Real-time Monitoring Of GMAW Welding Surface Quality Assisted By Visual Sensing Of Galvanized Steel

With the development and popularization of intelligent technology in modern manufacturing industry,the technical requirements in the field of welding are becoming higher and higher.The development of welding automation,flexibility and intelligence has become an inevitable trend.Welding quality determines the reliability and safety of welding products.Therefore,Real-time detection of weld quality is not only an important means to ensure welding quality,but also the only way to realize high-precision,high-efficiency and intelligent welding.In this paper,the quality problem of gas metal arc welding of galvanized steel is studied.It is proposed to build an automatic welding platform by using visual sensor combined with line structured light to realize weld image acquisition and processing,and then complete weld defect detection based on deep learning target detection algorithm.It provides theoretical guidance and technical reference for real-time monitoring of weld defects.Firstly,according to the welding working environment,a GMAW welding quality monitoring system platform based on visual sensing assistance is designed,including automatic welding system and machine vision system.Then the camera and structured light parameters are calibrated according to the detection requirements,and the accuracy meets the actual requirements.By analyzing the characteristics of the original weld image,Gaussian filtering combined with Gabor filtering algorithm is proposed,which can effectively remove the interference factors such as arc and splash.Then Otsu threshold segmentation and improved geometric center method are used to extract the weld centerline,and the result is better than other centerline extraction algorithms.Then Douglas puke algorithm is selected to identify the weld feature points,and the weld geometric size and three-dimensional reconstruction results are obtained.Secondly,by analyzing the feature point information of different weld types,eight weld defect feature vectors are combined with each other and input into different machine learning algorithm models.The experiment shows that the performance of BP neural network model based on variable momentum learning rate is better than that of logistic regression,decision tree and support vector machine algorithm.Finally,the shortcomings of traditional machine learning algorithms are analyzed.Based on the deep learning theory,a target detection algorithm yolo-v5 based on regression method is established through data enhancement.Through the backbone feature extraction network,spatial pyramid pooling,feature pyramid and path aggregation network,the average accuracy of yolo-v5 model is 97.92%,and the average running time of detection and recognition of single weld defect original image is less than 200 ms,Ensure the real-time performance of real-time monitoring of weld quality.In order to realize the automatic monitoring of weld defects,the software of GMAW welding quality monitoring system is developed.Taking Qt software as the development platform,C + + language is used for program development.By calling Open CV image function dynamic link library,the real-time monitoring system of GMAW weld quality of galvanized steel is realized.Finally,the reliability and stability of GMAW welding quality monitoring system in the actual scene are verified,and the work efficiency of users is improved.