STATISTICAL CALIBRATION: A BETTER APPROACH TO INTEGRATING SIMULATION AND TESTING IN GROUND VEHICLE SYSTEMS.

Computer models and simulations have become an indispensable tool for solving complex problems in many parts of vehicle development including powertrain engineering, mobility assessment, survivability analysis, and manufacturing and life cycle assessment. As computational power has increased and model accuracy has improved, engineers have come to depend on simulations to investigate and characterize systems. This raises the importance of model calibration and validation. Calibration is the process of tuning model parameters which are not directly measured in physical tests. These parameters maybe physical properties (material and soil properties, manufactured dimensions, engine operating points) which are difficult to measure or entirely non-physical model parameters. Calibration is necessary to ensure that models and simulation results are as close to physical reality as possible given modeling limitations and assumptions. This paper presents a calibration framework which implements automated statistical calibration using kriging emulators. Through a combination of advanced experimental designs and numerical techniques, this framework greatly reduces the computation required to fit emulators. The utility of this framework is demonstrated with examples including the calibration of turbo machinery simulations. Several different methods within the framework are also demonstrated: agreement and linkage based calibration, and automated sensitivity analyses.