Non Linear Estimator Design for Air to Fuel Ratio Measurement Using Neural Network Based Approach in Spark Ignited Engine

Abstract

<div class="section abstract"><div class="htmlview paragraph">Precise measurement of Air-fuel ratio (AFR) or Lambda value plays a substantial role in controlling exhaust emission from an internal combustion engine. Estimation of AFR is a significant factor to determine the engine performance and to optimize the catalyst conversion efficiency which has direct impact on increase or decrease of emissions. Most of the production two-wheeler engine determines AFR by using non-linear lambda sensor (Narrow band oxygen sensor) but it limits the AFR control due to restrictions in its performance and operating time. A wideband lambda sensor is more accurate and faster but may not be economical to place on low-cost vehicles. A time varying ion current signal can be easily captured on vehicle with minimal additional requirements. AFR has direct correlation with various engine parameters such as Engine speed, Throttle position sensor (TPS), Manifold air pressure (MAP), Fuel injection pulse width (FPW), etc. These signals can be captured with the pre installed vehicle sensors. Neural network-based model can be designed and trained to estimate AFR from different vehicle parameters. Results of neural network model can be improved by considering narrowband sensor as an additional input along with others inputs.</div><div class="htmlview paragraph">The work presented in this paper is implemented for a production vehicle ignition system for two wheeled vehicles. Multiple engine parameters as an input and Universal exhaust gas oxygen sensor (UEGO) as an output are acquired using a data acquisition system. The data is used to train and validate the neural network model to determine the possibility of using it for accurate AFR estimation. A wide range of lambda sensor is considered by performing experiment at different engine operating conditions. The model is validated for steady state as well as dynamic operating condition by running engine on dynamometer and by riding vehicle on day traffic scenario respectively. A detailed study is presented to compare the various models of artificial neural network (ANN) and recurrent neural network (RNN). The paper discusses the results by considering the effect of variation in different neural network model parameters along with engine parameters.</div></div>