A Novel RUL Prognosis Model Based on Counterpropagating Learning Approach

Abstract

The aviation industry is one of the fastest-growing sectors and is crucial for both passenger transport and logistics. However, the high costs associated with maintenance, refurbishment, and overhaul (MRO) constitute one of the biggest challenges facing this industry. Motivated by the significant role that remaining useful life (RUL) prognostics can play in optimising MRO operations and saving lives, this paper proposes a novel data-driven RUL prognosis model based on counter propagation network principles. The proposed model introduces the recursive growing hierarchical self-organisation map (ReGHSOM) as a variant of SOM that can cluster multivariate time series with high correlations and hierarchical dependencies typically found in RUL datasets. Moreover, ReGHSOM is designed to allow this clustering to evolve dynamically at runtime without imposing constraints or prior assumptions on the hypothesis spaces of the architectures. The output of ReGHSOM is fed into the supervised learning layers of Grossberg to make the RUL prediction. The performance of the proposed model is comprehensively evaluated by measuring its learnability, evolution, and comparison with related work using standard statistical metrics. The results of this evaluation show that the model can achieve an average mean square error of 5.24 and an average score of 293 for the C-MPASS dataset, which are better results than most of the comparable works.