The Integration of Federated Learning Techniques in Predictive Aircraft Maintenance Using Cloud Services

Federated Learning (FL) has emerged as a key research topic, providing a decentralized approach to model training while preserving data privacy. This research focuses on the application of FL in Predictive Aircraft Maintenance (PdM), a critical domain for aviation safety and efficiency. Traditional data sharing among airlines for PdM is hindered by legal restrictions, making FL a potential solution, since existing centralized learning models require data aggregation that poses privacy risks. In recent years, different opensource frameworks for FL have been developed, such as PySyft, TensorFlow Federated (TFF), FATE, and Flower. In this study, the Flower framework for FL is used to implement a federated learning workflow, comparing the performance of various federated aggregation algorithms against a centralized baseline. This framework was chosen for its clear documentation and gradual learning curve, making it well-suited for a research setup. Additionally, the compatibility of the Flower framework with the Google Cloud Platform (GCP) was investigated. Results indicate that FL provides competitive performance compared to Centralized Learning (CL) when applied to a PdM scenario while ensuring data privacy, adding to the FL knowledge pool and illustrating the potential of FL for fields involving privacy-sensitive data.