Research On Surface Defect Detection Method Of Automobile Water Tank Brazing Based On Machine Vision

Automobile water tank is one of the important parts of automobile engine cooling system,and ensuring its quality is very important.The core of the tube-strip water tank is formed by brazing after assembling various parts.The core may be due to certain factors during the pre-assembly process,and defects such as unsoldered and unsoldered may occur during the brazing process,which seriously affects the brazing quality of the water tank.At present,the quality inspection of water tank brazing is mainly based on human eyes,but the manual method is inefficient and the accuracy is not high.In order to detect defects more efficiently and accurately,and achieve the purpose of automated brazing quality inspection,this paper proposes a machine vision-based inspection method for surface defects of automobile water tanks.According to the characteristics of the surface defects of the automobile water tank and the detection technical indicators,the paper selects the appropriate software and hardware.Aiming at the more typical defects caused by the error of the matching size of the main plate and the cooling pipe,the classic machine vision method is adopted.First,preprocess the image.A region growing algorithm based on the connected parts of eight neighborhoods is used to find all the coordinate information of connected region.In the process of defect extraction,some interference is found at the image boundary.In order to eliminate the interference,this paper proposes an improvement based on the region growing algorithm.The boundary interference is removed and only the brazed joint area is retained.According to the defect shape and the number of defect pixels,the defects are extracted and identified from the connected areas of the brazing joint area.Due to the large difference in the shape of the surface defects caused by the collapse of the tube edge caused by the press-fitting during the research process,the classic machine vision method has a low recognition and detection rate of such defects,and the versatility is poor.Therefore,this paper proposes a water tank brazing defect detection algorithm based on Faster-RCNN.Firstly,the data set of automobile water tank is made,and the VGG16 pre-training model is migrated to the target detection network.The Faster-RCNN network was constructed,and the model was trained and tested by changing the parameters of the network model.According to the training loss convergence and test performance indicators,it is verified that the method can well realize the identification and location of defects.Experimental results show that the improved eight-neighborhood connected domain analysis method can quickly and accurately identify typical defects,with an accuracy rate of 90%.In order to further improve the accuracy and identify the collapse defect of the pipe edge,this paper uses the Faster-RCNN network model.The model is highly adaptable,and it performs well in anti-interference and accuracy.Moreover,it can identify various shape defects.The accuracy of defect detection is 95%,and the average detection time for each picture is 0.375 s,which has certain practical significance for promoting the development of automated automobile water tank brazing quality inspection.