An Artificial Neural Network-Based Expert System for Bi-directional Prediction of the Parameters of Single-Layer Deposition by Wire and Arc Additive Manufacturing

Abstract

Wire and arc additive manufacturing (WAAM), especially its new variant cold metal transfer (CMT) process is regarded as one of the most potential and advanced additive manufacturing processes. The process parameters play an extremely influential role in determining the dimensional accuracy of the part manufactured, stability of the process and obtainment of other important process outcomes of interest. Hence, subtle determination of a suitable combination of the process parameters stands extremely crucial, as a result of which, process modelling and parameter optimization of WAAM has been an intriguing subject of research over a past few decades. In this paper, a reliable predictive system has been attempted using computational intelligence to estimate the input conditions so as to achieve the desired nominal responses qualifying the key performances. Artificial neural network models have been developed, envisaging Levenberg–Marquardt training algorithm in order to achieve a bi-directional predictive capability for a set of 3 inputs and 12 responses. Optimal network parameters like initial weight and hidden layer neurons have been determined by validation performance while training the same samples in multiple trials. R-squared values of the training samples and mean absolute percentage errors of the test samples for each response have been found quite satisfactory suggesting fairly adequate predictive models. With this approach, both forward and backward mappings have been successfully achieved.