The need for bias modelling in MVEM based estimators

Abstract

Mean Value Engine Models (MVEM) have been used extensively in automotive controls especially over the last 20 years. An MVEM was derived from a detailed Within-Cycle, Crank-Angle based Model (WCCM) that modelled the fluctuating cylinder combustion driven dynamics of a Spark Ignition engine. The model was designed for eventual use in a Fault Diagnoser built for an automobile engine system. While using this model in Extended Kalman Filter based estimators for fault residue generation, it was noted that the model suffered from biases that impaired the quality of estimation results. The biases were found to originate from the inherent simplifications associated with MVEMs. This led to an understanding of the limits of accuracy of a traditional MVEM model, the need for accurate bias modelling and the development of more robust estimators. Estimation results were found to improve after bias correction using Least-Square Support Vector Regressors.