Suprayitno, Jyh-Cheng Yu
The deviations of design variables and parameters due to noise factors cause a response distribution in practices. Robust optimization aims to search for a design with optimality robustness and constraint reliability despite these variations. In light of the high sampling cost in engineering optimization, this study proposes a novel procedure that uses an evolving surrogate model to approximate the actual system, and uses a soft outer array to estimate the performance robustness. Since product life cycles often affect design variables with characteristic patterns, the Design Variation Hyper Sphere (DVHS) is applied to characterize the distribution of design variables, and the variation radius of DVHS is used to modify the constraint activity margin to ensure design reliability. A particle swarm optimizer is applied to search the subspace of the design variables in the surrogate for a robust optimum with the fitness function estimated from the soft outer array. During the evolution, a penalty function is applied to the fitness of the offspring in the infeasible region and within the margin of variation radius from the nominal constraints. The procedure ensures design feasibility without degrading design optimality using a relatively small sample. A design of a helical spring illustrates the effectiveness of the proposed procedure. © 2019 IEEE.
Department of Mechanical Engineering, State University of Malang, Malang, Indonesia; Department of Mechanical and Automation Engineering, National Kaohsiung University of Science and Technology, Kaohsiung, Taiwan
