Developing an Artificial Neural Network for Modeling Heavy Vehicle Rollover 2000-01-3418
A backpropagation through time algorithm was used to model and predict the rollover of a tank truck carrying varying liquid volumes, traveling at various speeds, and performing a number of steering maneuvers of up to 12 seconds duration. The training and testing data sets were built with data produced by simulations using first principle models. Because neural networks have trouble predicting behaviors beyond the boundaries of their training sets, the training set was weighted with 5 per cent of the input examples involving vehicle rollover due to sloshing. The network outputs under test data sets produced very strong correlations with first principle roll simulations in both rollover and non-extreme steering maneuvers.