Predicting component reliability and level of degradation with complex-valued neural networks

Abstract : In this paper, multilayer feedforward neural networks based on multi-valued neurons (MLMVN), a specific type of complex valued neural networks, are proposed to be applied to reliability and degradation prediction problems, formulated as time series. MLMVN have demonstrated their ability to extract complex dynamic patterns from time series data for mid- and long-term predictions in several applications and benchmark studies. To the authors' knowledge, it is the first time that MLMVN are applied for reliability and degradation prediction. MLMVN are applied to a case study of predicting the level of degradation of railway track turnouts using real data. The performance of the algorithms is first evaluated using benchmark study data. The results obtained in the reliability prediction study of the benchmark data show that MLMVN outperform other machine learning algorithms in terms of prediction precision and are also able to perform multi-step ahead predictions, as opposed to the previously best performing benchmark studies which only performed up to two-step ahead predictions. For the railway turnout application, MLMVN confirm the good performance in the long-term prediction of degradation and do not show accumulating errors for multi-step ahead predictions.
Document type :
Journal articles
Complete list of metadatas

Cited literature [60 references]  Display  Hide  Download

https://hal-supelec.archives-ouvertes.fr/hal-00879431
Contributor : Yanfu Li <>
Submitted on : Sunday, November 3, 2013 - 6:42:00 PM
Last modification on : Tuesday, March 5, 2019 - 9:35:06 AM
Long-term archiving on : Tuesday, February 4, 2014 - 4:28:10 AM

File

13_07_30_Ress_Fink_MLMVN.pdf
Files produced by the author(s)

Identifiers

Citation

Olga Fink, Enrico Zio, Ulrich Weidmann. Predicting component reliability and level of degradation with complex-valued neural networks. Reliability Engineering and System Safety, Elsevier, 2014, 121, pp.198-206. ⟨10.1016/j.ress.2013.08.004⟩. ⟨hal-00879431⟩

Share

Metrics

Record views

555

Files downloads

1678