| About this Abstract |
| Meeting |
MS&T24: Materials Science & Technology
|
| Symposium
|
Additive Manufacturing Modeling, Simulation, and Machine Learning: Microstructure, Mechanics, and Process
|
| Presentation Title |
Generative Property Optimization of Stochastic Microstructures |
| Author(s) |
Patxi Fernandez-Zelaia, Jiahao Cheng, Jason Mayeur, Amir Koushyar Ziabari, Guannan Zhang, Neil Zhang |
| On-Site Speaker (Planned) |
Jason Mayeur |
| Abstract Scope |
Materials inverse design problems are found in a number of domains from fuel cells to fusion materials. Purely experimental methods can be costly and physics-based models only solve forward problems. Machine learning based generative models have been demonstrated to be well suited for exploring inverse solutions. Denoising diffusion probabilistic models have exploded in popularity and are now ubiquitous across various applications (materials, text, images, audio, etc.). Currently there are no works demonstrating guided microstructural generation for structural engineering materials. This work establishes a model which generates two-phase stochastic microstructures with control over both feature statistics and homogenized properties. The latter is achieved by optimizing structures using extracted gradient information from an auxiliary structure-property surrogate model. The model is flexible enabling other tasks such as feature-constrained property optimization. These preliminary results demonstrate that DDPMs are powerful tools useful for solving inverse materials design problems. |