Prediction of clad characteristics using ANN and combined PSO-ANN algorithms in laser metal deposition process

Abstract

Process of laser metal deposition (LMD) involves a plethora of input parameters and the output must be a sound clad of desired geometry and mechanical properties. This desired shape of the clad can be deposited using a specific selection of process parameters. Motivated by the increasing popularity of evolutionary algorithms, its application in surface coating using laser metal deposition is demanded. Neural networks are such ally algorithms which is a form of supervised learning and is useful to predict the outcome from a developed network. Our focus in the present study is to model some of the outcomes of the deposition namely clad layer height, width and powder capture efficiency using artificial neural network (ANN) and combined particle swarm optimization (PSO)-ANN. These two methods are comparatively studied in predicting the responses of response surface methodology based experiments and the results are explored. The combined PSO-ANN based supervised learning paradigm better predicts the clad characteristics with predcition mean absolute percentage errors (MAPE) within 4%. Later, an another approach to obtain suitable process parameters for a targeted clad dimension deposition is performed using PSO-ANN model. The MAPE in PSO-ANN predictions of input parameters for depositing a targeted geometrical feature are within 10%. The reliability of presented models are also compared with previously developed empirical relations and discussed.