Abstract
Keywords - Controlled Auto Ignition, Closed Loop Control, Model based Predictive Control, Neural Net, Variable Valve Train
The combustion in future engines will work with a high amount of recirculated exhaust gas. In part load conditions this enables a low peak combustion temperature, which shows lowest emissions but suffers from instabilities of the process and has a highly nonlinear behaviour. These properties make a closed loop control a requirement for transient operation but also a challenge. The paper presents the use of neural nets for building a model of the gasoline engine equipped with an electromechanical valve train. This nonlinear model is used to set up a closed control loop simulation. Two different Model-based Predictive Controllers are created, which are capable of controlling multiple tasks coevally. These are chosen to the indicated mean effective pressure IMEP and the location of the maximum pressure apmax. One of the controllers is implemented with the potential for respecting constraints on the maximum pressure rise. The controllers are validated in simulation tests running the nonlinear model in a dynamic closed control loop; the results are compared.