Development of efficient numerical heat transfer model coupled with genetic algorithm based optimisation for prediction of process variables in GTA spot welding

Although numerical heat transfer models based on conduction mode of heat transfer have become a strong basis for the quantitative analysis of fusion welding, they still find limited use in actual design for three primary reasons. First, these traditional models consider a volumetric heat source term, which ironically requires a-priori knowledge of the final weld pool dimensions. Second, the numerical models need confident values of a few parameters, e. g. arc efficiency and arc radius, which are usually uncertain and requires many trial and error simulations to realise their suitable values. Third, these models are rarely attempted for the prediction of possible weld conditions for a requisite or target weld dimensions, which is of paramount interest in design for welding. The present work attempts to circumvent these issues by linking a genetic algorithm (GA) based global optimisation scheme with a finite element based three-dimensional numerical heat transfer model. The numerical model includes a volumetric heat source that adapts itself to the computed weld pool geometry at any instant. The GA module identifies the optimum values of a set of uncertain parameters needed for the reliable modelling calculations and next, identifies the suitable values of the process variables, e. g. weld current, for a target weld dimension. In each case, the GA module guides the numerical model to compute weld dimensions for a given set of inputs, traces the sensitivity of the error in prediction on the inputs being optimised, updates them accordingly and reuses the numerical model to finally obtain their optimised values. The complete integrated model is validated with a number of experimental results in gas tungsten arc spot welding processes