Updating of Selected Vehicle-Terrain Interface Equations Through Bayesian Calibration
Published: 2016
Publication Name: Proceedings of the 8th ISTVS Americas Regional Conference
Publication URL: https://www.researchgate.net/publication/308024826_UPDATING_OF_SELECTED_VEHICLE-TERRAIN_INTERFACE_EQUATIONS_THROUGH_BAYESIAN_CALIBRATION
Abstract:
Effective modelling of off-road vehicle mobility at diverse fidelities and on various terrains is an important task for many vehicle design applications. The Vehicle-Terrain Interface (VTI) model is a low-fidelity model often used to support virtual prototyping of off-road vehicle designs. The VTI model is an assembly of several equations which predict vehicle performance through high-resolution empirical modelling of the forces and displacements at the traction-terrain element interface. A recently developed database of wheeled vehicle performance tests over loose, dry sand, DROVE (Database Records for Off-road Vehicle Environments) is used along with a two-stage Bayesian calibration process to update the coefficient values in the VTI equations for drawbar pull and traction. Improvements to both models are evaluated through their performance in root-mean square error (RMSE) and square of the coefficient of determination (R 2). The updated coefficients result in RMSE improvements of 17.3% for drawbar pull and 5.5% for traction. The calibrated equation for drawbar pull also shows improvement in R 2 of 2.8% and the calibrated traction equation performs similarly to the original VTI equation in R 2. The improvements for both models were validated through a test dataset from DROVE and convergence of the Bayesian calibration process demonstrated with two indicators of convergence.