0
Research Papers: Nuclear Power

Application of Possibilistic C-Means for Fault Detection in Nuclear Power Plant Data

[+] Author and Article Information
Annalisa Perasso

Dipartimento di Matematica,
Università di Genova,
Genova 16146, Italy
CNR—SPIN,
Genova 16146, Italy
e-mail: perasso@dima.unige.it

Cristina Campi

CNR—SPIN,
Genova 16146, Italy
e-mail: campi@dima.unige.it

Cristian Toraci

IRCCS San Martino IST,
Genova 16146, Italy
Dipartimento di Matematica,
Università di Genova,
Genova 16146, Italy
e-mail: cristian.toraci@unige.it

Francesco Benvenuto

Ansaldo Nucleare S.p.A.,
Genova 16146, Italy
e-mail: Francesco.Benvenuto@ann.ansaldo.it

Michele Piana

Dipartimento di Matematica,
Università di Genova,
Genova 16146, Italy
CNR—SPIN,
Genova 16146, Italy
e-mail: piana@dima.unige.it

Anna Maria Massone

CNR—SPIN,
Genova 16146, Italy
e-mail: annamaria.massone@cnr.it

Metallic rings inserted between the power channels and the external pipes in order to keep the coaxiality.

Contributed by the Nuclear Division of ASME for publication in the JOURNAL OF ENGINEERING FOR GAS TURBINES AND POWER. Manuscript received June 30, 2014; final manuscript received October 7, 2014; published online December 9, 2014. Assoc. Editor: Igor Pioro.

J. Eng. Gas Turbines Power 137(6), 062901 (Jun 01, 2015) (7 pages) Paper No: GTP-14-1317; doi: 10.1115/1.4028809 History: Received June 30, 2014; Revised October 07, 2014; Online December 09, 2014

This paper describes a classification method for automatic fault detection in nuclear power plant (NPP) data. The method takes as input time series associated to specific parameters and realizes signal classification by using a clustering algorithm based on possibilistic C-means (PCM). This approach is applied to time series recorded in a CANDU® power plant and is validated by comparison with results provided by a classification method based on principal component analysis (PCA).

FIGURES IN THIS ARTICLE
<>
Copyright © 2015 by ASME
Your Session has timed out. Please sign back in to continue.

References

Figures

Grahic Jump Location
Fig. 1

Analysis of the flux mapping routine in the RRS. Top left panel: initial data set made of 55 time series where each signal has been normalized and rescaled to arbitrary unit (each curve corresponds to one time series). Top right panel: the two classes produced by Algorithm 2 after the first iteration. Bottom left panel: membership values for one of the 95 runs where the second iteration does not provide any further classification (from A through I: nine signals classified as normal; L: one signal classified as outlier). Bottom right panel: membership values for one of the five runs where the second iteration provides a third class (from A through H: eight signals classified as normal; L: first signal classified as outlier; I: second signal classified as outlier.

Grahic Jump Location
Fig. 2

Analysis of temperature time series in the SCWS. Top left panel: initial data made of 90 time series where each signal has been normalized and rescaled to arbitrary unit (each curve corresponds to one time series). Top right panel: centroids associated to one of the final third iteration for one of the 75 runs of Algorithm 2. Bottom left panel: memberships for one of the 75 four-class outcomes (from A through G: seven signals classified as normal; L, I, H: first, second, and third outlier). Bottom right panel: memberships for one of the 25 three-class outcome of Algorithm 2 (from A through H: eight signals classified as normal; L, I: first and second outlier).

Grahic Jump Location
Fig. 3

Analysis of the power channels in the reactor system. Top left panel: initial data made of 95 time series where each signal has been normalized and rescaled to arbitrary unit (each curve corresponds to one time series). Top right panel: classification of Algorithm 2 after two iterations. Bottom panel: memberships for nine signals classified as normal (from A through H) and for the outlier provided by Algorithm 2 (L).

Grahic Jump Location
Fig. 4

Representation of the results provided by the PCA-based Algorithm 3, accounting for the fact that for this algorithm the memberships can be either zero or one. Top left panel: classification of the same data as in Fig. 1 (from A through I: nine signals classified as normal; L: one signal classified as outlier). Top right panel: classification of the same data as in Fig. 2 (from A through G: seven signals classified as normal; L, I, H: first, second, and third outlier). Bottom panel: classification of the same data as in Fig. 3 (from A through E: six signals classified as normal; L: the actual outlier; I, H, G: misclassified data).

Tables

Errata

Discussions

Some tools below are only available to our subscribers or users with an online account.

Related Content

Customize your page view by dragging and repositioning the boxes below.

Related Journal Articles
Related eBook Content
Topic Collections

Sorry! You do not have access to this content. For assistance or to subscribe, please contact us:

  • TELEPHONE: 1-800-843-2763 (Toll-free in the USA)
  • EMAIL: asmedigitalcollection@asme.org
Sign In