Development of extended MVEM based fault residue generators using UKF state observers

Mean Value Engine Models (MVEM) are used to model the averaged dynamics of an automobile engine system for control and fault diagnosis. One approach to automobile fault diagnosis is to employ the use of a bank of residual generators each of which use a fault model and produces fault residues. These fault residues could then be used to detect or isolate faults using fault detection logic in the Fault Diagnoser. In this paper, the process of building such residue generators as Unscented Kalman Filter (UKF) based state observers that use different MVEM fault models has been described. Analytical redundancy can be used to detect sensor faults and is used to estimate sensor bias in a Mass Air Flow (MAF) sensor. A new fault model to detect flows has been described and is used to estimate Exhaust Gas Recirculation (EGR) flow rates for EGR valve sensor fault detection and leak flow for exhaust manifold leak detection.