Efficient prediction of thermal history in wire and arc-directed energy deposition combining machine learning and numerical simulation

Abstract Among metallic additive manufacturing technologies, wire and arc-directed energy deposition (WADED) is recently adopted to manufacture large industrial components. In this process, controlling the temperature evolution is very important since it directly influences the quality of the deposited parts. Typically, the temperature history in WADED can be obtained through experiments and/or numerical simulations, which are generally time-consuming and expensive. In this research, we developed a robust surrogate model (SM) for predicting the temperature history in WADED based on the combination of machining learning (ML) and finite element (FE) simulation. The SM model was built to predict the temperature history in the WADED of single weld tracks. For this purpose, FE model was first developed and validated against experiments. This validated FE model is then used to generate the data to train the ML modes based on the feed-forward neural network (FFNN). The trained SM model can fast and accurately predict the temperature history in the cases which were not previously used for training with a very high accuracy of more than 99% and in a very short time with only 38 s (after being trained) as compared with 5 h for a FE model. The trained SM can be used for studies that require a large number of simulations such as uncertainty quantification or process optimization.