Degradation Monitoring Using Probabilistic Inference.

Degradation Monitoring;Particle Filtering;Sample Impoverishment;Nuclear Engineering and Radiological Sciences;Engineering;Nuclear Engineering & Radiological Sciences

Alpay, BulentMartin, William R. ;

University of Michigan

关键词: Degradation Monitoring;    Particle Filtering;    Sample Impoverishment;    Nuclear Engineering and Radiological Sciences;    Engineering;    Nuclear Engineering & Radiological Sciences;   

Others  :  https://deepblue.lib.umich.edu/bitstream/handle/2027.42/60690/balpay_1.pdf?sequence=1&isAllowed=y

瑞士|英语

来源: The Illinois Digital Environment for Access to Learning and Scholarship

【 摘 要 】

In order to increase safety and improve economy and performance in a nuclear power plant (NPP), the source and extent of component degradations should be identified before failures and breakdowns occur. It is also crucial for the next generation of NPPs, which are designed to have a long core life and high fuel burnup to have a degradation monitoring system in order to keep the reactor in a safe state, to meet the designed reactor core lifetime and to optimize the scheduled maintenance. Model-based methods are based on determining the inconsistencies between the actual and expected behavior of the plant, and use these inconsistencies for detection and diagnostics of degradations. By defining degradation as a random abrupt change from the nominal to a constant degraded state of a component, we employed nonlinear filtering techniques based on state/parameter estimation.We utilized a Bayesian recursive estimation formulation in the sequential probabilistic inference framework and constructed a hidden Markov model to represent a general physical system. By addressing the problem of a filter’s inability to estimate an abrupt change, which is called the oblivious filter problem in nonlinear extensions of Kalman filtering, and the sample impoverishment problem in particle filtering, we developed techniques to modify filtering algorithms by utilizing additional data sources to improve the filter’s response to this problem. We utilized a reliability degradation database that can be constructed from plant specific operational experience and test and maintenance reports to generate proposal densities for probable degradation modes. These are used in a multiple hypothesis testing algorithm. We then test samples drawn from these proposal densities with the particle filtering estimates based on the Bayesian recursive estimation formulation with the Metropolis Hastings algorithm, which is a well-known Markov chain Monte Carlo method (MCMC). This multiple hypothesis testing algorithm using MCMC in particle filtering helps the filter to explore the state space more effectively in the direction of the degradations. We extended this algorithm for degradation detection and isolation to complete the degradation monitoring framework. We successfully tested our algorithms in degradation monitoring of balance of plant of a boiling water reactor.