Introducing Explainable Artificial Intelligence to Property Prediction in Metal Additive Manufacturing

Industry demands not only accurate predictability, but also the ability to explain predictions made by machine learning models. In the same endeavor, this work introduces eXplainable Artificial Intelligence (XAI) to mechanical property prediction of additively manufactured materials. Historical data for as-built Ti-6Al-4V alloy manufactured via selective laser melting (SLM) was mined to generate a comprehensive dataset. Robust Gaussian Process Regression (GPR) and Neural Network (NN) were built using a hefty 189 training data arguments. Alongside primary SLM process parameters, novel features such as sample porosity and build direction, which are known to have direct impact on strength, were also utilized. The optimized GPR model exhibited mean absolute errors of 23.9 MPa and 0.58% when inferencing on test tensile strength and elongation, respectively. On the other hand, the optimized NN model performed slightly worse with errors of 28.24 MPa and 0.97% for respective tensile properties. Beyond training and testing of the models, they were tested for explainability against different core levels of human-centric understanding put forth by XAI community. It was ultimately concluded that explainability may come at the cost of accuracy. (c) 2024 The Authors. Published by ELSEVIER Ltd. This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0).