Vehicle rollover detection in tripped and untripped rollovers using recurrent neural networks

Abstract

Comparing to other types of vehicle accidents, fatality rate of tipped rollover accidents shows significant number. Thus, tripped rollover prevention systems are important in order to keep driver safe. In other hands, different rollover indices are defined to handle the risk. The variable unknown parameters of each index, for instance, current load of the vehicle or center of gravity, are considered as a difficulty. In this work, the recurrent neural networks, which are designed to work on sequential data in order to provide data estimation without additional estimation algorithm, are investigated in purpose to estimate the tripped and untripped rollover index. The vehicle simulation software with industrial standard CarSim is applied to validate the result. The Tanh recurrent neural network is stated in the result to be the most accurate tripped rollover index estimator for the uncertain parameters, for example, sprung mass and the height of the center of gravity. The suitable input features for tripped and untripped rollover index and neural network structure are verified. To prevent and provide warning of rollover, an advance future prediction can also be designed for the future tripped and untripped rollover prediction.