About this Abstract |
| Meeting |
MS&T24: Materials Science & Technology
|
| Symposium
|
Materials Processing and Fundamental Understanding Based on Machine Learning and Data Informatics
|
| Presentation Title |
Fast and Accurate Prediction of Temperature Evolution in Additive Friction Stir Deposition Through In-Situ Calibration and Exploration of Unknown Physics |
| Author(s) |
Xiaofeng Wu, Nikhil Gotawala, David Higdon, Yunhui Zhu, Hang Yu |
| On-Site Speaker (Planned) |
Xiaofeng Wu |
| Abstract Scope |
As an emerging solid-state metal additive manufacturing technology, additive friction stir deposition has a thermomechanical processing nature, wherein the resulting quality, microstructure, and properties are crucially dependent on the temperature and strain evolution during deposition. However, precise prediction and control of the coupled thermal-mechanical evolution are challenging due to the lack of both full-field monitoring and accurate physics modeling. To address these challenges, here we propose an explainable AI framework via Bayesian learning, which establishes a surrogate model based on physics simulation data and calibrates this model using in-situ monitoring. We construct both univariate and multivariate Bayesian learning models conditioned on different observation data scales, resulting in an accurate and fast prediction model that is both physics-informed and experimental data-driven. With modest computational resources and in-situ measurements, the methods offer insights into previously unknown parameters in physics modeling with an error rate < 5% in the predicted temperature evolution. |