Research On Bearing Surface Defect Detection And Classification System Based On Machine Vision

Bearings are an extremely wide variety of parts and components used in industry and are important supporting basic parts for modern machinery and equipment.In the bearing production process,the detection of the surface defects still takes the advantage of manual detection,which has the disadvantages of poor stability,large workload,and the like.Machine vision system has the advantages of rapid detection,high degree of automation,non-destructive testing,etc.The detection of defects in the appearance of bearings has natural advantages.Based on this,a system of bearing surface defect detection and identification based on machine vision is designed.The main work contents and innovation points are summarized as follows.(1)Hardware design and algorithm design of a machine vision-based bearing surface defect detection and identification system have been completed.First of all,the whole system was designed.By analyzing the structural characteristics and common defects of the bearing,the goal of system design was established.According to this design requirement,the overall scheme of the system was given;secondly,the hardware of the system was designed.The system hardware is divided into image acquisition subsystem,loading and unloading subsystem,station transfer subsystem,etc.,and each subsystem is designed and illustrated one by one.Again,the software algorithm part of the system is designed,according to the image engineering.In the theoretical framework,the overall flow of the algorithm is designed,and the algorithm design and description are performed for each step in sequence.Finally,the experimental verification of the proposed algorithm is performed through experiments.(2)The automatic image acquisition and loading and unloading device designed by the system can automatically capture the images of the four surfaces of the bearing and realize automatic loading and unloading.This is the main innovation point of this paper in structural design.(3)In order to improve the image segmentation accuracy of the machine vision system,a local multi-neural network image segmentation algorithm based on wavelet transform is proposed,which is abbreviated as Lc-MNN algorithm.The algorithm uses wavelet feature extraction,Lc-MNN region classification,and Lc-MNN classification to achieve accurate segmentation of defect images.The experiment proves that thealgorithm can effectively improve the segmentation accuracy of the defect area,retain more detailed features of the defect area,and lay a good foundation for the feature extraction of the next step.This is one of the theoretical innovations in this paper.(4)Aiming at the shortcoming of feature selection in bearing surface defect recognition based on machine vision,a practical feature selection algorithm based on correlation analysis,scalar selection and vector selection is proposed.Firstly,the similar features are removed by correlation analysis.Then,scalar selection algorithm is used for further screening.Finally,the final classification features are selected by the vector selection algorithm.Comparative experiments show that this algorithm can achieve effective feature selection with a recognition rate as high as 99.5% and avoid large-scale computation.This is another theoretical innovation in this paper.