Artificial neural network for sensor failure detection in an automotive engine

This paper presents a sensor failure detection method with artificial neural network for critical complex equipments, where their dynamic performance usually can not be actually obtained. However, if the number of measured variables is higher than the order of the system, using the inherent redundant relationship among sensors, we can detect the sensor failure and even recover or estimate the correct readings of the failed sensor. Two kinds of artificial neural network were trained to accomplish these aims, one is used to detect the sensor failure and the other is applied to recover the readings of failed sensor. The feasibility of this method was proved with computer simulation through a mathematic model of an automotive engine in a hovercraft.