Asymptotic Bayesian Optimization: A Markov sampling-based framework for design optimization. (January 2022)

• Record Type: Journal Article
• Authors: Jerez, D.J.; Jensen, H.A.; Beer, M.; Chen, J.
• Is Part Of: Probabilistic engineering mechanics. Volume 67(2022)
• Publication Date: 2022-01
• Subjects: Discrete-continuous optimization -- Dynamic systems -- Markov sampling method -- Metropolis–Hastings algorithm -- Performance-based design -- Proposal distributions -- Stochastic optimization; Engineering -- Statistical methods -- Periodicals; Mechanics, Applied -- Statistical methods -- Periodicals; Probabilities -- Periodicals; Ingénierie -- Méthodes statistiques -- Périodiques; Mécanique appliquée -- Méthodes statistiques -- Périodiques; Probabilités -- Périodiques; 620.100727
• DOI: 10.1016/j.probengmech.2021.103178
• Languages: English
• ISSNs: 0266-8920

Abstract

Abstract: This paper presents a Markov sampling-based framework, called Asymptotic Bayesian Optimization, for solving a class of constrained design optimization problems. The optimization problem is converted into a unified two-phase sample generation problem which is solved by an effective Markov chain Monte Carlo simulation scheme. First, an exploration phase generates designs distributed over the feasible design space. Based on this information, an exploitation phase obtains a set of designs lying in the vicinity of the optimal solution set. The proposed formulation can handle continuous, discrete, or mixed discrete-continuous design variables. Appropriate adaptive proposal distributions for the continuous and discrete design variables are suggested. The set of optimal solutions provides valuable sensitivity information of the different quantities involved in the problem with respect to the design variables. Representative examples including an analytical problem involving nonlinear benchmark functions, a classical engineering design problem, and a performance-based design optimization problem of a structural system under stochastic excitation are presented to show the effectiveness and potentiality of the proposed optimization scheme. Validation calculations show that the scheme is a flexible, efficient and competitive choice for solving a wide range of classical and complex engineering design problems. Highlights: A stochastic framework for solving complex engineering design problems is presented. The design process is converted to a unified two-phase sample generation problem. The scheme can handle continuous, discrete or mixed-design variables. Designs uniformly distributed over the optimum solution set are obtained. The proposed scheme provides valuable sensitivity information of the optimal design.