A Method of Fast Fault Detection Based on ARMA and Neural Network

Abstract

Current fault detection systems lack the ability to generalize from previously observed patterns to detect even slight variations of unknown faults. In this paper, ARMA model combining with a Hopfield-model net is proposed for describing a approach that provides the ability to generalize from previously observed behavior to recognize future behavior. The approach can be used for fault detection in order to analyze and detect novel anomaly patterns. Meanwhile, a feedback neural network was used to predict the 'expected values' of the anomaly; Using the Neural Network is especially better since it can improve the detection rate without increasing the false positives. Experiments show events and variance of anomaly patterns.

Multivariate Fault Detection using Vector Autoregressive Moving Average and Orthogonal Transformation in Residual Space

“…Previous work using autoregressive models in the fault detection area can be found in Schöener, Moser, and Lughofer (2008), M. Yang and Makis (2010) and T. Yang (2006). These use AR, ARX and ARMA models respectively, but we have not found evidence of VARMA models applied for the purpose of fault detection.…”