ANOMALY DETECTION APPROACH WITH THE HILBERT - HUANG TRANSFORMATION IN PROCESS SIGNALS
D. A. Murzagulov, A.V. Zamyatin, O.V. Romanovich
Tomsk State University, Tomsk, Russia
Keywords: anomaly, process signal, spectral analysis, statistical models, predictive analytics
Abstract
Issues of anomaly detection in nonstationary process signals are considered because anomalies could be an early indication of defects and equipment failure. An approach to detection of anomalies utilizing the Hilbert-Huang transformation and statistical model is presented. The main idea of the proposed approach is to analyze the statistical parameters of the components of the Hilbert-Huang transformation, which has high adaptive properties for nonstationary signals and provides precise resolution in the time-frequency domain. The paper describes the principal scheme and algorithm of the approach, detailed definition of the statistical model, numeric experiments on real and augmented data, and comparative analysis with similar anomaly detection methods.
|
