A TCN-NTM Model for Remaining Useful Life Prediction for Aircraft Engine

In the prediction of aircraft engine's remaining useful life (RUL), deep learning has been extensively employed due to its capability of extracting RUL-related features from vast amounts of engine data. This paper introduces a novel deep learning architecture, TCN-NTM, based on temporal convolutional networks (TCNs) and neural Turing machines (NTMs), which effectively processes large datasets and captures temporal dependencies among data. Furthermore, during preprocessing, the RUL of aircraft engines is segmented and sampled using a sliding window algorithm to facilitate subsequent deep learning processing. Experimental results on the C-MAPSS dataset demonstrate that the TCN-NTM approach yields remarkable performance. Through ablation experiments, the effects of TCN and NTM on RUL prediction have also been verified. This method is expected to play a significant role in the health management and fault detection of aircraft engines in the future.