On-line fault diagnosis based on B-spline neural networks using asymptotic local approach

Abstract

Fault diagnosis is crucial in monitoring industrial processes. Faults can be often detected from residuals generated from the system model. For systems with known models, residuals can be readily generated. However, for systems with unknown models, neural networks can be used to model the system. For small or incipient faults, it is difficult to detect faults directly from the residuals. The asymptotic local approach, which transforms the fault diagnosis problem into one that detects statistical changes in a random variable, is proposed. The proposed scheme is illustrated by a simulation example, and comparison with faults obtained directly from the residuals is also made.