Preference-Based Surrogate Modeling in Engineering Design

摘要

Traditional techniques for validating and updating engineering models attempt to comprehensively improve model fidelity over the entire design space. Alternatively, this paper presents a preference-based strategic procedure for model building in simulation-based design optimization. It is based on the hypothesis that model validation over the entire design space will be time-consuming and may not be necessary, and a preferred approach is to validate them at potential optimal locations. On this basis, this paper introduces an integrated procedure that incorporates the statistics-based Kriging modeling method with an innovative preference-based approach to recursively build, validate, and update predictive surrogate models in the context of engineering design decisions. A distinguishing feature of this approach lies in its strategic investigation of model fidelity from the perspective of its relevance, usefulness, and completeness, thus maximizing its ability to find the most accurate results during design optimization while minimizing the computational cost involved in such model building. Two case studies, including a 21-bar truss design problem, are used as test-bed applications to illustrate the applicability of the preference-based modeling procedure, and the results are discussed from the perspectives of the efficiency of the overall design process and the accuracy of the resulting optimal design outcomes.