Virtual Experimentation for Robust Design

Abstract

Achieving design robustness is a goal for any industrial product development process. The theme of achieving robustness is to de-sensitize the target product against noise (i.e., variation-causing) factors through a fine calibration of control factors. It is commonly conducted through empirical testing along with statistical experimental design methods. However, empirical experiments are usually cost-, time-inefficient, and posterior to product development. Virtual experimentation using computer simulations is an efficient alternative to improve product performance robustness. Virtual experimental design is defined as a combination of statistical and mathematical procedure applied in the planning, design, running, analysis, and validation of computational experiments to improve the performance and stability of a target system. There are numerous differences in the engineering and statistical considerations for virtual vs. empirical experiments as below: 1. roles in product development process; 2. automated process vs. manual input/output; 3. Size of experiments; 4. Modeling error vs. test error; and 5. Design optimization vs. enumeration/validation. In this presentation, the author will illustrate how to apply virtual experimentation to improve product robustness.