Int J Performability Eng ›› 2020, Vol. 16 ›› Issue (4): 510-519.

• Orginal Article •

### Availability Assessment of Complex Systems under Parameter Uncertainty using Dynamic Evidential Networks

Mohammed Bougofaa,*, Abderraouf Bouafiab, and Ahmed Bellaouara,*

1. aTransport Engineering and Environment Laboratory, Frères Mentouri Constantine 1 University, Constantine, 25000, Algeria;
bChemical Engineering and Environment Laboratory, Skikda, University, Skikda, 21000, Algeria
• Submitted on  ;  Revised on  ; Accepted on
• About author:
Mohammed Bougofa received his M.S degree in health and industrial safety from the Institute of Hygiene and Industrial Safety at the University of Batna in 2015. He is a Ph.D. student at Frères Mentouri Constantine University. His research interests include complex system safety, reliability, and availability.
Abderraouf Bouafia received his M.S degree in health and industrial safety from the Institute of Hygiene and Industrial Safety at the University of Batna in 2015. He is a Ph.D. student at the University of Août. His research interests include quantitative risk assessment and the domino effect of complex processes.
Ahmed Bellaouar is a professor in the Transportation-Engineering Department at the University of Frères Mentouri Constantine. His research interests include mechanics, maintenance, transportation, logistics, and safety.

Abstract:

In many dynamic complex systems, the insufficiency of data makes components state probability estimation more difficult in most cases. This paper presents a new approach for evaluating complex system availability under epistemic uncertainties, based on a combination of the Dempster-Shafer Theory (DST) and a dynamic Bayesian network (DBN). This combination is called a dynamic evidential network (DEN). DST is well known for its utility and purpose to express uncertain experts' judgments about components state beliefs for treating epistemic uncertainty. In addition, the dynamic evidential network makes it possible to propagate this uncertainty while taking into consideration dynamic evolution and the dependency between components using conditional mass tables. Finally, a case study is presented as a validation of this approach.