Research On Process Stability Monitoring Of Resistance Welding Based On Industrial Big Data

Resistance welding is usually applied to mass production manufacturing.With the digitalization of companies,a large amount of the process status information data is collected at the production site.The collected data is often referred to as resistance welding industrial big data.In addition,resistance welding has the problem of quality stability compared to other welding methods due to the short welding time and the rapid heating and cooling.Therefore,how to carry out process stability monitoring is becoming an urgent problem to be solved based on the development trend of industrial big data.Compared with the data collected in the laboratory,industrial big data presents the characteristics of huge data volume,the low information content of single data,and high data fluctuation.The above causes correlation between joint quality parameters and welding process state information is poor,and the traditional modeling idea of building a quantitative prediction model for each welding joint is difficult to extend.Therefore,the main research content of this paper is proposing a new process stability monitoring method for two typical resistance welding methods,resistance spot welding and resistance butt welding,combined with the characteristics of industrial big data.The typical anomaly that may occur during the welding process for resistance spot welding was summarized into three aspects,material stack-up changes,welding expulsion and electrode wear.This study proposed indirectly monitoring the quality stability by monitoring the anomaly during the welding process.Monitoring material stack-up and welding expulsion were transformed into a classification recognition problem in machine learning.By optimizing the data pre-processing method,the material stack-up recognition accuracy could reach 99.3%,and the expulsion recognition accuracy could reach up to 98.25%.The wear of electrode heads is a gradual process,and it is impossible to monitor changes in electrode status using a classification method.So this study firstly demonstrated by numerical simulation that the electrode wear process is correlated with the shift in dynamic resistance curve and then proposed the concept of shape change factor and trend change factor to quantify the change law of dynamic resistance curve.The research results shown that the trend change factor was highly consistent with the actual electrode wear process and could realize the online monitoring of electrode status.Unlike resistance spot welding,resistance butt welding is a solid-phase joining method with better stability of welding quality.The typical anomalous described above do not exist,and it is impossible to monitor the welding process’ s stability by monitoring whether anomalous occur.Therefore,for resistance butt welding,this study proposed to make a direct judgment on whether the welding quality and the welding process are abnormal based on the state curve during the welding process.Specifically,it could be divided into micro and macro aspects.Micro refers to whether each weld point is abnormal,and macro refers to whether the welding process is stable.Microscopically,since industrial big data lack quality labels,this study used anomaly detection algorithms to predict whether a weld joint is abnormal,including isolation forest,autoencoder,and local outlier factors.It also proposed using separation and signal-to-noise ratio to solve the lack of evaluation criteria for anomaly detection algorithms.In addition,this study also proposes a model fusion method based on the relative distance to improve the accuracy of model prediction results.After model fusion,the predicted abnormal link ratio was 4.17%,closer to the actual abnormal link ratio of 4.52%,and the prediction accuracy rate was better than using a single prediction model.Based on the anomaly detection results,it was found that the characteristics of the second upsetting speed and the dynamic resistance rising speed had the strongest correlation with the welding quality.From a macroscopic point of view,this study proposed using statistical process control to monitor the stability of the welding process.Because the characteristic values fluctuated largely during the production process,this study further optimized the traditional control chart.It proposed a sliding window control chart method to monitor the stability of the resistance butt welding process.In addition,based on the sliding window control chart,five process anomaly judging criteria were proposed from the perspective of statistical distribution.In the last part of this study,an anomalous weld joint monitoring system was developed for resistance butt welding.The system used the STM32H7 and STM32MP157 as the core of the hardware system and built a user interface program based on the Py Qt framework to realize user interaction,process curve and anomaly detection result display,and other functions.