Research On Spot Welding Data Preprocessing Method And Realization Of Spot Welding Quality Monitoring Software

Spot welding is widely used in the manufacture of automobiles,electrical appliances,batteries,rail transit and aerospace products due to its advantages of low cost,simple operation,low noise,high production efficiency and easy automation.The quality of spot welding directly affects the reliability and safety of products,so ensuring the quality of spot welding has important social and economic value.With the rapid development of sensor technology,computer technology and artificial intelligence technology,the datadriven artificial intelligence method has brought a new solution to the comprehensive and non-destructive spot welding quality monitoring,the monitoring of spot welding quality through the establishment of a model of spot welding data is a key research direction in the field of spot welding.Due to the complex production environment and unstable working conditions of collectors and sensors,failures often occur in the process of data collection and transmission.At the same time,because the personnel,date and equipment for each spot welding operation may be different,there are differences between each batch of spot welding data that are not related to spot welding quality,so spot welding data in actual factory environments often have missing values and batch effects.Missing values and batch effects reduce the reliability and versatility of the spot welding quality evaluation model,and bring difficulties to the establishment of an intelligent spot welding quality monitoring system.Based on the spot welding data collected in the industrial production process,this thesis preprocesses the problem of missing values and batch effects in the spot welding data,improves the existing preprocessing methods,and establishes a spot welding quality monitoring software system with preprocessing function.The specific work of this thesis is as follows:(1)In order to fill in the missing values of spot welding time series data more accurately,a time series data filling method based on BiGRUI-WGANGP is proposed to solve the problems of incomplete information utilization and unstable training of the time series data filling method of GRUI-WGAN.This method makes full use of the time characteristics and missing laws of missing time series data,and uses WGANGP with dynamically adjust the penalty factor,which is more stable and efficient,for training,which improves the accuracy of missing value filling.(2)In order to better correct the batch effect of spot welding data,a batch effect correction based on EM-ComBat is proposed for the problem that the batch effect correction method of ComBat has poor effect on batch effect correction of unknown label data.This method utilizes the feature that EM algorithm can estimate model parameters with latent variables,improves the parameter estimation process of ComBat method,improves the accuracy of parameter estimation,and thus improves the effect of batch effect correction.(3)Aiming at the requirements of spot welding quality visualization,recordability and traceability in spot welding production,a spot welding quality monitoring software system with preprocessing function is designed and implemented to monitor spot welding data and spot welding quality.The test shows that the preprocessing of spot welding data can improve the accuracy of spot welding quality evaluation and detection,which is beneficial to spot welding quality monitoring.