Title: Hierarchical spline for time series forecasting: An application to Naval ship engine failure rate

Abstract: Predicting equipment failure is important because it could improve availability and cut down the operating budget. Previous literature has attempted to model failure rate with bathtub-formed function, Weibull distribution, Bayesian network, or AHP. But these models perform well with a sufficient amount of data and could not incorporate the two salient characteristics; imbalanced category and sharing structure. Hierarchical model has the advantage of partial pooling. The proposed model is based on Bayesian hierarchical B-spline. Time series of the failure rate of 99 Republic of Korea Naval ships are modeled hierarchically, where each layer corresponds to ship engine, engine type, and engine archetype. As a result of the analysis, the suggested model predicted the failure rate of an entire lifetime accurately in multiple situational conditions, such as prior knowledge of the engine.