Published Paper Details:

WAAM has been proven a promising alternative to fabricating medium and wide-scale metal parts with a high depositing rate and automation level. However, the production quality may deteriorate due to the poor deposited layer surface quality. In this paper, a 3-D profilometer is used for surface roughness measurement. To improve the surface integrity of deposited layers by WAAM, different machine learning models, including ANN, Decision tree, Random forest regression, and linear regression, were developed to predict the surface roughness. Furthermore, the ANN model was optimized by using a different number of hidden layers and adam optimizer. Complete factorial experiments were conducted to obtain the training data, and the K-fold Cross-validation strategy was applied to train and validate machine learning models. The comparison results indicate that Random forest regression has superiority in predicting surface roughness. The RMSE, MAE and MAPE for Random forest regression were 65.0021, 1.93516, 14.15% respectively. This study could also provide inspiration and guidance for surface roughness modeling in multipass arc welding and cladding.