Treffer: A Surrogate-Assisted Multiconcept Optimization Framework for Real-World Engineering Design

Design problems often have multiple conceptual solutions, referred to as concepts, typically represented using different variables. Ideally, designers would optimize across these concepts to identify the best-performing concept(s) and their corresponding design(s). However, existing optimization methods operate with a fixed set of variables, restricting their use in concept design. Multiconcept optimization (MCO) algorithms bridge this gap by allowing searches across multiple concept spaces. Yet, two key elements in MCO require further development for practical use: (1) efficient use of approximations to guide the search and (2) handling analysis failure during performance assessment. To address these challenges, we introduce an MCO framework that solves unconstrained and constrained single-objective optimization problems with a limited computing budget. The framework incorporates four surrogate-assisted optimization algorithms: predictor believer (PB), infeasibility preserved believer (IPB), and two novel approaches—enhanced constrained expected improvement (EcEI) and Bradley–Terry-based probabilistic sorting (BTPS). All these algorithms can handle analysis failure. For maximum flexibility in functional representation, the algorithms dynamically select surrogates from 23 classifiers and 41 regressors during the search. We demonstrate the framework on various analytical and practical examples, including single-concept constrained optimization problems (G-series), a modified G24 problem with analysis failure, a beam design problem involving six concepts, and a 3D shape-matching problem for coronary stent designs involving four concepts. Furthermore, we present the first parameterization scheme to represent single-helix stent designs. We believe that our contribution will enhance the adoption of optimization methods in concept design.