Measurement of health evolution tendency for aircraft engine using a data-driven method based on multi-scale series reconstruction and adaptive hybrid model

Abstract

With the improvement of structure complexity and the strict requirement for stable operation, maintenance pattern of aircraft engine has experienced the transformation from passive response to active prevention. Accurate result of health evolution trend measuring is the core part of conducting the prevention maintenance. In this paper, a data-driven method based on multi-scale series reconstruction and adaptive hybrid model is proposed to measure the health development tendency of aircraft engine. Firstly, to quantitatively characterize the health levels of engine, a comprehensive health state index (CHSI) is innovatively constructed by an improved ensemble auto-encoder (EnAE) and self-organizing map (SeOM) neural network, which realizes the feature-level fusion of multi-source sensory data. By designing novel network structure and training pattern in EnAE, original auto-encoder model is optimized and the more robust state features can be captured from raw signals. Secondly, considering the influence of series random fluctuations on forecasting results, a multi-scale intrinsic mode functions (IMFs) reconstruction strategy using the fast ensemble empirical mode decomposition based dispersion entropy (FEEMD-DE) theory is provided to efficiently extract the regular and irregular components from original CHSI series. Finally, an adaptive hybrid model, combining a recurrent reconstructed Grey Markov (RRGM) model and long short-term memory (LSTM) network, is developed to capture the complex characteristics of reconstruction components and then complete the measurement of health evolution trend. The feasibility and superiority of the proposed method is validated by using a multi-source sensory dataset collected from aircraft engines, and the experimental results indicate that the measuring accuracy of the proposed method is significantly higher than that of other existing methods.