Dynamic reconstruction-based fuzzy neural network method for fault detection in chaotic system

This paper presents a method for detecting weak fault signals in chaotic systems based on the chaotic dynamics reconstruction technique and the fuzzy neural system (FNS). The Grassberger-Procaccia algorithm and least squares regression were used to calculate the correlation dimension for the model order estimate. Based on the model order, an appropriately structured FNS model was designed to predict system faults. Through reasonable analysis of predicted errors, the disturbed signal can be extracted efficiently and correctly from the chaotic background. Satisfactory results were obtained by using several kinds of simulative faults which were extracted from the practical chaotic fault systems. Experimental results demonstrate that the proposed approach has good prediction accuracy and can deal with data having a −40 dB signal to noise ratio (SNR). The low SNR requirement makes the approach a powerful tool for early fault detection.