Dr. Hong Xiao is a full professor at the School of Power and Energy of Northwestern Polytechnical University. He is the deputy director of the key laboratory of digital intelligence technology for performance and control of aircraft engines. 

Prof. Xiao is author and co-author of more than 100 peer-reviewed publications and 4 books on a wide range of digital intelligence topics for aircraft engines. His main research activities are dedicated to the artificial intelligence algorithms in the field of aircraft engines – including gas path, oil system, air system, and vibration faults. In particular the integration of artificial intelligence into the jet engine teat is in his focus. His work has contributed to numerous innovations in the digital twins, digital experimentation, and Prognostics Health Management, etc.

Between 2012 and 2015 Prof. Xiao was employed by Gyeongsang National University (South Korea) as a Research Professor for the numerical simulation in the aerospace field. He completed his academic work in the field of “mathematical problems in engineering” and served as research fellow at the department of engineering and the department of pure mathematics and mathematical Statistics of University of Cambridge between 2016 and 2019.

Keynote: Data-Driven Method Embedded Physical Knowledge for Entire Lifecycle Degradation Monitoring in Aircraft Engines

Abstract:

The degradation of aircraft engine performance necessitates real-time monitoring. We introduce a data-driven method embedded physical knowledge for entire lifecycle degradation monitoring in aircraft engines.

This model is based on Long Short-Term Memory and integrated with the engine’s physical topology. To counteract the data distribution disparities across various engine states during model training, a data augmentation method is given.

In a case study, data from 600h of engine running were employed, with thrust as the target monitoring parameter. The initial 50-h dataset covers the entire range of working conditions for the subsequent 550 h, serving as the training dataset for both models.

The results demonstrated the accuracy, with the model achieving a mean absolute relative error below 0.5% and a maximum absolute relative error below 8% in the initial 50 h. Simultaneously, the real-time model yielded a mean absolute relative error below 0.7% and maximum absolute relative error below 10% for the subsequent 550 h. A comparison of the real-time thrust degradation between this study and traditional methods highlights the feasibility and advantages of this present approach.