| About this Abstract |
| Meeting |
2025 TMS Annual Meeting & Exhibition
|
| Symposium
|
Additive Manufacturing Modeling, Simulation and Machine Learning
|
| Presentation Title |
Multi-fidelity surrogate for integrating melt pool models across diverse input spaces |
| Author(s) |
Nandana Menon, Amrita Basak |
| On-Site Speaker (Planned) |
Amrita Basak |
| Abstract Scope |
Multi-fidelity (MF) modeling is a robust statistical method that intelligently combines data from sources of varying fidelities. This technique is particularly useful for predicting melt pool geometry in laser-directed energy deposition. A significant challenge in employing MF surrogates for melt pool modeling is the disparity in input spaces. This talk presents a novel method to construct an MF surrogate that predicts melt pool geometry by integrating models of different complexities, operating on diverse input spaces. The first thermal model uses five input parameters: laser power, scan velocity, powder flow rate, carrier gas flow rate, and nozzle height. The second model only considers laser power and scan velocity. A mapping between these heterogeneous input spaces transforms the five-dimensional space into a pseudo two-dimensional space. Predictions are then merged using a Gaussian process-based co-kriging method, enhancing predictive accuracy and computational efficiency while reducing the need for high-fidelity model evaluations. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Additive Manufacturing, Modeling and Simulation, Other |