Comparative Study of Feed-Forward and Recurrent Neural Networks in Modeling of Electron Beam Welding
In this study, back-propagation neural network (BPNN) and recurrent neural network (RNN) were utilized for the modeling of electron beam welding of AISI 304 stainless steel. The input parameters considered in this study were accelerating voltage, beam current, and scanning speed. These primary parameters were modeled along with the responses like bead geometric parameters. The developed approaches of the modeling had been evaluated in terms of computational time and accuracy in prediction for the test data. The modeling capability of both BPNN and RNN was found to be significantly good, when it was compared with the experimental data. However, Elman recurrent neural network had shown the better accuracy in prediction compared to BPNN, due to the presence of feedback connection in RNN, which leads to capture the dynamics of the nonlinear system efficiently. In addition to this, BPNN was found to be computationally faster than RNN, as expected.
KeywordsElectron beam welding Modeling Back-propagation neural network Recurrent Elman neural network
The authors thank the support received from Board of Research in Nuclear Science (BRNS), India, and from the Indian Institute of Technology (IIT) Kharagpur, for carrying out this study. The authors thank Dr. M. N. Jha for his valuable comment and suggestion.
- 2.Jaypuria, S., Doshi, N., Pratihar, D.K.: Effects of welding parameters on mechanical properties in electron beam welded CuCrZr alloy plates. IOP Conf. Series Mater. Sci. Eng. 338(1), 012013 (2018)Google Scholar
- 3.Schultz, H.: Electron Beam Welding. Abington Publishing, England (2004)Google Scholar
- 5.Pratihar, D.K.: Soft Computing: Fundamentals and Applications. Narosa Publishing, New Delhi (2014)Google Scholar
- 7.Medsker, L.R.: Recurrent Neural Networks: Design and Applications. CRC Press LLC, Boca Raton (2001)Google Scholar
- 14.Manjunath Patel, G.C., Shettigar, A.K., Krishna, P., Parappagoudar, M.B.: Back propagation genetic and recurrent neural network applications in modelling and analysis of squeeze casting process. Appl. Soft Comput. 59, 418–437 (2017)Google Scholar