Probabilistic risk assessment based model validation method using Bayesian network

- Probabilistic approach to validation.
- Performance-based probabilistic risk assessment (PRA) validation.
- Bayesian approach to validation in risk-informed decision making environment.
- Concept of overlapping coefficient as validation index.

Past few decades have seen a rapid growth in the availability of computational power and that induces continually reducing cost of simulation. This rapidly changing scenario together with availability of high precision and large-scale experimental data has enabled development of high fidelity simulation tools capable of simulating multi-physics multi-scale phenomena. At the same time, there has been an increased emphasis on developing strategies for verification and validation of such high fidelity simulation tools. The problem is more pronounced in cases where it is not possible to collect experimental data or field measurements on a large-scale or full scale system performance. This is particularly true in case of systems such as nuclear power plants subjected to external hazards such as earthquakes or flooding. In such cases, engineers rely solely on simulation tools but struggle to establish the credibility of the system level simulations. In practice, validation approaches rely heavily on expert elicitation. There is an increasing need of a quantitative approach for validation of high fidelity simulations that is comprehensive, consistent, and effective. A validation approach should be able to consider uncertainties due to incomplete knowledge and randomness in the system's performance as well as in the characterization of external hazard. A new approach to validation is presented in this paper that uses a probabilistic index as a degree of validation and propagates it through the system using the performance-based probabilistic risk assessment (PRA) framework. Unlike traditional PRA approaches, it utilizes the power of Bayesian statistic to account for non-Boolean relationships and correlations among events at various levels of a network representation of the system. Bayesian updating facilitates evaluation of updated validation information as additional data from experimental observations or improved simulations is incorporated. PRA based framework assists in identifying risk-consistent events and critical path for appropriate allocation of resources to improve the validation.