Probabilistic Inspection of Multimodally Distributed Signals

0444754 - ÚTIA 2016 RIV GB eng C - Konferenční příspěvek (zahraniční konf.)
Ettler, P. - Puchr, I. - Jirsa, Ladislav - Pavelková, Lenka
Probabilistic Inspection of Multimodally Distributed Signals.
The 12th International Conference on Condition Monitoring and Machinery Failure Prevention Technologies (CM 2015/MFPT 2015). Oxford, UK: The British Institute of Non-Destructive Testing, 2015. ISBN 978-1-5108-0712-9.
[The 12th International Conference on Condition Monitoring and Machinery Failure Prevention Technologies (CM 2015/MFPT 2015). Oxford (GB), 09.06.2015-11.06.2015]
Grant CEP: GA MŠMT 7D12004
Institucionální podpora: RVO:67985556
Klíčová slova: condition monitoring * probabilistic logic * binomial opinion * multimodally distributed signal * Gaussian mixture
Kód oboru RIV: BC - Teorie a systémy řízení
http://library.utia.cas.cz/separaty/2015/AS/jirsa-0444754.pdf

Condition monitoring of complex systems starts from an inspection of single measured signals. Results of health evaluation for each signal then enter the monitoring framework which uses a special type of probabilistic logic to assess condition of the whole inspected system while respecting its internal structure. A distribution of the involved signal can be multimodal in general. Therefore, a Gaussian mixture is more suitable for its approximation than a single theoretical distribution. The mixture estimated from a sufficiently long record of a particular signal can be used not only for its prediction but also for evaluation of the signal health. However, finding the values in a low probability region does not necessarily indicate bad condition of the signal – it can be interpreted as an increase in the information uncertainty instead. This approach is being tested on industrial data within the running project which aims to develop the novel probabilistic condition monitoring system.
Trvalý link: http://hdl.handle.net/11104/0247505