Research On Resistance Spot Welding Quality Based Onmachine Learning

The body quality is closely related to the quality of spot welding joints,andnugget diameter and tensile shear strength are important indicators forevaluating the quality of spot welding joints.Since welding is a complex process of multivariate interaction,in order to ensure the quality of spot welding joints,itis traditionally necessary to repeatedly adjust the welding process parameters inorder to obtain qualified nugget diameter,and in order to study the mechanical properties of the weld joints,it is necessary to Tensile-shear test is performed ona welding sample,the method is costly,time-consuming and low in efficiency.Inaddition,for post-welding quality inspection,ultrasonic solder joint quality inspection has been maintained at the level of manual inspection.The quality ofspot welding joints is more affected by human factors,and the accuracy of qualityinspection and the inspection efficiency is low.Therefore,the rapid prediction of the quality of spot welding joints and rapid inspection after welding is still anurgent issue to be solved.In view of the above problems,this paper proposes amethod based on machine learning to construct a spot welding quality prediction model and a post-weld quality inspection model,which provides a new idea forthe rapid prediction and detection of spot welding quality.The main work of this paper is as follows:1)This article first constructed a process database for spot welding quality prediction.Due to limited spot welding data,this article only focuses on the resistance spot welding process of high-strength steel materials.The welding current,welding time,and welding pressure are used as input characteristics,and the nugget diameter and tensile shear strength are output.Based on support vector machine,decision tree,random forest and XGBoost algorithm,a spot welding quality prediction model was established,and the prediction accuracy and time-consuming of each model were compared and analyzed.The results show that the spot weldingquality prediction model based on the XGBoost algorithm has higher prediction accuracy,the prediction result is closer to the actual value,and the time is very short.It shows that the spot welding quality prediction method proposed in this paper can ensure the reliability of the results while greatly improve the efficiency of spotwelding quality research work.2)In order to obtain higher prediction accuracy,the optimal model XGBoost for the above-mentioned spot welding quality prediction accuracy,the gray wolf algorithm(GWO)is used to optimize the model hyperparameters,and the GWO-XGBoost composite model is established,and the prediction results of the model before optimization are compared analysis.The results show that the root mean square error of the GWO-XGBoost spot welding quality prediction model is reduced,which proves the effectiveness of the GWO optimization model parameters to improve the prediction accuracy of the model.3)Use ultrasonic testing equipment to collect ultrasonic echo signals,select the four characteristics of bottom echoes,intermediate echoes,attenuation rate and bottom echo spacing as the feature values for evaluating solder joint quality,based on support vector machine,decision tree,Random forest and XGBoost algorithm build spot welding quality classification model,and compare the classification accuracy,time and classification results of each model.The results show that the spot welding quality classification model based on the XGBoost algorithm has the highest comprehensive accuracy and the test time is very short.The gray wolf optimization algorithm is used to optimize the parameters of the XGBoost spot welding quality classification model.The results show that the comprehensive classification accuracy of the spot welding quality classification model after parameter optimization has been improved,and the gray wolf optimization algorithm selected in this paper can improve the classification accuracy to a certain extent to optimize the parameters of spot welding quality classification model.