About this Abstract |
| Meeting |
MS&T25: Materials Science & Technology
|
| Symposium
|
2025 Graduate Student Poster Contest
|
| Presentation Title |
SPG-5: Comparison of Derivative-Free and Gradient-Based Minimization for Multi-Objective Compositional Design of Shape Memory Alloys |
| Author(s) |
Sravya Josyula, Yuval Noiman, Eric Payton, Tommaso Giovannelli |
| On-Site Speaker (Planned) |
Yuval Noiman |
| Abstract Scope |
Designing shape memory alloys (SMAs) that meet performance targets while remaining affordable and sustainable is a complex challenge. In this work, we investigate methods for multi-objective optimization of SMA compositions to achieve an arbitrary target martensitic start temperature (Ms) while minimizing cost. We trained two machine learning models, a tree-based ensemble and a neural network, using a dataset of experimentally characterized alloys and physics-informed features. The tree-based model was used with a derivative-free optimizer (COBYLA), while the neural network was paired with a gradient-based optimizer (TRUST-CONSTR). Though the tree-based ensemble exhibits slightly better accuracy, gradient-based optimization on the neural network is more consistent at finding optimal solutions, whereas derivative-free optimization on the trees model often converges to suboptimal results. The present study demonstrates some practical limitations to using machine learning models for composition optimization as well as a viable approach which can be extended to other materials. |