Remaining Useful Life Prediction of Aero-Engine using CNN-LSTM and mRMR Feature Selection

Abstract

Remaining useful life (RUL) prediction is critical in prognostics and health management (PHM) applications, where the trend of data-driven approaches using operational data from complex systems to establish degradation processes has attracted increasing attention. In order to solve the problem of feature screening and life prediction in the remaining service life prediction of aero-engine, this paper proposes Max-Relevance and Min-Redundancy (mRMR) method to screen sensor features by calculating the mutual information between features, uses the advantage of LSTM network in processing time series data, constructs samples by time window sliding, and designs the RUL direct prediction framework of CNN-LSTM network. The validation experiments were carried out on the C-MPASS dataset. Experimental results show that compared with other single deep learning and traditional models, the proposed hybrid model has lower regression analysis error and degradation prediction error, and can obtain more accurate remaining useful life prediction results.