SGBRT: An Edge-Intelligence Based Remaining Useful Life Prediction Model for Aero-Engine Monitoring System
Tiantian Xu, Guangjie Han, Linfeng Gou, Miguel Martínez-García, Dong Shao, Bin Luo, Zhenyu Yin
Abstract
In this paper, we develop an <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">edge intelligence</i> based aero-engine performance monitoring system. The proposed approach can effectively predict the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">remaining useful life</i> of aero-engines, which is the main focus within the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">prognostics and health management</i> framework – thus it provides support for optimal operation planning and maintenance decisions. The proposed model, which we term SGBRT, follows a hybrid machine learning approach, combining a <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">self-organizing mapping</i> network with a <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">gradient boosting regression tree</i> model. In particular, the SGBRT computes the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">remaining useful life</i> of an aero-engine in two steps: it first employs a <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">self-organizing map</i> to cluster the sample data; and then it fits each cluster by way of a <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">gradient boosting regression tree</i> . Detailed simulation results with the C-MAPSS dataset show that this method achieves a higher prediction accuracy and better generalization than other conventional approaches; the compared methods range from classical approaches such as a <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">switching Kalman filter</i> to state-of-the-art deep learning models.