Robust Reliability Redundancy Allocation Guided By Multi-Component Importance Measure

Reliability has been considered as a significant design measure in various industrial systems. One of the most efficient approaches to maximize system reliability is to employ redundant components in parallel, which is known as the reliability redundancy allocation problem.Systems may be subjected to abnormal external failuremodes(AEFMs), whose probabilities are hard if not impossible to predict, such as failures caused by natural disasters(earthquakes, floods, tornados, etc.) and intentional attacks. To deal with this uncertainty and based on robust optimization, this research proposes a multi-objective optimization approach that can simultaneously optimize system reliability under both normal and the worst cases of abnormal situations for reliability redundancy allocation problems.Considering the impact of unpredictable external risks the system may face, and using optimization techniques to allocate the resources reasonably will enhance the capacity of a system to withstand unpredictable external risks and make the system more secure and efficient.Since the amount of AEFMs is not known, the worst case for each possible amount of AEFs needs to be identified. For this purpose, a new importance measure, multi-component importance measure(MCIM), that can quantify the importance of combinations of multiple components with arbitrary amount is proposed.Based on MCIM, we also propose a heuristic method to find the worst case of the system when AEFMs appear.Reliability redundancy allocation problems belong to the NP-hard class, exact methods cannot give optimal solutions for large scale problems efficiently. Due to the complexity of the problem, this research revises a widely used multi-objective evolutionary algorithm, MO-PSDA, to solve the proposed problem. Experimentation shows that the revised algorithm is able to provide a better approximation to the true Pareto optimal solutions for the proposed problem in a more stable way than the original algorithm.