Visual Detection Method And System Design For Metal Weld Defect

As industrial processing technology develops and the demand for intelligent operations continues to rise,the demand for welding quality detection technology is increasing.The welding process is susceptible to various disturbances that can affect the quality of the welded parts and the reliability of the products.At present,weld defect detection mainly relies on human resource,which is inefficient,costly and susceptible to the impact of subjective factors,which make it difficult to ensure the accuracy of detection.According to the practical application requirements of smart factories,this paper designs a weld defect visual detection system by applying machine vision to weld defect detection.The main research of this paper is as follows:(1)According to the requirements of the visual inspection system for weld defects,suitable vision sensors and other hardware devices are selected to build the system hardware platform.A coordinate transformation model is established,the calibration of the industrial camera is performed by the Zhang’s calibration method,and the laser plane is calibrated using the camera calibration results and the calibration plate to obtain the equation of the laser plane in the camera coordinate system.(2)Aiming at the detection of excessive weld residual height defects,a detection method is proposed to compare the residual height dimension by structural light measurement with a threshold value.An improved bilateral filtering algorithm is used to filter the structured light image of the weld seam,and pre-processing operations such as threshold segmentation are performed on it.In order to extract the structured light centerline accurately,a combination of Hessian matrix and mean coordinate method is proposed as a centerline extraction algorithm.And an improved linear fitting method is used to extract the structured light feature points and to compare the calculated residual height dimension with the threshold value to determine whether the weld has a residual height overload defect.(3)Aiming at the detection of multiple types of defects on the weld surface,an improved YOLOv5 algorithm for weld defect detection is proposed.The data set is expanded using the Mosaic+Mixup data augmentation strategy,and a lightweight Ghost Net network is introduced to replace the residual module in the YOLOv5 backbone network to reduce the computational effort and number of parameters in the backbone network.The improved model is trained and tested on the data set after expansion.The experiments show that the improved YOLOv5 model is more effective compared to the traditional deep learning model,and the mean average precision reached 96.88%.(4)Based on the proposed weld defect detection algorithm,the corresponding software system is designed to provide two types of defect detection algorithms for different scenarios to choose to use.The software platform is also used to perform experimental verification of the visual inspection system for weld defects designed in this paper.The experiments show that the measured residual height size and the actual size error are small in the detection of weld residual height overload defects;in the detection of multiple types of defects on the weld surface,the recognition rate of weld defects is high,with an average recognition rate of about 86.79%.