Data-driven dynamic health index construction for diagnosis and prognosis of Engine Bleed Air system

Abstract

The Engine Bleed Air system is a critical component in aircraft operations, providing necessary air supply for various onboard systems. Failures in the Engine Bleed Air (EBA) System can lead to flight delays, extended downtime, and safety risks. The current practice of using fixed pressure thresholds for EBA monitoring has limitations in terms of maintenance efficiency and aircraft safety. This paper presents a data-driven approach to dynamic thresholding and health index construction for the Airbus A330 EBA. A substantial EBA flight dataset is constructed using Quick Access Recorder (QAR) data, incorporating normal and faulty states. To explore the extensive QAR data of the EBA system, a data-driven baseline mining model is proposed in this study. To efficiently process high-dimensional feature data and model the pressure baseline, the LightGBM tree-based algorithm is employed. Additionally, this study proposes a health index (HI) construction method based on the baseline model, along with the EBA diagnosis and prognosis experiments based on the HI index. The Diagnosis and Prognosis methods, utilizing the proposed HI, demonstrate superior diagnostic effectiveness compared to fixed threshold methods and uncover a clearer trend of EBA health degradation. These contributions highlight the potential of data-driven approaches in managing aircraft EBA systems, emphasizing the advantages of dynamic thresholds and health index models for improved diagnosis and prognosis.