Monitoring nonlinear and non-Gaussian processes using Gaussian mixture model based weighted kernel independent component analysis
Monitoring nonlinear and non-Gaussian processes using Gaussian mixture model based weighted kernel independent component analysis
A kernel independent component analysis (KICA) is widely regarded as an effective approach for nonlinear and non-Gaussian process monitoring. However, the KICA-based monitoring methods treat every KIC equally and cannot highlight the useful KICs associated with fault information. Consequently, fault information may not be explored effectively, which may result in degraded fault detection performance. To overcome this problem, we propose a new nonlinear and non-Gaussian process monitoring method using Gaussian mixture model (GMM)-based weighted KICA (WKICA). In particular, in WKICA, GMM is first adopted to estimate the probabilities of the KICs extracted by KICA. The significant KICs embodying the dominant process variation are then discriminated based on the estimated probabilities and assigned with larger weights to capture the significant information during online fault detection. A nonlinear contribution plots method is also developed based on the idea of a sensitivity analysis to help identifying the fault variables after a fault is detected. Simulation studies conducted on a simple four-variable nonlinear system and the Tennessee Eastman benchmark process demonstrate the superiority of the proposed method over the conventional KICA-based method.
122-134
Cai, Lianfang
b696f3db-590f-4a8f-b31c-b9a6eb9fe0ae
Tian, Xuemin
5b7f2306-69c1-41c7-8cab-49932ac1ae01
Chen, Sheng
9310a111-f79a-48b8-98c7-383ca93cbb80
1 January 2017
Cai, Lianfang
b696f3db-590f-4a8f-b31c-b9a6eb9fe0ae
Tian, Xuemin
5b7f2306-69c1-41c7-8cab-49932ac1ae01
Chen, Sheng
9310a111-f79a-48b8-98c7-383ca93cbb80
Cai, Lianfang, Tian, Xuemin and Chen, Sheng
(2017)
Monitoring nonlinear and non-Gaussian processes using Gaussian mixture model based weighted kernel independent component analysis.
IEEE Transactions on Neural Networks and Learning Systems, 28 (1), .
(doi:10.1109/TNNLS.2015.2505086).
(PMID:26685274)
Abstract
A kernel independent component analysis (KICA) is widely regarded as an effective approach for nonlinear and non-Gaussian process monitoring. However, the KICA-based monitoring methods treat every KIC equally and cannot highlight the useful KICs associated with fault information. Consequently, fault information may not be explored effectively, which may result in degraded fault detection performance. To overcome this problem, we propose a new nonlinear and non-Gaussian process monitoring method using Gaussian mixture model (GMM)-based weighted KICA (WKICA). In particular, in WKICA, GMM is first adopted to estimate the probabilities of the KICs extracted by KICA. The significant KICs embodying the dominant process variation are then discriminated based on the estimated probabilities and assigned with larger weights to capture the significant information during online fault detection. A nonlinear contribution plots method is also developed based on the idea of a sensitivity analysis to help identifying the fault variables after a fault is detected. Simulation studies conducted on a simple four-variable nonlinear system and the Tennessee Eastman benchmark process demonstrate the superiority of the proposed method over the conventional KICA-based method.
Text
TNNLS2017-Jan.pdf
- Version of Record
Restricted to Repository staff only
Request a copy
Text
ngmm-wkica.pdf
- Accepted Manuscript
More information
Accepted/In Press date: 30 November 2015
e-pub ahead of print date: 17 December 2015
Published date: 1 January 2017
Organisations:
Southampton Wireless Group
Identifiers
Local EPrints ID: 404253
URI: http://eprints.soton.ac.uk/id/eprint/404253
ISSN: 2162-237X
PURE UUID: 5558c85d-3c3a-465e-b024-655593efd36e
Catalogue record
Date deposited: 05 Jan 2017 10:17
Last modified: 15 Mar 2024 04:03
Export record
Altmetrics
Contributors
Author:
Lianfang Cai
Author:
Xuemin Tian
Author:
Sheng Chen
Download statistics
Downloads from ePrints over the past year. Other digital versions may also be available to download e.g. from the publisher's website.
View more statistics