Improving accuracy of vehicle-terrain interface algorithms for wheeled vehicles on fine-grained soils through Bayesian calibration

The Vehicle-Terrain Interface (VTI) model for off-road vehicle performance is often used in virtual prototyping to support preliminary vehicle design. Five VTI algorithms for wheeled vehicles on fine-grained soils are calibrated in this study using a Bayesian calibration technique and the Database Records for Off-road Vehicle Environments (DROVE): powered and unpowered sinkage, drawbar pull, motion resistance, and gross traction. These algorithms are improved through a two-stage Bayesian calibration technique utilizing the Metropolis algorithm with three separate calibration strategies. The VTI and proposed algorithms are compared for performance in coefficient of determination and root-mean square error. The results from each algorithm were validated through k-fold cross validation with five folds. The final algorithms from the best performing strategy after accounting for increased complexity for each of the five performance parameters are reported as calibrated algorithms. Validated improvements in coefficient of determination are recorded for all five parameters: 7.8% for powered sinkage, 1.4% for unpowered sinkage, 6.4% for drawbar pull, 0.9% for motion resistance, and 12.5% for gross traction. Improvements are also seen in the normalized root-mean square error performance: 13.4% for powered sinkage, 1.9% for unpowered sinkage, 7.6% for drawbar pull, 17.5% for motion resistance, and 23.2% for gross traction.