Journal
APPLIED SOFT COMPUTING
Volume 135, Issue -, Pages -Publisher
ELSEVIER
DOI: 10.1016/j.asoc.2023.110041
Keywords
Aircraft engine; Remaining useful life; Ensemble deep learning; Health state-related
Ask authors/readers for more resources
This study proposes a health-state-related ensemble deep learning method for predicting the remaining useful life (RUL) of aircraft engines. The method divides the lifetime degradation into multiple health states and trains three deep learning methods on different health states to learn different degradation laws. By calculating the self-adaptive ensemble weight sets, the prediction results of each algorithm model in different health states can be comprehensively utilized. The experimental results demonstrate that the proposed method can significantly improve prediction performance.
Remaining useful life (RUL) prediction for aircraft engines is crucial to enabling predictive maintenance. Current RUL predictions for aircraft engines mainly focus on model-based and data-driven methods that employ a single model or algorithm. Few studies on RUL prediction have been conducted by using an ensemble method that combines prediction results from multiple algorithms. As an emerging frontier technology, ensemble learning has become a topic of interest in the field of RUL prediction because it can achieve better prediction performance than single model. In this study, a health-state-related (HSR) ensemble deep learning method that considers different degradation laws of the aircraft engine is proposed for RUL prediction. First, a health baseline is constructed and lifetime degradation is divided into several health states to represent different degradation laws. The Mahalanobis distance to the health baseline is utilized to recognize the current health state of the aircraft engine. Second, three deep learning methods, namely stacked autoencoder, convolutional neural network and long short-term memory, are selected as member algorithms and trained on different health states. Thus, different member algorithm sets are constructed for different health states, learning different degradation laws in different health states. Third, self-adaptive ensemble weight sets for different health states are calculated by applying ridge regression, which can comprehensively utilize the prediction results of each algorithm model in different health states. A case study is conducted by using a dataset of the PHM data challenge to demonstrate the effectiveness of the proposed method. The experiment result shows that the proposed HSR ensemble deep learning method can considerably improve prediction performance compared with methods that are based on a single prediction algorithm and ensemble learning method that does not consider the health state.(c) 2023 Elsevier B.V. All rights reserved.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available