Abstract:

A general mixture signal model (MSM) together with support vector data description (SVDD) are proposed to address the monitoring of non-Gaussian processes.Mixture signal model involves Gaussian, non-Gaussian and measurements noises.Methods to extract and monitor the corresponding mixture signals are presented.A general SVDD kernel function parameterization and optimization approach is proposed to monitor the non-Gaussian signal sources.Industrial application demonstrate that the general proposed kernel function is capable of characterizing the non-Gaussian behaviors encapsulated in process data and detect abnormal events promptly.