Trend detection and data mining via wavelet and Hilbert-Huang transforms

This paper presents the formulation and evaluation of effective algorithms of reliable data analysis for real-time monitoring of incipient faults and anomalies, data fusion and event classification. The objective is to alleviate the shortcomings of the existing techniques for data mining by taking advantage of nonlinear filtering to handle non-Gaussian and non-stationary multiplicative noise and uncertainties. New concepts have been developed toward characterization of the data features and behavior interpretation of the underlying processes to evaluate their performance. In particular, the techniques of wavelet transform, Hilbert-Huang transform, and symbolic encoding are investigated to explore their effectiveness and relative simplicity to interpret and implement data mining tasks.