Bayesian optimization and guided over-instrumented gripper design for thermoforming of composite materials

This work investigates the application of Bayesian Optimization for optimizing process settings in the thermoforming of continuous fiber-reinforced thermoplastic composites. The optimization addresses a high-dimensional design problem involving the positions and forces of tension grippers, aiming to minimize wrinkle formation based on computationally expensive finite element simulations. Advanced Bayesian Optimization techniques, including Log Expected Improvement and Trust Region Bayesian Optimization, demonstrate superior performance compared to standard methods, achieving large wrinkle reduction with fewer simulation evaluations. A sensitivity analysis based on the learned Gaussian process surrogate model is performed, providing quantitative insights into the relative influence of individual gripper parameters. Building on the optimized baseline configuration, the Trust Region based Bayesian Optimization is utilized to identify promising regions for over instrumentation beyond basic parameter tuning. The approach leverages the previously obtained solution in the 16-dimensional optimization domain together with a one-dimensional positional exploration of the added gripper to identify Trust Regions. The results show that strategic over-instrumentation combined with local re-optimization can yield significant improvements in forming quality with low additional effort, demonstrating the effectiveness of Bayesian Optimization for simulation-based optimization of immature manufacturing processes. • Bayesian Optimization is applied to a high-dimensional, simulation-based gripper optimization problem in composite thermoforming. • LogEI and TuRBO outperform EI and Sobol sampling for efficient wrinkle reduction in a 16-dimensional design space. • ARD-based sensitivity analysis of Gaussian Process surrogate models identifies gripper positions as the dominant parameters influencing wrinkle formation. • A two-stage over-instrumentation strategy combining regional EI and local re-optimization enables further forming quality improvements with low additional computational effort.