Modelling on-off virtual lambda sensors based on multi-spread probabilistic neural networks

In this work, we have explored a novel model of learning machine which seems to be able to emulate effectively the way of functioning of the traditional on-off lambda sensors (i.e. O₂ sensor). These sensors are a low cost solution used in the SI (spark ignition) engines to monitor the air-fuel ratio and so to maintain a strict control of the air-fuel mixture close the stoichiometric condition. The idea behind this work is to suggest a scheme of air/fuel control system for SI engines in which there is not need of a lambda sensor. The last is replaced by a model, named as virtual lambda sensor (VLS), trained in order to predict the air-fuel ratio values in function of features suitably selected by the in-cylinder pressure sensor signal