Mastering of complex systems by transdisciplinary rupture innovations GTR EN Management, Methods, Standard  Tools


Methods and tools


Probabilistic networks applied to risk mastering and dependability



The GTR follows the « bayesian networks : methods and application to risk mastering and dependability » working day jointly organized by IMdR and RUFEREQ in September 2012 in order to share experience between industry and academics on the use of bayesian networks to solve modelling problems linked with risk mastering and dependability.

Bayesian networks are probabilistic networks considered as a solid mathematic formalism, supported by software platforms of simulation. The appearance of ergonomic tools for modelling and probability computation allows bayesian networks exploitation  by a large community of industrial and academic users with a great spreading of usages  in many activity fields and numerous application types among which risk analysis, reliability computation, failure diagnosis, maintenance prevision, a.s.o… 

This mathematic formalism is recognized nowadays by the international scientific community. The researches  are numerous on the development of computation algorithms in networks bigger and bigger, integrating temporal variables, continuous. 

Probability networks have a strong modelling capacity and their computing precision is widely demonstrated, namely on reliabity computation. Their main interest  for complex system comprehension , owing to scenario simulation, insertion of new knowledges and diagnosis of causes of functionment and dysfunctionment… 

Usage precautions are to be taken, specifically in dynamic process modelling where temporal process complexity leads to big size models. Inference algorithms are continuously  improved to model more and more complex systems.

Despite the great maturity of formalism , approaches and model construction it is required to formalize the approaches, construction and structuration of the model . In fact an important effort is to be given to upstream model construction and knowledge structuration . Modelling  robustness lies upon  methodology construction accuracy . 

We propose indeed  that the WG focuses on ever open questions and some perspectives of model robustness, dependancy materialization, take into account uncertainty of nature whatever, opening towards more elaborated and more efficient models whose objective is to face very big size modelling problems.  The promising formalisms to investigate are the relational probabilistic model (PRM), the modelling by networks of faith function,a.s.o..These new formalisms have to be studied to master their potential and limits. We propose to follow the evolution and new adaptations of these modelling tools to answer  more and more complex concrete problems.

This GTR gives an  opportunity to meet actors specialized in operation of modelling formalism  under probabilistic networks to share knowledge and experience.Structures models under graph form such as bayesian networks should be part of the inescapable tools of dependability and risk mastering. The group objective is, on one hand to promote capacities of modelling and probability computation of graphic,probabilistic (  bayesian networks ,  evidential  networks, Relational probabilistic models, dynamic bayesian networks , bayesians continuous and hybrid networks), on other hand to illustrate their implementation on problems of risk mastering and dependability of various levels of complexity. Finally the WG aims at raising problematics which could be accroached by industrial and academic communities.


The  persons who wish to participate to the GTR are praid to register at : Philippe WEBER (CRAN) , Christophe SIMON (CRAN)