Neural network data-driven engine torque and air-fuel ratio control

In this paper the problem of simultaneous torque tracking and AFR stabilization in spark ignition (SI) engines at the stoichiometric level is addressed. To provide a suitable solution for this problem, an approach based on artificial neural networks with adaptive tuning of their synaptic weights is proposed. For the design of the control law, empirical mathematical models of air-fuel ratio (AFR) and torque are generated and verified. The models are presented as grey boxes with chosen input variables and parameters identified. Identification of the parameters and further verification of the models are provides with the use of data obtained during a driving cycle for a vehicle equipped with a V8 engine. The dual channel neural network controller is tuned in on-line operation according to error back propagation algorithm and the performance index directly depending on the control errors. The controller is free from noise amplification and is robust with respect to uncontrollable engine disturbances. Simulation results show excellent performance of the controller over the entire range of operating conditions.