Bayesian Inference for PRA
This session covers the application of Bayesian inference methods in Probabilistic Risk Assessment (PRA). The objective is for participants to be able to describe inference processes as part of PRA applications. We will describe how to update Bayesian priors and apply tools such as OpenBUGS using the techniques described in the Springer book Bayesian Inference for Probabilistic Risk Assessment (coauthored by the lecturer, Dr. Curtis Smith). In the session, we will address a variety of issues related to using probabilistic models for estimating PRA parameters. We will provide background to the analysis framework, then proceed to demonstrate the analysis of varying-complexity problems from traditional conjugate-types of inference through applications including uncertain data and trending.
Workshop on INL RAVEN Software
RAVEN is a software tool to characterize the probabilistic behavior of complex systems. It might be used for risk analysis, reliability analysis, uncertainty quantification and code validation. In most cases, RAVEN employs a “black box” approach with respect to the external code representing the physical systems (more advanced options like dynamic event trees are also available) and provides sampling strategies to effectively explore the input space. Standard statistical post-processing capabilities are provided to compute mean, variance, etc. of selected figures of merit of the output space. RAVEN relies heavily on artificial intelligence algorithms to construct surrogate models of complex physical systems to perform reliability analysis (limit state surface), uncertainty quantification and parametric studies.
The code license may be requested from Cristian Rabiti (email@example.com) or Andrea Alfonsi (firstname.lastname@example.org). The software (including source) is currently free for non-commercial uses. Commercial usage will be possible in the future under a new license structure.
The first objective of the workshop is to acquire a general understanding of the RAVEN package and its main capabilities. Secondly, a series of practical examples are going to be provided, in ascending level of complexity, starting from the simplest statistical analysis to the generation of the complex surrogate models and their utilization in reliability analysis. Users that already have access to the code will be able to run the examples directly on their laptops. Those that do not have access to the software yet, will receive a copy of the example inputs in electronic format. Depending on the Conference room availability “guest” accounts are going to be provided in order to execute workshop examples in a remote server
Workshop on EDF R&D tool PyCATSHOO
The safety requirements of its nuclear and hydraulic fleet, has allowed EDF to have long-standing experience in using and developing PRA and PSA tools for complex systems. During PSA 2017 EDF will present to the general public their latest development: PyCATSHOO a tool dedicated to dependability analysis of hybrid systems; i.e., systems including deterministic continuous phenomena and discrete stochastic behavior. Currently PyCATSHOO is used at EDF to perform several safety studies. During 2017, EDF will release PyCATSHOO under a freeware license. PyCATSHOO is a C++ written library. It has two APIs (Application Programming Interfaces) in Python and C++. These APIs provide a set of tools, based on distributed hybrid stochastic automata, which help in modeling and assessment of complex hybrid systems. PyCATSHOO thus combines the power of an object oriented language (Python or C++) and LEGO type modeling approach. A hybrid stochastic automation may exhibit random transitions between its states according to a predefined probability law and it may exhibit deterministic transition governed by the evolution of physical parameters. One can summarize the modeling approach with PyCATSHOO as follows: 1.) a system is divided into a functional or any other wway into elementary subsystems/components, 2.) Each of elementary subsystem/component is described as a set of hybrid stochastic automata, state variables, and message boxes, 3.) Message boxes ensure message exchange between subsystems//components, that ensure their dependencies, 4.) the system behavior is simulated using Monte Carlo sampling, and 5.) Sequences that lead to desirable end states are traced and clustered. PyCATSHOO offers a flexible modeling framework that allows to define generic components (classes) that can be resulted in different studies. It can greatly reduce model development costs.
During the workshop the tool will be presented as well as a workshop on several practical examples using the Python interface. Basic knowledge of Python is desired but not required. In order to manipulate the tool, EDF will provide a Linux Virtual machine under Oracle VirtualBox (available for Mac, Windows, Linus) that has to be downloaded on a personal computer.