Robustness and reliability estimation in multi-objective genetic optimization

A common trade-off in engineering design is between the reliability of a system and its expected performance in uncertain environments. Previous techniques using the genetic algorithm for engineering design optimization solved optimization problems with either robustness or reliability formulations. This thesis proposes an explicit-sampling multi-objective genetic algorithm to find robust and reliable solutions to multi-objective optimization problems. Because of the explicit sampling to quantify uncertainty, the approach presented here is limited to problems where the reliability constraints do not require high values of reliability. A novel presentation of the Pareto-optimal solutions provides the designer with limited variance information about the solutions' performance without requiring additional objectives or computation beyond that used for the basic explicit-sampling multi-objective genetic algorithm. The results of the approach applied to a set of example problems (including continuous analytic problems and mixed discrete non-linear engineering problems) demonstrate the feasibility of explicit sampling methods for robust and reliable multi-objective genetic optimization.