Neural network-based expert systems for predictions of temperature distributions in electron beam welding process

摘要

In the present paper, neural network-based expert systems have been developed for online predictions of temperature distributions on electron beam-welded plates. Finite element method is a popular tool to carry out this analysis. However, this analysis could be time consuming, and the obtained results might be dependent on a number of mesh parameters, namely shaping ratio, number of element divisions, and others. Thus, an expert system might be necessary for making online predictions of temperature distributions in welding after considering the said uncertainties. Neural network-based expert systems have been developed using the data collected through finite element analysis, and their performances are compared on some test cases. Once trained, the neural network-based expert systems could make the predictions in a fraction of a second.