Remaining useful life prediction of aero-engine enabled by fusing knowledge and deep learning models. (January 2023)

• Record Type: Journal Article
• Authors: Li, Yuanfu; Chen, Yao; Hu, Zhenchao; Zhang, Huisheng
• Is Part Of: Reliability engineering & system safety. Volume 229(2023)
• Publication Date: 2023-01
• Subjects: Knowledge -- Deep learning -- CNN -- LSTM -- RUL prediction; Reliability (Engineering) -- Periodicals; System safety -- Periodicals; Industrial safety -- Periodicals; Fiabilité -- Périodiques; Sécurité des systèmes -- Périodiques; Sécurité du travail -- Périodiques; 620.00452
• DOI: 10.1016/j.ress.2022.108869
• Languages: English
• ISSNs: 0951-8320

Abstract

Highlight: A novel framework for Prognostics and Health Management situations with numerous sensors is proposed. The framework fuses knowledge and deep learning models for RUL prediction. Knowledge is utilized to guide the construction of deep learning models. The proposed approach is compared with state-of-the-art deep learning methods on the C-MPASS dataset. Results indicate that constructed model has higher interpretability and higher prediction accuracy. Abstract: The remaining useful life (RUL) prediction of a complex engineering system is extremely significant for ensuring system reliability. The conventional prediction of the RUL based on only extracted degradation features of sensor data is tedious for decreasing costs and providing a decision-making foundation. However, knowledge is available for improving RUL prediction accuracy. This paper proposes a novel RUL prediction approach that combines knowledge and deep learning models. The proposed approach represents the sensor relationships as flow charts to be transformed as embedding vectors for clustering. These clustering results are subsequently utilized to guide the sensor data arrangement and hybrid deep learning model construction. Compared to various deep learning models, the robustness and reliability of the proposed method are demonstrated on the NASA open dataset C-MAPSS. The results show that the proposed approach had improved prediction accuracy by 5.5% compared to the best prediction from the literature methods. Furthermore, the constructed deep learning model by utilizing knowledge can be interpretable. Most importantly, the prediction results reveal the feasibility and reliability of fusing knowledge and deep learning models. And the proposed approach is promising for widespread application to other prediction situations with data from numerous sensors.