الفهرس 
Chapter 1: IntroductionOnly 14 pages are availabe for public view
 |  | from | 143 |  |  |
| | | from | 143 | | |
Abstract
In the recent years, the rapid progress of digital technology to safety-critical
systems such as the Reactor Protection System (RPS) is increased. The RPS is
one of the most important digital Instrumentation and Control (I&C) systems
utilized in Nuclear Power Plants (NPPs) for safe operation and shut-down of the
reactor in emergency events. It ensures a safe reactor trip when the safetyrelated parameters violate the operational limits and conditions of the reactor.
Reliability assessment of RPS is an essential issue to maintain a high degree of
reactor safety and cost savings. Achieving high reliability and availability of
RPS ensures that the NPPs able to perform required tasks under given
conditions over a time interval. This work aims to perform reliability analysis
and improvement of the RPS; this is done through three new suggested models.
First, a quantitative evaluation reliability analysis model for the RPS with 2-outof-4 architecture using the state transition diagram is proposed. The model
assesses the effects of independent hardware failures, Common Cause Failures
(CCFs), and software failures on the RPS failure through calculating the
Probability of Failure on Demand (PFD) which characterizes the safety of the
RPS. The effectiveness of the suggested model is verified by comparing the
obtained results with that of the Fault Tree Analysis (FTA) technique.
Second, a general methodology for improving reliability of the RPS in NPP
based on a Bayesian Belief Network (BBN) is suggested. The structure of the
BBN model is based on the incorporation of failure probability and downtime of
the RPS I&C modules. Various architectures with dual state nodes for the RPS
I&C components are developed for reliability sensitive analysis and to
demonstrate the effect of RPS I&C modules on the failure of the entire system.
A reliability framework clarified as a Reliability Block Diagram (RBD)transformed into a BBN representation is constructed for each architecture to
identify which one will fit the required reliability. Two common component
importance measures are applied to define the impact of RPS I&C modules,
which revealed that some modules are more risky than others and have a larger
effect on the failure of the RPS.
Third, a hybrid machine learning reliability evaluation model of the nuclear
RPS in NPP is proposed. Initially, the significant reliability factors in RPS are
classified to four sections: Interlocks, over-power DT-shutdown, overtemperature DT-shutdown, and reactor shutdown logic circuit. Each section
represents a subsystem of the integrated RPS safety system. Feature selection
strategy based on Principal Component Analysis (PCA) is applied for low
impact factors exclusion. Then, the Support Vector Regression (SVR) algorithm
is utilized to promote machine learning models for the reliability evaluation of
each subsystem. Particle Swarm Optimization (PSO) is applied for each
learning model for parameter optimization of SVR. Thereafter, Artificial Neural
Network (ANN) is employed to correlate the reliability of the four subsystems
with the reliability of the whole RPS with utilizing PSO for an optimization
process to minimize the number of training phases. At last, residual error
Correction of Markov chain is adapted to improve the predictive performance of
the proposed learning model.
In comparison with some various literature methods, the achieved results
demonstrate the validity and effectiveness of the proposed work for reliability
evaluation and improvement of the RPS.
Keywords: Reactor Protection System (RPS), Nuclear Power Plant (NPP),
State Transition Diagram, Probability of Failure on Demand (PFD), Bayesian
Belief Network (BBN), Support Vector Regression (SVR), Artificial Neural
Network (ANN), Particle Swarm Optimization (PSO).