1 / 14

Use of Bayesian Networks in the risk analysis for an industrial plant :

Use of Bayesian Networks in the risk analysis for an industrial plant : Epistemological perspective of the modelling process of a system with technical and organisational dimensions. Régis.Farret@ineris.fr (1) Jean-Christophe.LeCoze@ineris.fr Myriam.Merad@ineris.fr

vienna
Download Presentation

Use of Bayesian Networks in the risk analysis for an industrial plant :

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Use of Bayesian Networks in the risk analysis for an industrial plant : Epistemological perspective of the modelling process of a system with technical and organisational dimensions Régis.Farret@ineris.fr (1) Jean-Christophe.LeCoze@ineris.fr Myriam.Merad@ineris.fr Carole.Duval@edf.fr (2) Aurelie.Leger@edf.fr (CRAN)(2), (3) (1) Institut national de l ’environnement industriel et des risques, France (2) EDF – Electricité de France, Research & Development, Dept MRI, France (3) Centre de Recherche en Automatique de Nancy, UMR 7039, CNRS-UHP-INPL, France

  2. Other valves Inspection of valves Valve Toxic effects on population Toxic gaz dispersion Overpressure Thermic effect Vapour Cloud Explosion Training of drivers Other accident scenarios Fault tree + Event tree Manual or Organisational Ei 1 EI ET Ei 2 EI  Over-filling  OU EM Ei 3 Rise of pressure Ph D EI OU EM Vessel Rupture Ei 4 OU EM Ei 5 Ph D EI ET EM Physical Agression EC 6 EI OU  Vehicle accident  Ein 7 EI OU EC 8

  3. Modelling Approach Uncertainty Decision making Bayesian Network (BN) The frame of our project > Our epistemological questions 2. Should a model be a faithful image of reality ? 3. Are we confident in the results ? 1. What is the goal of the modelling process ? • 4. What types of • uncertainties : • are we faced with ? • can we estimate ? Risk Analysis Organisational Analysis 5. Does the expert influence the model ? The results ? 6. What are the (expected) advantages of this BN tool ? What are its limits ?

  4. * Set of elements linked together in order to achive a goal* System > Sum (elements) Interaction with the envt, yet the model-ling process has to set boundaries * «cum plexus» = bound with* «Organised complexity» (Weaver, 47):auto-org°, feed-back, reconfiguration* e.g. engine failure < living organism WHAT ? The system studied • Industrial equipment / installation • Technical • Human and organisational • Open • Complex • Dynamic • Objective : • help optimum decisions ensuring security • Examples of practical questions: • Estimate the efficiency of a safety barrier, including all human and organisation  probability of 1 accident • Choose between 2 safety barriers (human / technical) • Estimate the impact of one change in the organisation  probability of various accidents

  5. A model is a representation of reality, in order to help in a decision or answer a specific question It is NOT a faithful image of reality - necessarily simplified (esp. for complex system) - drawn from a given point of view « The map is not the territory » WHY ? The objectives of modelling • Risk analysis is a particular case of modelling • 1. Identification of risks What can go wrong ? • 2. Characterisation of risks How often (how likely) ? •  If it goes wrong, what consequences ? • 3. Interpretation / Decision+ How confident am I in the result ? • Describe (represent) • Understand • Predict • Decide (or communicate) P G Uncertainty

  6. HOW ? (1) Our global modelling process Technical Organisational Expertise Expertise A. Definition of system + question Building the model B. Risk Identification C. Audition of operators Risk Analysis = Modelling D. Representation Quantifying E. Risk characte-risation (quantif.) F. Décision

  7. HOW ? (2)(A priori) Interest of Bayesian Networks • 1. Graphical (representation & communication, easy to handle) • 2. Probabilities included • Generalise the concept of Fault / Event Tree, with probabilistic links (influences) between variables (various “states”) > deterministic tree Other advantages : • Integrate figures / expert advice • Integrate (partially) correlated links • Possibility of dynamic networks • Integrates uncertainty through probabilities, as a mean of expressing both 1°) our ignorance and 2°) our knowledge

  8. (A priori) interest of Bayesian Networks : an example Thermic effect Cigarette Fire in the building Non application of security rules Presence of fuel Toxic effect High temperature outside Probabilistic link Deterministic link

  9. Uncertainty (1) : a typology • No absolute classification of uncertainty is established • Most known typology : • Variability(stochastic, objective) = intrinsic property of reality • Epistemologic(lack of knowledge, subjective) = depends on us ! • structural : ignorance of phenomena • choice of tool, model, method of audition / analysis • lack of data + difficulties of observations/measurement • Our proposal (adapted to our system + our approach) • Building the model • Quantifying

  10. Uncertainty : typology Modelling Rules for risk analysis + building the model Quantification Integration of proba + uncertainty by the B.Network

  11. Uncertainty : towards a strategy • Two main ways of tackling uncertainty : • Develop social/political strategy to : • live with uncertainty : e.g. precaution principle • manage uncertainty : e.g. choice of acceptability criteria  outside our scope • Develop better tools to know better : • Get more data / more precise data  outside our scope • Quantifying :  Improve estimation (quantification), through B.N. • Building the model (risk analysis + building the model) :  conceptual frame + guidance + « validation » by experts • Another human source of uncertainty : subjectivity • No analysis is possible without a sequence of (small) decisions based on the analyst ’s judgement • All the more so in the human/organisational field ! • NB : the analyst is NOT part of the system, but influences the observation/modelling process

  12. Our conceptual frame Level 3 : organisational(management, envtal factors) Level 2 : human (individual actionsor decisions) Level 1 : technical (bow-tie from risk analysis)

  13. (A posteriori) Interest of Bayesian Networks • 1. Graphical (representation & communication, easy to handle) • 2. Probabilities included in a rigourous way • Generalise the concept of Fault / Event Tree, • with probabilistic links (influences) between variables (various “states”), • which can express uncertainty and be integrated instantaneously • Main limits and specific recommendations • Specify the system boundaries & precise objective of the study (question ?) • Be aware that model = support for reasoning (e.g decide), NOT reality • Avoid too complicated graphs • Conceptual frame + Procedure (included a specific « protocole ») = build the model + manage the related uncertainty • 2 experts : both technical + organisational competences • Legitimation step : conceptual frame + approbation by a working group • Open questions : • How to limit subjectivity ? How to transfer natural language into probabilistic figures ? (identify scales, objective criteria…) • Is it legitimate to translate organisational influences into numbers ? • Is it useful to develop 3 BN for the 3 levels of our conceptual frame ?

  14. Thank you for your attention Régis.Farret@ineris.fr (1) Jean-Christophe.LeCoze@ineris.fr Myriam.Merad@ineris.fr Carole.Duval@edf.fr (2) Aurelie.Leger@edf.fr (CRAN)(2), (3) (1) Institut national de l ’environnement industriel et des risques, France (2) EDF – Electricité de France, Research & Development, Dept MRI, France (3) Centre de Recherche en Automatique de Nancy, UMR 7039, CNRS-UHP-INPL, France

More Related