Sensitivity study of process parameters of wire arc additive manufacturing using probabilistic deep learning and uncertainty quantification

Abstract

This study employs a deep learning (DL) based stochastic approach to comprehensively interpret the effects of current intensity and velocity variations on temperature evolutions and cooling rates in the wire arc additive manufacturing (WAAM) process of a thin wall. Uncertainty raised from process parameters, material properties, and environmental conditions significantly impacts the final product quality. Furthermore, understanding the relationship between the process and temperature evolution within the WAAM process is complex. This study contributes to quantifying the uncertainty to the final product quality, such as temperature evolutions and cooling rates via a fast and accurate DL-based surrogate model. This contribution helps to precise adjustments and optimizations to enhance the overall WAAM process. Initially, a DL-based surrogate model is constructed using data obtained from a high-fidelity validated finite element (FE) model, ensuring an impressive 99% accuracy compared to the FE model while reducing computational costs. Subsequently, probabilistic methods are used to characterize uncertainties in current intensity and velocity, and the Monte-Carlo method is applied for uncertainty propagation. The findings illustrate that small variations in the input parameters can lead to significant fluctuations in temperature evolutions. Additionally, a sensitivity analysis is conducted to precisely quantify the influence of each input parameter. Finally, an uncertainty reduction is performed to enhance the variation of cooling rate. In general, this study is expected to make precise adjustments and optimizations to enhance the overall WAAM process for better quality of printed piece.