Collaborative sparse classification for aero-engine’s gear hub crack diagnosis

Abstract

It is a big challenge to robustly detect the early crack fault of the differential gear train’s gear-hub of an aero-engine from the vibration signals of its engine casing, because of imprecise dynamic model guidance, extremely weak signature, complex modulation effects and limited training data. In this paper, a novel collaborative sparse classification framework (CSC), which collaborates the prior knowledge based sparse filtering and data-driven classification strategy, is proposed as a new endeavor for health condition assessment of aero-engine’s gear-hub. The sparse filtering model collaborates the empirically established fault pattern and its intrinsic local self-similar properties, by which the feature to interference ratio is enhanced. Subsequently, a sparse classification method is adopted to further explore the latent discriminative signatures and thus the health conditions of gear-hub can be automatically recognized. This work can not only recognize the abnormal vibration with high accuracy but also locate is source component to some extent. The effectiveness, superiority, parameter robustness and generalization performance of the proposed framework are thoroughly demonstrated by enormous comparison experiments with the state-of-the-arts.