Aggregation-based fault diagnosis algorithms

Abstract

We propose and discuss a fault diagnosis algorithm in which the aggregative modeling approach is explored. We assume that a user has a model of a properly working system, a number of models of the known system faults and an auxiliary generic model based on Volterra expansion to represent the unknown malfunctions. The algorithm has two phases:The aggregation algorithm is used to evaluate the Volterra model from the set of the system input-output measurements {(xn, yn)}, n = 1,..., N.All models are compared with another set of system measurements, and the one with the smallest error indicates the type of the system fault (or lack thereof).The illustrative experiments are performed on a Wiener-Hammerstein system in presence of a heavy noise.