Construction Method of Turbine Engine Health Indicator Based on Deep Learning

Abstract

Traditional turbine engine health indicator (HI) construction methods generally require manual feature extraction, feature selection and even feature fusion, besides, training labels need to be designed in advance, which make the whole procedure time consuming and not universal. Therefore, this paper proposes a novel unsupervised construction method of turbine engine health indicator based on stacked denoising autoencoders (SDAE). In this method, the deep structure of autoencoders adaptively extracts features of raw turbine engine monitoring signals in an unsupervised way to obtain its health indicator. Experimental results on CMAPSS engine dataset show that the HI curves constructed by the proposed method can well reflect the degradation process of turbine engine during the whole life cycle, and have better correlation and monotonicity compared to the traditional HI construction methods. Moreover, the proposed method does not need to rely on complex signal processing measures, the whole process is carried out in an unsupervised manner with a certain degree of versatility.