Research On Evaluation And Trend Prediction Method Of Strip Welding Quality Under Non-Balanced Data

In the continuous production line of strip steel,the front and back two strips need to be lapped together for welding,so the quality of welding directly affects the smooth progress of the subsequent process.The traditional welding quality is evaluated by manual excavation of the welding edge to carry out punching cup burst experiment,which is a post-repair method to find out the cause of welding problems,which is inefficient and a waste of resources.Therefore,this article from the perspective of data driven,using the data from the welding process monitoring equipment for welding quality,welding quality and on the basis of the trend prediction research,it can find the welder potential ahead of fault to predict maintenance,greatly reduced because of the economic losses caused by welding quality problems.During the study,this paper considers the welding faults in the actual production data is very less is unbalanced situation,puts forward an improved hybrid sampling method and parameter optimization method,in view of the trend prediction,this paper proposes a hybrid prediction model in order to obtain higher prediction accuracy,and through the experiment proved the superiority of the proposed method and model.Finally,the system platform is developed based on PyQt platform to provide support for its application in engineering.The main research contents of this paper include:(1)Aiming at the problem that the traditional feature selection method has poor effect on the extraction of unbalanced data,the feature extraction of unbalanced data based on RELIEF algorithm is proposed.In order to solve the problem that the traditional data equalization sampling method is easy to introduce abnormal points or lose important information,an improved PNS mixed sampling method is proposed,and experiments are carried out with the monitoring data of welding process,which proves that the results obtained by the proposed method have higher accuracy and faster calculation speed.(2)In view of the classification model parameters is difficult,the problem of low efficiency of traditional parameters optimization method,is put forward in this paper,based on the sparrow improved gray wolves optimization search algorithm,has solved the original algorithm the problems of slow convergence speed and weak global search capability,at the same time through standard test functions and real welding experiment data,the superiority of this method was verified.(3)Aiming at the problem that the traditional time series prediction method does not learn enough about the trend details and the prediction accuracy is low,this paper based on the time series decomposition method combined with the TCN-BILSTM hybrid model to forecast,improve the prediction accuracy,and combined with the predicted value and historical data for trend analysis,to provide support for predictive maintenance.(4)Based on PyQt tool,the welding quality evaluation method and trend prediction method proposed in this paper were systematically integrated,and a welding quality evaluation and trend prediction platform which could be applied to Windows platform was established.