Meta-Heuristic Algorithms-Tuned Elman vs. Jordan Recurrent Neural Networks for Modeling of Electron Beam Welding Process

摘要

A boost in the preference of high energy beam, such as electron beam, laser beam etc. has led to the requirement of its automation through accurate input-output modelling. Modeling of electron beam welding is conducted in the present study through Elman and Jordan recurrent neural networks (RNNs), both having a single feed-back loop, to meet the said requirement. The RNNs are trained using some nature-inspired optimization tools, namely cuckoo search, firefly, flower pollination, and crow search utilizing input-output welding data, obtained from a computational fluid dynamics-based heat transfer and fluid flow welding model. RNN predictions are validated through real experiments. Thus, the effect of change in the position of the feed-back loop on the accuracy of prediction of RNNs is investigated. In addition, a few popular statistical tests have been used to evaluate the performances of the RNNs tuned by various optimization algorithms, where flower pollination-tuned Jordan RNN is observed to yield the best results.