Experimental validation of recurrent Neuro-Fuzzy Networks for AFR estimation and control in SI engines

An accurate model of Air to Fuel Ratio (AFR) dynamics is critical for high-quality AFR control in SI engines. These modeling and control problems are very sensitive because the dynamics of intake manifold air-fuel flow is severely nonlinear and multivariable. This study focuses on Recurrent Neuro-Fuzzy Network (RNFN) estimation and control of AFR nonlinear dynamics in SI engines. First, a nonlinear autoregressive with exogenous inputs (NARX) model is chosen for modeling the AFR nonlinear dynamics in the fuel injection system. Then, the strategy based on RNFN, is employed to fine-tune the model parameters. A controller is also designed based on inverse model-based method. The objective of control scheme is to keep the AFR constraint conditions by providing the proper fuel injection commands. This strategy is performed on an informative data-set obtained by a real-time in-vehicle experimental test. The effectiveness of the proposed approach is evaluated and validated by the resulting improvement in comparison with ECU performance.