ENBIS: European Network for Business and Industrial Statistics

A Sequential Approach to Integrate Physical and Computer Experiments

22 September 2009, 16:55 – 17:15

Abstract

Submitted by

Daniele Romano

Authors

Alessandra Giovagnoli§ and Daniele Romano#

Affiliation

§ Dept. of Statistics, University of Bologna, Italy #Dept. of Mechanical Engineering, University of Cagliari, Italy

Abstract

In competitive industrial sectors R&D activities are often based on the combination of experimentation in the lab and computer simulation. In spite of this, there still is no formalized approach for integrating physical and simulation experiments in the applied statistical literature. Computer experiments is an autonomous discipline since the end of the eighties (Sacks et al., 1989) but provides a limited view of what a “computer experiment” can be in an industrial setting. Simulation is rigidly assumed to be expensive and deterministic and the “integration” problem is totally ignored.
We present a sequential approach, still at a seminal stage, where both physical and computer experiments are used in a synergistic way with the goals of improving a real system of interest and validating/improving the computer model. The approach is somehow inspired by the framework for sequential experimentation proposed by Box and Wilson (1951). Different decision levels are handled. High level decisions are whether to stop or continue, whether to to run the next experiment on the physical system or on its simulator and which is the purpose of the experiment (exploration, improvement, confirmation, model validation). Lower level decisions refer to the choice of the experimental region, the run size and the design.

View paper

Return to programme