2015 QPRC

Using Simulation to Incorporate Response Variability into Pareto Front Optimization

Christine M. Anderson-Cook, Statistical Sciences Group, Los Alamos National Laboratory
Jessica L. Chapman, Department of Mathematics, Computer Science and Statistics, St. Lawrence University
Lu Lu, Department of Mathematics and Statistics, University of South Florida

Abstract:

Pareto front optimization has been used to evaluate and balance trade-offs between different estimated responses to seek optimum input locations for achieving best outcomes. Due to the natural variability of the responses, the estimated response surfaces are subject to uncertainty, which will then influence the Pareto front, and thus the final decisions.  An approach using simulation to directly incorporate the uncertainty of the estimated model parameters into Pareto front optimization is presented. This approach provides realistic information about variability in the front as well as how decisions about best design space locations are influenced. The method is illustrated with a manufacturing example with three responses and two input factors. Ideas are shown for scaling the methods when the dimension of the problem increases for either the input space or the number of responses.