Bayesian Process Optimization

Overview

Rather than running thousands of experiments to find optimal process parameters, LAMMA develops inverse design frameworks that start from a target material state — a desired crystallinity distribution, void content, or bonding strength — and compute the process parameters needed to achieve it.

Approach

We use Proper Generalized Decomposition (PGD) and Gaussian Process surrogates to build computationally efficient approximations of the high-fidelity AFP-ISC forward model, enabling Bayesian optimization over the full process parameter space.

Goals

• Real-time process parameter adjustment during AFP layup
• Uncertainty quantification in predicted crystallinity fields
• Extension to multi-objective optimization (crystallinity + void content + cycle time)