A hybrid-learning decomposition algorithm for competing risk identification within fleets of complex engineering systems

There is an increasing interest in the reliability of complex engineering systems, especially in the systems’ through-life risk analysis. A complex system, like the civil aircraft engine studied in this paper, contains multiple potential failure modes throughout its life that are contributed by various sub-system and component failures going through different deterioration processes. In order to fulfill the requirements of efficient swap and replacement maintenance strategies in the aviation industry, it is important to quantify the individual component risks within a complex system to enable an accurate prediction of spare parts demands. We propose a novel data-driven hybrid-learning algorithm with three building blocks: pre-defined reliability model based on the Weibull distribution, automated unsupervised clustering, and the quality check & output. The algorithm enables the identification of the riskiest sub-systems and the associated reliability models are quantitatively calculated. As all component risks follow the Weibull distribution, the parameters can be obtained. A case study carried out on a fleet of civil aircraft engines shows that the algorithm enables a better understanding of sub-system level risks from system level performance records, improving the efficient execution of the maintenance strategy.