Online Transfer Learning-based Method for Predicting Remaining Useful Life of Aero-engines

Aero-engine is the heart of aircraft, and its reliability and safety are extremely important. It is necessary to predict its remaining useful life to achieve the purpose of early maintenance decision. In this paper, the existing aero-engine prediction methods are analyzed. Aiming at the problem of inaccurate training of prediction models due to the operation of aero-engines under multiple operating conditions and insufficient amount of data for specific operating conditions, an online transfer learning-based method for remaining useful life prediction is proposed to realize the transfer of prediction models between different operating conditions. After introducing the idea of online transfer learning, based on the HomOTL-I algorithm, this paper makes improvements to the regression problem to make it applicable to remaining useful life prediction, and validates the algorithm using aero-engine test data. The results show that the online transfer model can obtain higher accuracy and convergence speed compared with the offline and online models, proving the effectiveness of the method on the small sample prediction problem.