FLARE – Federated Learning for Aircraft Reliability Enhancement

Modern aircraft engines are equipped with multiple sensors that continuously record data over time in a complicated environmental setting. Using these measurements, predictive maintenance applications to estimate the Remaining Useful Life (RUL) to predict the engine's failure and decide when maintenance should be performed. This article introduces a secure federated learning framework for collaborative aircraft engine RUL prediction system development named FLARE (Federated Learning for Aircraft Reliability Enhancement), that preserves user data confidentiality. The framework compares centralized learning, local-only learning, and federated learning using multiple machine learning models like Linear Regression, Random Forest, Gradient Boosting, Support Vector Regression and XGBoost along with a federated LSTM architecture. Furthermore, four federated optimization strategies, namely FedAvg, FedProx, FedNova, and a quality- weighted dynamic aggregation method (FedDWA), are implemented and analyzed to improve performance under varying data distributions using NASA C-MAPSS turbofan engine dataset.