Symbolic dynamics of wavelet images for pattern identification

Author

Xin Jin ; Gupta, S. ; Mukherjee, K. ; Ray, A.

Author_Institution

Dept. of Mech. Eng., Pennsylvania State Univ., University Park, PA, USA

fYear

2010

fDate

June 30 2010-July 2 2010

Firstpage

3481

Lastpage

3486

Abstract

Symbolic time series analysis has been introduced in recent literature for pattern identification in dynamical systems. Relevant information, embedded in the measured time series, is extracted in the form of symbol sequences by partitioning of the data sets, and probabilistic finite state automata are constructed from these symbol sequences to generate pattern vectors. This paper presents a symbolic pattern identification method by partitioning of two-dimensional wavelet (i.e., scale-shift) images of sensor time series data. The proposed method is experimentally validated on a laboratory apparatus for identification of evolving patterns due to fatigue damage in polycrystalline alloy specimens.

Keywords

feature extraction; finite state machines; image recognition; probabilistic automata; time series; wavelet transforms; data set partitioning; dynamical system identification; fatigue damage; feature extraction; pattern vectors; polycrystalline alloy specimens; probabilistic finite state automata; sensor time series data; symbol sequences; symbolic pattern identification method; symbolic time series analysis; two-dimensional wavelet image partitioning; wavelet image symbolic dynamics; Automata; Data mining; Feature extraction; Filtering; Neural networks; Principal component analysis; Signal analysis; Time series analysis; Wavelet analysis; Wavelet transforms; Anomaly detection; Feature extraction; Pattern Identification; Symbolic dynamics; Wavelet Images;

fLanguage

English

Publisher

ieee

Conference_Titel

American Control Conference (ACC), 2010

Conference_Location

Baltimore, MD

ISSN

0743-1619

Print_ISBN

978-1-4244-7426-4

Type

conf

DOI

10.1109/ACC.2010.5531077

Filename

5531077