Carole Duval

Methodological developments for probabilistic risk analyses of socio-technical systems

Nowadays, the risk analysis of critical systems cannot be focused only on a technical dimension. Indeed, well-known accidents in nuclear or aerospace areas underlined initiating causes also related to technical and organizational viewpoints. This led to the development of methods for risk assessment considering three main aspects on the system resources: the technical process, the operator constraining the process, and the organization constraining human actions on the process. However, only few scientific works have tried to join these methods in a unique and global approach. Thus this paper focuses on a methodology that aims to achieve the integration of the different methods in order to assess the risks probabilistically. The integration is based on (a) system knowledge structuring and (b) its unified modelling by means of Bayesian networks also supporting quantification and simulation phases. The methodology is applied to an industrial case to show its feasibility and to draw conclusions regarding the model relevance for system risk analysis. The results of the methodology can be used by decision makers to prioritize their actions when faced with potential or real risks.

A safety barriers-based approach for the risk analysis of socio-technical systems

Usually, an efficient interaction between different resources of an industrial system (technical, human and organizational) leads to an efficient operation of this system. If this interaction is too weak due to missing or failing resources, the system can evolve to inoperative or risky situations, which can be hazardous for critical systems (as nuclear power plants and chemical processes). Thus methodologies are needed to support risk analysis by integrating together these system dimensions. Nevertheless few existing methodologies are able to perform this task and are mainly dedicated to partial or specific application domains. To face this gap, the paper presents a new methodology based on a system knowledge unification and its structuring in order to quantitatively estimate risks. Then the proposed approach integrates explicitly safety barriers, considered as key parts for risks prevention, and modeled by means of Bayesian networks. Finally a barrier example is depicted in the paper to highlight the feasibility of the methodology.

A Bayesian network-based integrated risk analysis approach for industrial systems: application to heat sink system and prospects development

In less than a decade, the European context on industrial risk management has evolved in order to propose frameworks to improve knowledge of both hazardous events and systemic analysis. Some frameworks are addressing more precisely socio-technical systems, considered as complex systems, operating under environmental constraints and for which multiple risks exist. Indeed, the physical and regulatory environment strongly influence the different stakes of a socio-technical system, mainly its availability, but also its safety. Nevertheless as these systems cannot be studied as a set of independent sub-systems owing to complexity, the conventional risk analysis is not applicable to them. More integrated risk analysis development is required, globally covering all the risks in a same view, taking into account system models (e.g. functional and organizational), system life cycle phase, system environment, the potential role of maintenance, and the human actions. In relation to this context, Electricité De France (EDF), which is managing socio-technical systems dedicated to energy production, took the opportunity to contribute to this issue. Thus, this article is defending a ‘system thinking’-based integrated risk analysis approach. Integrated risk analysis covers different disciplines (i.e. dependability, human reliability, and organizational analysis) and is designed for developing methods and appropriate tools in order to support innovative risks analysis of such systems. This approach is justified with regard to other risk analysis approaches in order to highlight the benefits of an integrated approach compared with the usual studies that are specific (technical or environmental or human centred). These main concepts and principles, and the adequacy of Bayesian networks to integrated risk analysis models, are demonstrated by applying them to an industrial case that is a sub-set of an EDF energy power plant (a heat sink). Finally, based on the results of sensitivity studies performed to ensure the robustness of Bayesian network-based integrated risk analysis models, major prospects development and ways to tackle them are identified mainly related to the robustness of risk assessment, the modelling of the human barrier, and the resilient aspects of the organization.

Expert judgments collecting and modeling: Application to the Integrated Risks Analysis (IRA) methodology

Assessment of different types of risks is today one of the challenges for an Integrated Risks Analysis (IRA) methodology. Indeed, whereas technical or environmental risks assessments can generally be done by means of statistical way, human and organizational considerations are more taken into account with the use of expert judgments. These considerations lead, from a scientific point of view, to address issues such as how the information provided by the experts can be collected and then modeled. Thus, this paper aims at reviewing different ways needed to express expert knowledge but also different frameworks for representing the information collected. These two items have to support the full development of the IRA methodology.

Contribution à la modélisation et au traitement de l’incertain dans les analyses de risques multidisciplinaires de systèmes industriels

Cette communication a pour but de montrer comment sont identifiees, propagees et exploitees les incertitudes de toutes natures (aleatoire et epistemique) dans la nouvelle approche developpee par EDF et le CRAN pour faire le lien entre des risques de natures differentes. Elle definira les differentes natures des incertitudes associees aux connaissances manipulees dans cette approche, identifiera et argumentera la methode retenue pour leur representation et leur traitement dans les modeles multi-disciplinaires de risques et presentera les resultats obtenus sur un cas concret pour l’aide a une prise de decision dans l’incertain.

Evidential Network-Based Extension of Leaky Noisy-OR Structure for Supporting Risks Analyses

Bayesian Networks (BN) are used in risks analysis because their capacities allow supporting complex system modeling. Nevertheless, to achieve some modeling, one BN issue is still the effort required for quantification even if some solutions are addressing the use of logical structures like OR, AND, Noisy-OR, Leaky Noisy-OR, etc. These structures are useful to represent different uncertainties but they do not allow taking into account uncertainty on their parameters, logically present in risks analysis. To face this challenge, this paper aims at proposing imprecise extensions of the Leaky Noisy-OR structures and a solution to implement these imprecise structures by using Evidential networks.