In view of the difficulty in identifying the fault signal of the inter-shaft bearing of the aero-engine

a deep gradient boasting model (Deep-GBM) was proposed to improve the precision score by learning the feature of vibration signal step by step. Fault simulation experiment was conducted on a type of aeroengine intershaft bearing. Vibration fault signal was decomposed through empirical mode decomposition (EMD) method

and intrinsin mode function (IMF) component sample entropy of nonlinear dynamics parameters was collected as base features. With the model proposed

the aero-engine inter-shaft bearing was diagnosed respectively with fault in inner ring

comprehensive fault in inner ring and rolling element

and in normal condition

with stick-spalled fault and stick-scratched fault. The experimental results showed that the fault diagnosis accuracy of the Deep-GBM reached 87%

28% higher than that of the traditional machine learning model. Besides

the model has been proved to have good generalization ability.