Abstract
Many real-word systems exhibit nonlinear and nonstationary dynamics, which defy understanding based on the traditional reductionist's approach. However, traditional analytical methods designed to effectively handle nonlinear dynamics are not well integrated with multisensor data fusion for process monitoring and control objectives. Realizing full potentials of multiple sensor signals calls upon the development of new methods for anomaly detection and transition analysis of nonlinear dynamics in complex systems. This article presents a novel pattern-frequency tree (PFT) approach for multisensor signal fusion and dynamic transition analysis. We leverage both pattern and frequency information in the PFT model to develop efficient algorithms for modeling and analysis of abnormal transitions in the nonlinear state space. Experimental results demonstrate that the proposed PFT method achieves a superior performance for multisensor data fusion and anomaly detection in nonlinear dynamical systems.
