Multifractal information fusion based condition diagnosis for process complex systems

Abstract

Abnormal conditions are hazardous in process complex systems, and the aim of condition diagnosis is to detect abnormal conditions and thus avoid serious accidents. Comparing with conventional techniques of condition diagnosis without concerning the nonlinearity of complex system, multifractal analysis elaborately reveals scale-invariance or self-similarity properties of time series data, which is one of the intrinsic characteristics of complex systems. Moreover, the monitoring data within multiple feature variables should be investigated by combining multifractal analysis and information fusion techniques, so that significant patterns of the whole system would be discovered. In this article, a condition diagnosis framework is proposed for industrial complex systems, by which nonlinear features are extracted from univariate time series through multifractal analysis using multifractal detrended fluctuation analysis algorithm, and multiple feature variables are investigated through Mahalanobis–Taguchi system as an information fusion method to determine the condition of the whole system. The effectiveness of the approach is illustrated using data from both simulated model and real production system in an industrial enterprise.