Real-Time Rollover Warning in Tripped and Un-tripped Rollovers with A Neural Network

Nearly 35% of passenger vehicle accident deaths are from rollover crashes. In vehicle stability control system, the active rollover prevention is presented to prevent rollovers. An imminent rollover should be detected immediately through accurate and reliable detection for active vehicle rollover prevention. A traditional rollover index is able to detect un-tripped rollovers. However, it fails to detect tripped rollovers from external inputs such as tripping by the force of a vehicle striking a curb or a road bump. Thus, a new neural network algorithm to detect both tripped and un-tripped rollovers is needed so any estimation algorithms to determine unknown parameters will no longer be used. The neural network algorithm uses multi-variables from available sensors on the vehicle to calculate and categorize the rollover warning into 3 levels: “Safe”, “Low Risk”, and “High Risk”. Moreover, the algorithm can detect both tripped and un-tripped rollover by testing with a 1/5th scaled vehicle. In order to show dynamic similarity between the 1/5th scale vehicle and a full-sized vehicle, the Buckingham π theorem is used. From experiment results, it is clear that the neural network algorithm can be used to accurately enable the rollover warning for the tripped and un-tripped rollover.