Parameter Screening in Statistical Dynamic Computer Model Calibration Using Global Sensitivities

Abstract

Computer, or simulation, models are ubiquitous in science and engineering. Two research topics in building computer models, generally treated separately, are sensitivity analysis and computer model calibration. In sensitivity analysis, one quantifies the effect of each input factor on outputs, whereas in calibration, one finds the values of input factors that provide the best match to a set of test data. In this article, we show a connection between these two seemingly separate concepts for problems with transient signals. We use global sensitivity analysis for computer models with transient signals to screen out inactive input factors, thus making the calibration algorithm numerically more stable. We show that the computer model does not vary with respect to parameters having zero total sensitivity indices, indicating that such parameters are impossible to calibrate and must be screened out. Because the computer model can be computationally intensive, we construct a fast statistical surrogate of the computer model which is used for both sensitivity analysis and computer model calibration. We illustrate our approach with both a simple example and an automotive application involving a road load data acquisition (RLDA) computer model.