A new era in synthetic data generation in aviation may be on the horizon with the launch of the SynthAIr project - Synthetic Data Generation for ATM Systems. The initiative is a response to the scarcity of relevant data for aviation and the inherent limitations of AI models in handling diverse datasets.
The project, which kicked off in September in Trondheim – Norway, comprises four partners from four different countries, Sintef, Deep Blue, Eurocontrol and Delft University of Technology that, under the coordination of Massimiliano Ruocco from Sintef, who will lead the research in this innovative field, during its 30 months duration.
With the rapid advancement of AI technologies, the aviation industry is at a critical juncture where data availability and ubiquity pose significant challenges to the development of AI-based tools, particularly those related to safety-critical data. SynthAIr is poised to address this pressing issue with an innovative approach.
"As SynthAIr launches, we stand at the cusp of a transformative era in aviation and ATM”, says the Project Coordinator Massimiliano Ruocco (Sintef). “Synthetic data promises to overcome data scarcity, enhancing our predictive modeling capabilities. This project is our pathway to fully unlock AI's potential in ATM systems, driving groundbreaking innovations and efficiencies. Together, we're charting a bold new course for the SESAR community."
Co-funded within the framework of Horizon Europe and SESAR Joint Undertaking, the primary objective of SynthAIR is to pioneer AI-based methods for synthetic data generation in the ATM system domain. This initiative is a response to the scarcity of relevant data and the inherent limitations of AI models in handling diverse datasets. By exploring AI-based methods for synthetic data generation, SynthAIR aims to overcome these obstacles while simplifying the process for users with varying levels of expertise.
At the heart of the project lies the groundbreaking concept of the Universal Time Series Generator (UTG). Inspired by recent advancements in computer vision and language technology, the main idea is to learn a model from multiple datasets and generate synthetic data that accurately represents new, unseen datasets. This process is achieved by conditioning the UTG on a compressed representation of the target data. In the aviation domain, for example UTG can be trained on data from a few specific airports and subsequently utilized to generate synthetic data for entirely new airport environments. This approach extends to creating a Universal Time Series Forecaster (UTF) capable of making accurate predictions for new environments without the need for additional training.
“Synthetic data stands as a beacon of hope, promising to elevate the predictive modeling capabilities to unprecedented heights” continues Massimiliano Ruocco. “It's not just about having more data; it's about having the right kind of data that can train our systems more effectively. With SynthAIr, the aim is not only to enhance the quality of predictions but also to fast-track the integration of AI in ATM systems. The potential of AI in aviation is vast, and with the right data, SynthAIr could unlock efficiencies, safety measures, and innovations that were previously beyond anyone’s reach.”
