Reliability-based optimization using evolutionary algorithms

  • Authors:

  • Kalyanmoy Deb;Shubham Gupta;David Daum;Jürgen Branke;Abhishek Kumar Mall;Dhanesh Padmanabhan

  • Affiliations:

  • Department of Mechanical Engineering, Indian Institute of Technology Kanpur, India and Helsinki School of Economics, Helsinki, Finland;Operations Research Center, Massachusetts Institute of Technology, Cambridge, MA;Solar Energy and Building Physics Laboratory, Swiss Federal Institute of Technology, Lausanne, Switzerland;Warwick Business School, The University of Warwick, Coventry, UK;Deutsche Bank, Mumbai, India;Hewlett Packard, Bangalore, India

  • Venue:
  • IEEE Transactions on Evolutionary Computation
  • Year:
  • 2009

Quantified Score

Hi-index 0.00

Visualization

Abstract

Uncertainties in design variables and problem parameters are often inevitable and must be considered in an optimization task if reliable optimal solutions are sought. Besides a number of sampling techniques, there exist several mathematical approximations of a solution's reliability. These techniques are coupled in various ways with optimization in the classical reliability-based optimization field. This paper demonstrates how classical reliability-based concepts can be borrowed and modified and, with integrated single and multiobjective evolutionary algorithms, used to enhance their scope in handling uncertainties involved among decision variables and problem parameters. Three different optimization tasks are discussed in which classical reliability-based optimization procedures usually have difficulties, namely 1) reliability-based optimization problems having multiple local optima, 2) finding and revealing reliable solutions for different reliability indices simultaneously by means of a bi-criterion optimization approach, and 3) multiobjective optimization with uncertainty and specified system or component reliability values. Each of these optimization tasks is illustrated by solving a number of test problems and a well-studied automobile design problem. Results are also compared with a classical reliability-based methodology.