Minimum rotation partitioning for data analysis and its application to fault detection

Symbolic dynamics provide a new set of tools for data analysis, fault detection and investigation of the dynamical systems. The main concept is partitioning the phase space into a finite number of non-overlapping segments that provide a low-dimensional representation of time series. By simplifying the dynamics this way, a novel method for nonlinear analysis of systems, including fault progression, can be constructed from observed data. This paper presents a novel space partitioning technique, referred as minimum rotation partitioning for the purpose of fault detection and quantification. The results obtained from a permanent magnet synchronous machine is presented as an example of fault detection and quantification.