| As the basic components of mechanical equipment,bearings usually determine the quality and stability of mechanical equipment,and are the basis for safe and reliable operation of contemporary mechanical equipment.Among them,bearing rollers,as rolling elements,are key parts that are relatively easy to damage in bearings,and their quality conditions have a certain impact on the performance,precision and service life of bearings.Among the existing bearing roller surface quality inspection methods,it is difficult to adapt to the needs of industrial development because of the low inspection efficiency and high labor cost of the manual visual inspection method,while the machine vision inspection method is widely used in industrial scenarios because of its high efficiency,low cost and high accuracy.In addition,with the development of artificial intelligence technology,the application of deep learning in the field of defect detection has increasingly become a current research hotspot.Based on the above research significance,based on photometric stereo vision and deep learning algorithm,this paper carried out corresponding research on the task of surface defect detection of bearing rollers.The main research work is as follows.Firstly,design a reasonable image acquisition system.Combined with the actual needs of the image acquisition site,a reasonable selection of hardware such as cameras,lenses,and light sources was carried out,and an image acquisition system based on photometric stereo vision was designed and built to ensure uniform illumination and clear defects on the end surface of bearing rollers.The experimental environment provides a hardware basis for subsequent algorithm design experiments.Secondly,the research on the deformation elimination method of the reconstructed surface.Aiming at the deformation phenomenon that occurs in the three-dimensional reconstruction of metal samples,two solutions are discussed: one is to use the principle of series expansion method to fit the reconstructed surface;the other is to propose an improved Frankot-Chellappa integral algorithm to solve the deformation problem of the surface.Through the analysis and research of these two solutions,it can be seen that the improved Frankot-Chellappa integral algorithm has relatively better reconstruction effect and computational efficiency,and is more in line with the application requirements of actual scenarios.Thirdly,research on the effect of image fusion methods.Defects are often visual features with weak semantic information in images,and it is difficult to ensure good detection results.In order to improve the quality of the image,the Dense Fuse fusion network is used to fuse the Gamma correction map and the curvature map to obtain defect features with strong semantic information.At the same time,in order to verify the effectiveness of the fusion data set,the YOLOv5 n algorithm is used to detect defects on the four data sets of traditional optical map,Gamma correction map,curvature map and fusion map.The experimental results show that the fusion data set has better detection accuracy than other data sets,and is more in line with the application scenario of bearing roller end face defect detection.Fourthly,the improved YOLOv5 n defect detection algorithm research.First,the Mosaic-9 data enhancement method is used to improve the model’s ability to detect small target defects;then the EIo U loss function is introduced to improve the positioning accuracy of the bounding box;finally,the lightweight upsampling operator CARAFE is used to replace the nearest neighbor interpolation method to expand The receptive field of the feature map is improved,which is conducive to more information participating in the process of feature fusion and improving the detection performance of the model.The experimental results show that the improved YOLOv5 n defect detection algorithm has better defect detection performance than the original algorithm. |
PDF Full Text Request