A hyperspherical particle swarm optimizer for robust engineering design

Abstract

This paper presents a novel multi-objective particle swarm optimizer, called hyperspherical particle swarm optimization (HSPSO), which efficiently deals with robust engineering design problems. In contrast to traditional optimization methods which rely on single-point design configurations, the HSPSO method evolves multi-dimensional design surfaces while simultaneously optimizing several potentially conflicting objectives and minimizing product/process variations. The hyperspherical representation is accommodated by incorporating manufacturing tolerances for design variables, and sensitivity analysis is performed to maintain feasibility within the design region. Hyperspherical particles are automatically evaluated, and non-inferior solutions are identified by the Pareto-dominance strategy. To enhance the local search ability of the particle swarm optimization algorithm, a gradient descent algorithm is applied, and fitness evaluation is performed by using a crowding factor, which defines the density of the population along the Pareto front. The performance of the proposed HSPSO algorithm is highlighted by reporting on three robust engineering design problems, which involve a mixture of single objective and multiple conflicting objectives along with integer, discrete and continuous design parameters. Monte Carlo simulations are used to assess the reliability of the obtained results.