Feasibility study on machine learning methods for prediction of process‑related parameters during WAAM process using SS‑316L filler material

Abstract

The geometry of objects by means of wire arc additive manufacturing technology (WAAM) is a function of the quality of the deposited layers. The process parameters variation and heat flow affect the geometric precision of the parts, when compared to the actual dimensions. Therefore, in situ geometry monitoring which is integrated in such a way to enable a backward control model is essential in the WAAM process. In this article, an attempt is made to study the effect of four input variables, namely voltage (U), welding current (I), travel speed and wire feed rate on the output function in the form of two geometrical characteristics of a single weld bead. These output functions which are determinant of the weld quality are width of weld bead (BW) and height of weld bead (BH). A machine learning approach is utilised to predict the bead dimensions based on the input parameters and to predict the parameters by assigning suitable scores. For predicting the bead dimensions, two models, namely linear regression and random forest, shall be utilised, whereas for the purpose of classification based on weld parameters, k-nearest neighbours model shall be employed. Through this work, a wide dataset of parameters in the form of input variable and output in the form bead dimensions are generated for 316LSi filler material which shall be used as a training data for a machine learning algorithm. Subsequently, the predicted parameters shall be cross-checked with actual parameters.