System reliability considering component failure and repairs: genetic and memetic algorithms

Due to the increasing complexity of structures and functions of high-tech systems, such as modern engineering and service systems in the communication, electronics, and space technology fields, the need for effective evaluation tools for such systems has increased in parallel. In the analysis and design of complex systems, the system reliability is one of the performance criteria commonly used in the literature to evaluate performance. The concept of system reliability generally refers to the operation of a system in accordance with its purpose. The reliability optimization of a system with various problem-specific constraints is an important and current problem. The Redundancy Allocation Problem (RAP), which is widely used in the literature, can be defined as the design of new systems with higher reliability using redundant components in a parallel arrangement. The RAP is an NP-hard problem. While most of the studies on RAP generally only consider failures, few studies are based on the assumption that failures and/or repairs occur at constant rates. However, in real-life problems, the components in the system have an increased failure rate due to wear and tear from regular use. Due to the computational difficulty in RAP studies, it is generally assumed that components are not repaired after they fail and thus become out of order. Without an opportunity to repair a failed part, the initial cost of the system increases. Under such circumstances, the system has to be designed with better components (i.e., highly reliable but expensive) to maintain the overall system reliability at the desired level. In real-life applications, a component failure has a negative impact on system reliability. After the failed component is repaired and put back to work, the reliability of the system increases. Structures designed with repair considerations are more realistic, and high reliability can be obtained at a lower cost by using this approach. In addition, the "k" parameter in the metrics (k-out-of-n) is dynamically adjusted according to changing user requirements. The (k-out-of-n) metric is used as a reliability constraint in RAP, and the introduction of a dynamic "k" facilitates the design of adoptable systems to seasonal changes in demand. In this thesis, two metaheuristic algorithms, Genetic Algorithm (GA) and Memetic Algorithm (MA) are proposed to optimize RAP with increasing failure rates, component repairs, and dynamic "k". The system reliability (objective function value) in the RAP optimization is estimated with a discrete event Monte Carlo simulation with discrete event simulation (DES) model, which is developed according to the occurrence of events with a realistic approach. The validity of the DES model has been demonstrated on the test problems given in the literature. The effectiveness of the developed GA and MA is shown on the test problems that are widely used in the literature. According to the computational analysis, GA and MA found quality solutions. It has been shown that MA gives better results than GA. Thus, problem-specific GA and MA for RAP designs with higher reliability at lower cost, where failure and repair are taken into account, have been developed and brought to the relevant literature.