Explainable AI for Layer-Wise Emission Prediction in Laser Fusion

The dynamic behavior of melt pools in powder bed-based laser fusion is very challenging to model using physics-based models and conventional black-box data-driven models. Explainable Artificial Intelligence is developed in this work to advance the understanding of convoluted links of non-sequential process physics, online time series sensing data, and process anomaly (e.g., overheating in the melt pool). A Shapley Additive Explanations (SHAP)-enabled Deep Neural Network-Long Short-Term Memory (DNN-LSTM) model has been developed as a mechanism to integrate process parameter knowledge with process history information through online sensing data while providing local and global model interpretation and transparency.