The aviation industry includes many actors ranging from airspace users, air traffic control, and airports to support service delivery. Operational safety, efficiency, and capacity are enhanced greatly through data sharing, however, with this also comes privacy concerns. Digital technologies, in particular, machine learning software benefit from access to high-quality datasets, prompting research into privacy-preserving data-driven models.

The AICHAIN solution enabled the privacy-preserving exploitation of large private datasets from different stakeholders to enrich operational machine-learning applications. This was achieved through privacy-preserving federated machine learning, where the training and serving of the federated models can be done at the data owners’ facilities in a cyber-secured and trustworthy manner without sharing any data. In this way, private data owners can remain in full control of their dataset’s privacy. A novel blockchain-based mechanism enhances the federated learning platform with two key features: 1) an audit trail to support model trustworthiness, as required for operational AI applications in airspace management; and 2) a system of tokens to implement direct incentives for the participants and fairness policies.

AICHAIN research began by proving the technical feasibility of the concept and its value to airspace management, along with consideration of governance and incentive mechanisms. Prototype architecture was used to demonstrate the solution through federated learning experiments and cyber-security assessments of the platform before two air traffic management use-cases enhanced private data from a federated airline. A framework of customizable tokens to implement rewarding policies was created, aimed at encouraging the effective collaboration of the private data owners in the federated processes.

These results could now be used to implement large-scale experiments with more airlines in the federation and under more realistic operational conditions.

Benefits

• Enhanced predictability, safety, capacity, and efficiency
• Enhanced cyber-security and privacy
• AI trustworthiness