Research On Geometrical Parameter Prediction For Weld Defect In Metro Train Based On Gaussian Process

Aluminum alloy car body is one of the important ways to ensure the low-carbon energy saving and green environmental protection of urban rail transit trains.Aluminum alloy car body of metro train is generally to be of an all-welded construction with the integral carrying capacity.However,the existing welding methods and techniques cannot completely avoid generating weld defects,and the traditional non-destructive testing technology has the difficulty in quickly quantifying the geometric parameters of defects.For analyzing the geometric parameters of defects quickly and reducing the complexity of manual evaluation,the research on the prediction of defects’ geometrical parameters,which employs the traditional non-destructive testing technology as the source of defect information and the intelligent algorithm as the method of pattern recognition,was carried out.The research will have important theory and application value to guarantee the quality of metro train body and safe operation.In order to solve the existing problems of weld defect detection technology and improve the analysis efficiency of defect geometric parameters,a prediction model of floating potential signal based on Gaussian process was established in this paper.First,a butt weld in the draw beam was chosen as the research object by analyzing the maximum stress contour plot of locomotive underframe under four types of longitudinal load cases.Taking the objective butt weld as an example,a weld defect signal collection model of locomotive carbody based on the ultrasonic guided wave testing was built.After setting specific model parameters according to the actual situation,groups of simulation experiments aimed to explore the correlation between welding defect geometrical parameter and floating potential signal were carried out,which can provide data source and characteristic references for predicting defect geometrical parameters.Then,gearing to the data characteristics of the floating potential signal,an adapted Gaussian process prediction model for floating potential signal was established through combining radical basis functions and optimizing function hyper-parameters.Based on the above Gaussian process prediction model,the analysis of time series was further employed to predict defect geometrical parameters.Verification results on the angle data set exhibit that the Gaussian process prediction model can be used to predict the floating potential signal and its corresponding geometric parameter value.Finally,an application system for predicting geometrical parameters of defect embedded in metro carbody weld was developed,which is able to provide technical support for quantitative analysis of defect geometric parameters.