Model reduction in vehicle dynamic systems

Abstract

In the future there will be a great demand for autonomous vehicles. New different companies have already reached the prototyping stage. Dynamic system plays a crucial role in the monetary success of an automobile project. The autonomous vehicles are guided by complex algorithms, which utilize the data from the sensors to monitor the environment and plan the motion of the car. For predicting the future position and orientation of the vehicle relative to the changing environment, a mathematical model describing the dynamic characteristics of the vehicle needs to be solved. The solution is found by solving the model numerically by iteration process. Hence, the process of solving the model is the bottleneck for the whole process. By reducing the time consumed to solve the model, the cycle time for the overall process can be decreased and hence, increasing the cycle frequency and the performance of the vehicle. The work done in the thesis attempts to achieve this by mathematically finding the redundant elements, which do not add to the dynamics of the system but increase the calculation time by increasing the complexity of the system. Such elements could be neglected from the system and hence, reducing the complexity without compromising on the accuracy. Here in this thesis, Eigenvalue separation and Eigenvalue sensitivity methods have been applied to the different bond graph models of bicycle vehicle with suspension system, antilock braking system and steering system. The same methods have also been applied for four wheel vehicle model. The bond graph model of the whole systems has been separated into fast and slow systems. The reduced system behaviour is similar to the actual bond graph model. Though this method is mainly applicable for linear systems, the non-linear systems can also be reduced after certain modifications.