The Online Detection Algorithm For Surface Defects Of Automobile Hubs Based On Deep Learning

Wheel surface defect detection is an important part of the wheel enterprise's production logistics process.Through on-the-spot investigation,this paper finds that in the traditional solution,the detection of the surface defects of the hub is to locate and mark the defects by visual observation by workers at the fixed station of the production line.Due to the complicated structure of the automobile wheel hub,the manual detection method has low efficiency and large workload,and this detection method is more and more difficult to meet the requirements of faster,more accurate and stable production of the wheel hub.In view of the above situation,this paper combines deep learning technology and field investigation of wheel defect detection tasks to propose an online detection algorithm for wheel hub defects based on deep learning.The main work of this paper is as follows:1.Process design of defect detection system for hub production line.The system includes a complete wheel online defect detection process,including image acquisition,image pre-processing,wheel image defect detection and more.2.Establish a database of wheel surface defects.The specific process is to collect the surface defect image of the hub through the industrial camera on the production line,and then carry out data cleaning,and mark the defect image by professional labeling software,and finally obtain the defect image database with accurate annotation.3.Recognition of blurred images.In order to improve the efficiency of the defect detection algorithm system,an algorithm is designed to recognize the blurred image before the blurring is eliminated.If the image is blurred,the blurred image is sent to the deep learning network to eliminate the blurring.If it is recognized as a clear image,the defect detection of hub image is carried out directly.4.Elimination of blurred image on hub surface.Before the detection of the hub defect image,it is necessary to preprocess the defect image to meet the algorithm requirements.Since the image acquisition stage in the defect detection system proposed in this paper is carried out while the hub is being transported on the production line,there is a motion blur phenomenon in the collected hub image.In response to this problem,this paper proposes a deep generative adversarial algorithm to eliminate motion blur without the need for accurate motion information.5.Wheel image defect detection.This paper implements the deep learning target detection algorithm and improves it based on the wheel defect task.Based on the Faster-RCNN target detection algorithm,three improvements are made,including adding SE module,replacing ROI-Pooling with ROI-Align and FPN multi-scale feature fusion network.Finally,a model for hub surface defect location and classification is obtained.In this paper,the final realization of the hub image defect detection algorithm can detect two kinds of defects:defect points and scratches.At the same time,the location and type of defects are given.On the 300 hubs surface defect image data set established in this paper,for all defect areas,94%recall rate and 88%prediction accuracy can be achieved.