Exhaust Gas Recirculation Control in a Spark-Ignition LPG Engine Using Neural Networks

Abstract

This paper presents a neural network approach to control exhaust gas recirculation (EGR) in a liquefied petroleum gas (LPG) engine. In order to meet increasingly stringent automotive exhaust emission regulations, alternative fuels such as LPG engines have been developed in many countries. HC&CO emissions of LPG engines can be easily reduced through air-fuel ratio control, but the control effect of NOx is not good enough. Consequently EGR system is introduced to achieve a significant reduction of NOx emissions. Conventional EGR control uses the mapping method. The calibration time is long and the work is complex when adopting this mapping method. Neural networks are suitable for the identification and control of nonlinear dynamic systems. Neural networks own many advantages such as parallelism, self-organization, memorization, fault-tolerance, etc. If the structure is appropriate and train times are enough, using neural networks can improve the control precision of EGR and reduce the cost and time of calibration. The software and hardware design of EGR control system are also described in this paper. Neural networks for the EGR control have been developed on a 1.46-liter 4-cylinder spark-ignition LPG engine. The experimental results show that the EGR system can achieve satisfied control effect