Research on combination of periodic inspection and data-driven maintenance system of turbofan engine

Abstract

In the field of aviation, turbofan engine is a crucial high-precision component of an aircraft which requires rigorous maintenance to ensure the running stability, safety, and efficiency. To reduce the high cost of general maintenance, the system of combination of periodic inspection and Data-driven maintenance is a meaningful research trend. In this pattern, the engine's condition study over sensor data serves as the foundation for the maintenance, to meet the safety requirements, improved sensor data processing and interpretation and precise prediction model plays important role. In this study, Long Short-Term Memory (LSTM) architecture was used to estimate RUL (remaining useful life) and support periodic inspection. The experiments were taken by using the dataset of FD001 and FD003 in Turbofan engine degradation dataset that contains operating and degradation information of several turbofan engines. After being analysed by different evaluation methods, the algorithm is capable to discover the features that hidden among sensor data, which indicates that the model has promising results for predictive maintenance. Therefore, the Data-driven maintenance structure that integrated with LSTM architecture is proposed as an application method. Furthermore, considering all aspects of the real application, management tools, Kubernetes and Kubeflow are introduced in this paper. Finally, regularly scheduled inspection is introduced, with the help of estimated RUL, inspection can be applied by parts rather than removing engine for overhaul and replacement of all parts including parts in good condition.