates is carried out to ensure the accuracy and positioning visibility of surface defect recognition of the target bearing, and at the same time, it is convenient to correct the simulation bearing model or surface defect detection model with deviations to improve the reliability and effectiveness of bearing surface defect detection.

[0063] The setting of the threshold value is for the convenience of comparison. Regarding the size of the threshold value, it depends on the amount of sample data and the base quantity set by those skilled in the art for each group of sample data; as long as the proportional relationship between the parameter and the quantized value is not affected.

[0064] The size of the coefficient is a specific value obtained by quantizing each parameter for the convenience of subsequent comparison. Regarding the size of the coefficient, it depends on the amount of sample data and the corresponding operation coefficient initially set by those skilled in the art for each group of sample data; as long as the proportional relationship between the parameter and the quantized value is not affected.

[0065] As described above, it is only the preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and should be covered by the protection scope of the present invention.

Claims

1. A bearing surface defect detection system based on model fusion technology, characterized in that: It includes a surface defect detection platform, a surface dimension evaluation unit, a model processing unit, an identification output unit, a comparison evaluation unit, and an intelligent management unit; The surface defect detection platform is used to retrieve the basic information of the target bearing, and send the basic information to the surface size evaluation unit for surface size defect processing and simulation comparison analysis, compare and analyze the obtained surface size defect value, and obtain a feedback signal or a size defect signal. When the feedback signal is generated, a simulation bearing model of the target bearing is constructed based on the basic information and compared and analyzed to obtain a qualified signal or a failed signal; When a non-conforming signal is generated, a corresponding difference region between the simulated bearing model and the standard simulated bearing model is obtained and set as a model difference region; The model processing unit is used to perform model processing operations on the retrieved surface image data set, and to perform discrimination processing on the ratio between the total number of errors and the total number of tests in the obtained output result, so as to obtain an adjustment instruction or a standard-reaching instruction. When a standard-reaching instruction is generated, the algorithm model corresponding to the standard-reaching instruction is set to an end face defect detection model, an outer cylindrical surface defect detection model, and an inner cylindrical surface defect detection model, and the end face defect detection model, the outer cylindrical surface defect detection model, and the inner cylindrical surface defect detection model are fused to obtain a surface defect detection model; The recognition and output unit is used to perform surface defect recognition and positioning processing on the collected surface feature image set of the target bearing to obtain the output result of the surface defect detection model; the comparison and evaluation unit is used to perform position comparison visual analysis on the defect position coordinates, and the position comparison visual analysis process is as follows: the defect position coordinates in the output result of the surface defect detection model are obtained, and the position of the corresponding point on the simulated bearing model is obtained based on the defect position coordinates, and the position of the corresponding point on the simulated bearing model is compared and analyzed with the model difference area: if the position of the corresponding point on the simulated bearing model belongs to the model difference area, a precise signal is generated; if the position of the corresponding point on the simulated bearing model does not belong to the model difference area, a deviation signal is generated.

2. According to the model fusion technology of claim 1, the bearing surface defect detection system is characterized in that: The surface size defect processing and simulation comparison analysis process is as follows: The basic information of the target bearing on the detection workbench is obtained through laser scanning technology, and the basic information includes the inner diameter and the outer diameter; the standard basic information of the target bearing is obtained, and the parameters in the basic information are set to g, where g is a natural number greater than zero, and the difference between each parameter in the basic information and the corresponding parameter in the standard basic information is obtained, and the number of differences between each parameter in the basic information and the corresponding parameter in the standard basic information greater than a preset threshold is set as a surface size defect value, and the surface size defect value is compared and analyzed to obtain a feedback signal or a size defect signal.

3. The bearing surface defect detection system using model fusion technology according to claim 2 is characterized in that: When a feedback signal is generated, a simulated bearing model of the target bearing is constructed based on the basic information, and a standard simulated bearing model of the target bearing is obtained at the same time. The simulated bearing model and the standard simulated bearing model are difference analyzed to obtain the difference value between the simulated bearing model and the standard simulated bearing model. The difference value between the simulated bearing model and the standard simulated bearing model is discriminated and processed to obtain a qualified signal or an unqualified signal.

4. The bearing surface defect detection system using model fusion technology according to claim 1 is characterized in that: The model processing operation process is as follows: A surface image dataset containing normal and defective images is obtained, wherein the surface image dataset includes an end face image dataset, an outer circular surface image dataset, and an inner circular surface image dataset; the end face image dataset, the outer circular surface image dataset, and the inner circular surface image dataset are respectively divided into three parts, namely, a training set, a test set, and a validation set; Each training set is input into the corresponding set algorithm model for training until the samples in the training set are exhausted, and each set algorithm model that has been pre-trained is obtained. Each verification set is further used for verification processing with each corresponding set algorithm model to obtain the output result of each set algorithm model. The output result includes overfitting, underfitting and qualified fit. If the output result is overfitting or underfitting, an improved signal is generated. If the output result is a qualified fit, an accurate signal is generated.

5. The bearing surface defect detection system using model fusion technology according to claim 4 is characterized in that: When an accurate signal is generated, the set algorithm model corresponding to the accurate signal is tested through the test set, and the ratio between the total number of errors and the total number of tests in the output result of the test of the set algorithm model corresponding to the accurate signal is obtained. The ratio between the total number of errors and the total number of tests in the output result is judged and processed to obtain adjustment instructions or compliance instructions.

6. The bearing surface defect detection system using model fusion technology according to claim 1 is characterized in that: The surface defect identification and positioning process is as follows: The line connecting the midpoints of the long sides of the detection workbench is taken as the Y axis, and the starting point of the line connecting the midpoints of the long sides is the negative direction of the Y axis. The line connecting the midpoints of the short sides of the detection workbench is taken as the X axis, and the starting point of the line connecting the midpoints of the short sides is the negative direction of the X axis. The intersection of the Y axis and the X axis is set as the origin, and the ray perpendicular to the detection workbench surface with the origin as the starting point is taken as the Z axis, and then the spatial coordinate system is established; A surface feature image set of a target bearing on a detection workbench is obtained by a high-definition camera, the surface feature image set includes an end face feature image, an outer cylindrical surface feature image, and an inner cylindrical surface feature image, and the surface feature image set is preprocessed and gray-scaled. The preprocessing includes filtering and cropping, and the surface feature image set after preprocessing and gray-scale processing is input into a surface defect detection model; The surface defect detection model identifies and matches the surface feature image set after preprocessing and grayscale processing, and inputs the surface feature image set into the end face defect detection model, the outer cylindrical surface defect detection model and the inner cylindrical surface defect detection model for defect detection, and obtains the output result of the surface defect detection model.

Images