Abstract

The paper extends traditional Gaussian Mixture Model (GMM) techniques to provide recognition of operational states and detection of emerging faults for industrial or complex systems. A Variational Bayesian (VB) method allows a GMM to cluster with its Mixture Components (MCs) to facilitate the extraction of steady-state operational behaviour — this is recognised as being a primary factor in reducing the susceptibility of alternative prognostic/diagnostic techniques which can initiate false-alarms resulting from control set-point and load changes. Furthermore, a GMM with an Outlier Component (GMMOC) is discussed and applied for direct fault detection. To demonstrate the efficacy of the proposed techniques, real-time measurements from operational Industrial Gas Turbines (IGTs) show that the resulting VBGMM facilitates the selection of the number of required MCs to cluster the data, and thereby provide essential input for operational signature recognition. Moreover, GMMOC is shown to facilitate the early detection of emerging faults. An advantage of the VBGMM over traditional pre-defined thresholds is the extraction of steady-state data during both full- and part-load cases, and a primary advantage of the GMMOC method is its applicability for novelty detection when there is a lack of prior knowledge of fault patterns. Results based on measurements taken from IGTs operating in the field are therefore also included which show that the techniques provide an integrated pre-processing, benchmarking and novelty/fault detection methodology.