A hierarchical decomposition approach for large system reliability allocation | Zastosowanie metody dekompozycji hierarchicznej do alokacji niezawodności w dużych systemach

Xiaoling Zhang,XiaoZhong Huang,Yü Liu

2009-01-01

Abstract:Reliability has become a great concern in recent years, especially for large system consisting of a large number of subsystems, modules and components. To achieve the reliability goal in design stage, reliability allocation, a method to apportion the system target reliability amongst subsystems and components in a well-balanced way, has since received increasing attention. However, seeking the optimal reliability allocation scheme for a system with bunch of subsystems and components is not straightforward, and it is known as an NP-hard problem. An abundance of work has been carried out to investigate the computational efficient methods, e.g. exact algorithm, heuristic algorithm and meta-heuristic algorithm etc., to handle the optimization of reliability allocation for the complex system. Even though the proposed methods in past research work well for system consisting of a moderate set of components, they will still suffer "curse of dimensionality" and be impossible to converge if the system consisting of tens/hundreds of subsystems and components which maybe exist in industrial engineering. To mitigate the deficiency, a decomposition strategy is proposed, in which the reliability allocation problem for the system with a large number of components is decomposed into a set of smaller, coordinated sub-problems which can be solved via traditional optimization algorithm in an computational efficient manner. Target cascading method, as an efficient hierarchical decomposition method, is employed in this paper to decompose the large system reliability allocation problem into a set of hierarchical optimization problems in according with the system configuration. To illustrate the efficiency and effectiveness of the proposed method, a numerical example is presented, as well as some comparative studies.