A framework based on nonlinear FE simulations and artificial neural networks for estimating the thermal profile in arc welding

In this paper, a novel strategy based on nonlinear thermal analysis has been developed using finite element simulations and artificial neural networks in order to predict the time-temperature distributions in arc welding process. The highly nonlinear and transient thermal finite element methodology pertaining to simulations of arc welding process is investigated through various combinations of numerical features. Detailed algorithms are provided for an insightful understanding of the nonlinear computations associated with the physics of arc welding. Subsequently, data-sets have been generated from the simulations with inputs of various geometrical and weldprocess parameters. The output data describes the non-uniform and nonlinear variation of temperatures with heating and cooling of arc welds. Artificial neural networks have been developed through training and testing of the supervised data-sets. The nonlinear thermal profile is then predicted by the simulation-based neural networks for two sample weld specimens with new process parameters. The results show an excellent agreement with experimentally measured temperatures during welding. The present work provides a robust and effective methodology to obtain fast estimations of nonlinear thermal behaviour during arc welding of the components.

Automated summary

A novel strategy based on nonlinear thermal analysis using finite element simulations and artificial neural networks has been developed to predict the time-temperature distributions in arc welding. The highly nonlinear and transient thermal finite element methodology of arc welding process is investigated through different numerical features, and detailed algorithms are provided for an insightful understanding. Artificial neural networks have been developed through training and testing of supervised data sets, enabling accurate predictions of the nonlinear thermal behavior of welded components.