Methodology for development of welding procedures and empirical weld process models based on principal component analysis techniques

Abstract

With the increased automation of welding processes, there is a necessity for experimental design techniques that will permit rapid and efficient definition of the range of welding process parameters that may be used to produce acceptable welds for each new application. Such techniques are also required during development of empirical weld process models. In the present study, an experimental design technique based on principal component analysis (PCA) has been developed and shown to be more efficient than the traditional fractional factorial approach to experiment design, weld procedure development, and development of weld process models. Gas tungsten arc (GTA) welding of corner joint sheet steel was selected to exemplify the application of the technique. The boundary of a three­dimensional domain of weld process parameters (welding current, speed, and electrode gap) which could be used to produce acceptable welds when using either flush or edge touch corner joint configurations was satisfactorily approximated using only 25 welds. The PCA based technique was able to resolve differences between the procedural domains of these two joint geometries. The procedural field for the edge touch configuration was found to have a larger range of acceptable welding conditions than that for the flush configuration. Empirical weld process models for these joint configurations were developed using standard multivariable regression analysis techniques and only 25 further welds. Finally, an empirical model for weld width prediction in corner joint GTA welding was developed which includes the effect of current, voltage, torch travel speed, and sheet metal fitup. This model had an R² value of 86%, with all statistical tests yielding acceptable results.