A transient virtual-AFR sensor using the in-cylinder ion current signal

Abstract

This paper presents a neural network (NN) based air-to-fuel ratio (AFR) estimation scheme for spark ionization sensors in gasoline internal combustion (IC) engines. The proposed hardware and software estimation system results in a virtual wideband oxygen sensor for an individual cylinder. Principal component analysis (PCA) of the spark ionization signal is used with manifold absolute air pressure (MAP), fuel pulse width (FPW) and engine speed to train a NN offline to predict the AFR under transient engine load and speed settings. Experimental results from dynamometer tests on a port fuel-injected (PFI) four cylinder 1.6 l gasoline IC engine demonstrate that the NN based AFR prediction correlates well with AFR measured from a universal exhaust gas oxygen (UEGO) λ -sensor mounted in the exhaust manifold. The prediction is experimentally demonstrated to be robust to transients of load and engine speed. The experimental results significantly advance those of the previously published studies which have been largely restricted to simulation.