Estimation of deposition rate in robotic-controlled wire arc additive manufacturing: implementation of state-of-the-art machine learning algorithms

Abstract

Purpose Wire arc additive manufacturing (WAAM) is characterised by a higher deposition rate compared to other metal additive manufacturing processes. This study aims to estimate the deposition rate for single-layer deposited beads during the WAAM process using state-of-the-art machine learning algorithms. Design/methodology/approach In this study, six machine learning algorithms, lasso regression, ridge regression, random forest (RF), XGBoost, Gaussian process regression (GPR) and artificial neural network (ANN), were used for the development of predictive modelling of deposition rate in WAAM with three input parameters. The data set was collected by conducting experimental trials. Findings ANN was highly efficient in attaining a higher coefficient of determination (R2) = 0.97 for testing. Mean square error and mean absolute error were acceptable, and predictions were very close to the actual value and better than lasso and ridge regressions. Three models, GPR, RF and XGBoost, were found ineffective due to lower performance and accuracy, so they were not used for the final prediction. Research limitations/implications Due to the limitation of conducting a large number of experimental trials in the research laboratory, only a small data set was considered in this study. In addition, a single-layer bead deposition was carried out to simplify the data collection task for obtaining the experimental value of the deposition rate. Practical implications Implementation of machine learning techniques for manufacturing processes, particularly for optimisation or estimation of process parameters and output responses, is very important. This case study can help manufacturing engineers and production managers implement machine learning in an industrial environment to improve the quality and overall performance of manufacturing processes. Originality/value Using machine learning for the estimation of the actual deposition rate in WAAM with higher accuracy is possible with minimal assumptions. Hence, it gives more accurate results compared to analytical and mathematical calculations.