Degradation variation pattern mining based on BEAST time series decomposition integrated functional principal component analysis

Abstract

The variety of operational conditions among comparable systems in a fleet leads to the creation of numerous samples (having multiple degradation paths) and information regarding system performance (featuring multiple state variables) within the fleet. A common technique for modeling degradation variation patterns in such fleets is functional principal component analysis, albeit often resulting in a loss of information on mutations related to the degradation of the system. This paper proposes a method to mine degradation variation patterns through a Bayesian estimator of abrupt change, seasonal change, and trend time-series decomposition integrated functional clustering. Assume that the functional characteristics evolve over time in the degradation paths of repairable systems, prompting the utilization of functional data analysis methods for clustering the corresponding degradation variation patterns. The BEAST method is used to analyze the impact of individual degradation variations on repairable systems, which can differentiate between abrupt changes, seasonal variations, and trends in the population of repairable systems. We then use this analysis to develop preventive maintenance optimization models and analyse the impact of change-points in the degradation process on the maintenance strategy. The study offers a robust methodology for analyzing fleet degradation, thereby enhancing the understanding of degradation patterns and optimizing preventive maintenance strategies.