Robust operation ensures desired chemical production outcomes despite parameter uncertainties. In process design, Design Centering (DC) is a practical method for identifying a nominal operating condition while ensuring production robustness. Traditional DC methods often rely on multilevel optimization and explicit gradients, limiting their applicability when models are unavailable or experimental costs are high. To address this, we propose a Bayesian optimization framework that reformulates DC as a derivative-free, worst-case constrained optimization problem. The method treats the whole process as a black box, uses Gaussian process regression to model constraints, and guides adaptive sampling with a feasibility-aware acquisition function. The case studies show that the proposed framework consistently locates high-robustness solutions using only 20–40 samples per case, with computational times ranging from tens to just over one hundred seconds. This represents a significant reduction in evaluation cost compared to geometric or sampling-based DC methods, while remaining compatible with black-box and commercial simulation environments.
Zhao et al. (Thu,) studied this question.