Development Of Radiator Brazing Defect Detection System Based On Machine Visio

The car radiator prevents the engine from overheating by absorbing heat from the cylinder block and is one of the important components of the cooling system.The radiator core is brazed by each part through the brazing furnace at high temperature,and in the process of brazing processing near the radiator port is prone to weld defects and blocking defects,these two types of defects seriously affect the brazing quality of the radiator.At present,the detection of brazing defects is mainly done by means of human eye detection,but the accuracy of this manual detection method is not high and inefficient,which can no longer meet the needs of modern production.To this end,a machine vision-based brazing defect detection system has been developed to enable automated detection of radiator brazing defects.The main content of this article is as follows:(1)In the radiator brazing defect image,there are numerous extraneous interference regions outside the target.For this reason,an adaptive port ROI extraction algorithm is proposed in this paper,which achieves adaptive extraction of defective target regions through the steps of threshold segmentation,edge detection,expansion processing and minimum outer rectangle.The test results showed that the classical machine vision inspection algorithm was able to accurately identify and extract two types of brazing defects with an average accuracy of over 91%.(2)To improve the detection accuracy of the system and adaptability to different feature defects,an improved YOLOv5 s target detection algorithm is proposed in this paper.To address the shortcomings of the convolution kernel in terms of spatial location adaptation and inter-channel redundancy,this paper tries to optimize the Backbone backbone network by replacing the convolution layer in the Backbone network with the Involution operator,combining the advantages of the Involution operator.Secondly,a novel Head network model is constructed by introducing Ghostconv module and Bottlenneck CSP module to improve the robustness of the network in practical detection and increase the feature fusion capability of the network model.The improved YOLOv5 s model has better inspection accuracy and adaptability compared with the classical machine vision algorithm.The experimental results show that the improved YOLOv5 s model achieves 94.3% target recognition accuracy for brazing defects,and the detection speed reaches 32 frames/second,meeting the industrial inspection requirements.Completed the development of online inspection system for radiator brazing defects,including the construction of hardware system and the development of software operation interface,which realized the online inspection of radiator brazing quality.