Aircraft engine turbine RUL prediction using NADINE

Abstract

RUL prediction has become a widely researched topic in recent years. This paper describes the use of the deep learning approach Neural Network with Dynamically Evolving Capability (NADINE) to overcome RUL prediction challenges used in static deep learning methods - the need for predefined initial network structure and parameters. NADINE offers a fully flexible and self-growing network capable of growing its hidden layers and hidden nodes on demand without the use of problem-specific parameters. Despite its standard MLP structure, it adopts two strategies to overcome the problem without compromising the performance of the network - that is the adaptive memory strategy and soft forgetting. The use of a dynamic self-growing network has demonstrated decent performance on RUL regression prediction tasks.