The aviation and air traffic management (ATM) domain is undergoing a major transformation, driven by the need for enhanced automation and the potential of artificial intelligence (AI). SynthAIr stands at the forefront of this transformation, aiming to address data scarcity and improve ai-based models for atm automation. The project has several key objectives:

• Improve understanding of ai-based synthetic data generation methods: SynthAIr seeks to assess the current state of ai methods for synthetic data generation, particularly for time series data, and adapt them to the atm domain. The focus is on leveraging ai methods from other domains to generate historical multi-feature timestamped data specific to aviation.

• Develop universal ai methods for generating realistic synthetic atm data: The project aims to create high-fidelity, easily generalisable, and privacy-preserving synthetic datasets. It will explore universal time series models to enhance both synthetic data generation and predictive modelling in atm.

• Evaluate privacy, fidelity, and diversity of generated synthetic data: Recognising the benefits of synthetic data—cost reduction, bias minimisation, and enhanced diversity—SynthAIr will develop metrics to assess privacy preservation, fidelity, and diversity.

• Validate models and their application in operational use cases: Specific operational use cases will measure the impact of the proposed ai methods, focusing on downstream tasks such as forecasting and prediction, particularly in scenarios with unbalanced datasets.

• Recommend conclusions, principles, and best practices: Drawing on experimental results and collaboration with the advisory board, SynthAIr will provide recommendations and best practices for the ethical use of novel ai methods in atm.

• Accelerate ai uptake in the atm system and foster collaboration: SynthAIr aims to deliver novel ai methods leveraging synthetic data, accelerating ai adoption in atm. The project also promotes collaboration with practitioners, end users, and stakeholders through co-creation, experience sharing, and training opportunities.

To evaluate, test, and validate these ai methods, SynthAIr has identified five operational use cases:

1. Turnaround time prediction: Predicting aircraft turnaround times, which affect over 40% of primary delays at airports, using large datasets to create a single model applicable across different scopes.

2. Flight delay prediction: Reducing the impact of flight delays by improving predictive models through data augmentation and universal models for scenarios with limited data.

3. Passenger flow prediction: Predicting passenger flows at airport terminals to improve service quality, operational efficiency, and resource allocation, given that over 50% of delayed flights are caused by passenger-related issues.

4. Synthetic traffic generator for simulations: Generating synthetic traffic to support real-time simulations, test new atm concepts, and train reinforcement learning algorithms.

5. Flight diversion prediction: Predicting flight diversions due to factors such as adverse weather by augmenting training datasets with synthetic positive observations to improve model performance.

In essence, SynthAIr is poised to transform ATM by harnessing the power of synthetic data, bridging the gap between ai’s potential and practical application, and fostering collaboration among key stakeholders in the aviation ecosystem.