Trust-region Filter Algorithms utilising Hessian Information for Grey-Box Optimisation

Abstract: Optimising industrial processes is essential for enhancing profitability, efficiency, and sustainability. Conventional optimisation relies on equation-oriented (EO) glass-box models with fully specified equations and accessible derivatives. However, many real-world systems contain black-box components (such as high-fidelity computational fluid dynamics simulators or proprietary unit models) that lack explicit formulations or analytic gradients. This limits the applicability of derivative-based deterministic optimisation on grey-box (glass-box/black-box) problems. The classical trust-region filter (TRF) algorithm offers a promising framework for such grey-box optimisation, but often requires adjusting numerous tuning parameters and many black-box evaluations to converge. In this work, we extend the classical TRF method by introducing four new variants that incorporate second-order information through local Hessian projections, improving robustness and reducing sensitivity to tuning parameters. We further integrate a flexible surrogate modelling framework, enabling the optimiser to switch between low-fidelity (linear, quadratic, simplified quadratic) and high-fidelity (Taylor series, Gaussian process) surrogates for local approximation of black-box components. Performance is benchmarked on a suite of 25 grey-box test problems and several engineering case studies, including liquid-liquid extraction, pressure vessel design, alkylation process, and spring design. Across all cases, the new variants achieved up to an order-of-magnitude reduction in iterations and black-box evaluations relative to the classical TRF algorithm. These findings demonstrate the new variants as efficient alternatives for grey-box process systems optimisation.