Abstract
The remaining useful life (RUL) and utilization strategy of an aero-engine are related to the flight safety of an aircraft, which directly affects the flight itself and the safety of the occupants. Aiming at the complexity of aero-engine condition monitoring data, an aero-engine predictive maintenance planning framework based on RUL prediction is proposed, which aims to analyze the engine RUL and design predictive maintenance strategies. First, a deep learning integrated model (Trans-LSTM), including Transformer and Long Short Memory Network Model (LSTM), is proposed. Second, Bayesian optimization is used to optimize the hyperparameters of the integrated model to further improve the accuracy of the predictive model. Based on the prediction data, an engine alarm threshold was designed. When the threshold is triggered during engine operation, a predictive maintenance task is applied. The optimal alarm threshold under the Trans-LSTM model is calculated by comparing the total flight cost and other indicators under different flight hours. Experimental results demonstrate that the data-driven predictive maintenance strategy can monitor engine status in real time, promptly identify potential failure risks, and prevent engines from operating in an unknown state. This effectively reduces the risk of sudden engine failures and significantly enhances flight safety compared to the periodic maintenance strategy. In addition, through the accurate prediction of the engine state and reasonable arrangement of maintenance tasks, it can effectively reduce the cost of using the engine and avoid the waste of manpower, material and financial resources caused by excessive maintenance. Moreover, it enhances engine task availability, prolongs the engine's optimal operating period, better meets the actual needs of air transportation, and brings higher economic benefits and operational efficiency for airlines, thus showing great value and potential in practical application.