Federated Fleet Learning with Explainable AI: A Privacy-Preserving Architecture for Aviation Predictive Maintenance

Abstract

Engine maintenance stands at the centre of operational sustainability and cost governance in civil aviation, yet the sector confronts a paradox in data governance. Unscheduled failures impose severe financial penalties and safety hazards, but carriers remain reluctant to share operational data—treating it as proprietary intelligence—and competitive pressures suppress horizontal collaboration. Centralised machine-learning paradigms that require raw data transfer are structurally incompatible with these constraints. This paper presents a comprehensive literature review designed to address this deadlock from the standpoint of information systems and technology management. It conceptualises a Federated Fleet Learning architecture that enables a shared corporate intelligence network while preserving Data Sovereignty. Three interlocking components are examined as pillars of a Decision Support System: the 1DCNN-BiLSTM hybrid, suited to the constrained processing capacity of Aircraft Edge Units; the FedProx optimisation algorithm for managing statistical heterogeneity across disparate fleets; and asynchronous LEO satellite strategies for mitigating network latency. SHAP-based Explainable AI (XAI) integration is scrutinised within a technology governance frame to meet FAA and EASA DO-178C certification demands and to ensure algorithmic accountability. A noteworthy finding emerges: fleets with disjoint failure profiles—one versed solely in turbine degradation, another in compressor faults—can illuminate each other’s corporate “Blind Spots” through Cross-Fleet Knowledge Transfer without any raw data exchange. The study amalgamates algorithmic feasibility with industrial policy, laying a theoretical foundation for a privacy-preserving digital transformation paradigm in aviation.