Steady-state and transient operation discrimination by Variational Bayesian Gaussian Mixture Models

Zhang, Yu and Bingham, Chris and Gallimore, Michael and Chen, Jun (2013) Steady-state and transient operation discrimination by Variational Bayesian Gaussian Mixture Models. In: IEEE International Workshop on Machine Learning for Signal Processing (MLSP 2013), 22-25 September 2013, Southampton, UK.

Documents
PID2950955.pdf
[img] PDF
PID2950955.pdf
Restricted to Repository staff only

587kB
Item Type:Conference or Workshop contribution (Paper)
Item Status:Live Archive

Abstract

The paper presents a Variational Bayesian (VB) method to allow a Gaussian Mixture Model (GMM) to be clustered automatically with its mixture components in order to facilitate the discrimination of what can be regarded as steady-state and transient machine operation. The determination of whether a unit is considered to be in steady-state, or subject to external transients is an important pre-processing scenario for both sensor- and machine-fault detection algorithms, for instance, Principal Component Analysis (PCA) based Squared Prediction Error (SPE), which is known to produce excessive ‘false alarms’ when fed with measurements that include transient unit operation. Here, the resulting Variational Bayesian Gaussian Mixture Model (VBGMM) method is utilized to discriminate the operational behaviour of industrial gas turbine systems. Daily batches of measurement data from in-the-field systems are used to show that the VBGMM provides a useful pre-processing tool for subsequent diagnostic and prognostic algorithms.

Keywords:Steady-state operation, transient operation, Gaussian mixture model, variational Bayesian inference
Subjects:G Mathematical and Computer Sciences > G310 Applied Statistics
Divisions:College of Science > School of Engineering
ID Code:12554
Deposited On:20 Nov 2013 11:51

Repository Staff Only: item control page