Comparison of Model-based Vs. Data-driven Methods for Fault Detection and Isolation in Engine Idle Speed Control System

Abstract

An internal combustion engine operating at idle is regulated by a feedback controller so that it runs at a preset idle speed without stalling when no acceleration is requested from the driver. Idle speed control is affected by numerous disturbances, ranging from accessory loads to environmental conditions. Because of the regulating behavior of the controller, faults, especially actuator faults, may affect sensor measurements in a way very similar to disturbances, system uncertainty or noise. This poses a challenge to the fault detection and isolation (FDI) problem for this system. In this paper, two fundamentally different fault diagnoses approaches are used to detect and isolate faults. A model-based residual generation scheme as well as a data-driven linear discriminant analysis scheme is developed to solve the FDI problem even when faults are concurring in addition to system uncertainty, disturbance and noise. Their performances are compared side by side using data gathered from an experimentally validated simulator for an engine idle system that considers an actuator fault, a sensor fault, several system uncertainties and disturbance (operating conditions), and sensor noise. The results show that comparable performance can be achieved with both schemes and some comments are made about each approach.