1 of 1

v

v

Physics-Informed Bayesian Optimisation algorithm identified high-performing CFRP laminate designs using only 19 simulations

Cuba-Kae Vanderpuye, James Dear, Haibao Liu

v

v

Short Beam Shear (SBS) test setup.

v

Methodology

Challenge: Optimising composite layups requires expensive progressive damage simulations.

Objective: Identify laminates with high peak load and low post-failure load drop using minimal simulation data.

Conclusions:

✓ Only 19 simulations required ✓ Accurate surrogate predictions

✓ Efficient Pareto-front exploration ✓ Framework scalable to CDM and impact optimisation

Data-Driven Optimisation of Composite Laminates Using Multi-Objective Bayesian Optimisation (MOBO)

Machine Learning Framework Flowchart

Physics-informed metric extraction

Results

Evolution of GP surrogate predictions and EHVI-guided candidate selection

Selected optimal laminate configuration