It is in Skymantics’ DNA to address the needs of the aviation industry in the areas of data exchange and analysis. Some valuable lessons learned over the years is that it is a challenge environment due to three main factors:
- Complex potential scenarios. Aviation is a complex ecosystem of airspace users and air traffic managers. Current operations are more reactive with low predictability.
- Difficulty to capture data sources. Often post-operative data only reflects the reactive scenario after the operation is successfully completed and does not capture the planning scenario. However, in order to validate future concepts for Performance Based Navigation (PBM) and Performance Based Flow Management (PBFM), systems should be measured and validated using non-ideal conditions.
- Flight operations have commercial or safety sensitive information. While some of this information is likely available, it may be restricted due to LADD, SUI, or other means of data protections.
Take the use case of flight and flow scenarios in the National Airspace System (NAS), for example. If an Air Navigation Service Provider (ANSP), such as the FAA, wishes to prototype and test a new Air Traffic Flow Management service, it needs to validate the protocols and measures with a team of subject matter experts. The concept begins as a paper exercise, then is proven through a set of laboratory experiments. However, these experiments often use simulators which follow a set of scripted events due to the complexity of open-ended scenarios and the lack of availability of data which is also safe to use and share.
True, fast-time simulation can identify shortfalls and opportunities, but neither of these methods can demonstrate specific unpredictable events without incorporating Human-In-The-Loop (HITL). These HITL experiments are generally limited to a subset of scenarios which can be scripted and may be manipulated to support a “sunny day” scenario or a positive outcome despite a constrained scenario.
Meet synthetic data by Artificial Intelligence (AI)
Introducing AI generated synthetic data can break down these barriers. In a few words, synthetic data is a creation of realistic information but without the real data records. AI learns from real data to create similar but synthetic data. Just as AI can analyze thousands of faces and then create a new face from those images, it can also be used to create a day of air traffic scenarios by analyzing real data. Let’s make a comparison with traditional methods of analysis:
Cost and speed
Due to the difficulty to obtain relevant data to generate analysis scenarios, most simulations today operate based on scripted routes and expected flight operations rather than rainy day scenarios.
AI generated synthetic data can generate relevant scenarios mimicking real data such as weather, airspace user preferences, and demand conditions. This data can be generated at scale and much faster than the time it takes to configure and refine scenario rules.
Security and privacy
Production aviation data has sensitive information that cannot be shared or requires specific authorization. This especially affects information on operations and procedures executed by airspace operators. This barrier makes it difficult for data scientists to access the necessary information.
Using AI generated synthetic data obfuscates commercially sensitive information in which real data might be data mined. By using synthetic data, routes and preferences for scenario testing are not related to a single airline or user, thereby protecting corporate interests and privacy of travel.
Flexibility
Scenarios based on real data represent actual operations, limited in geography and time, and biased to certain situations (e.g., a good weather day). This limitation creates a barrier when simulating hypothetical scenarios that never or rarely occur. Often, these are the most important scenarios to test for!
Using AI to generate synthetic data, hypothetical scenarios can be generated including weather, typical airspace user reactions, and air traffic management playbooks to demonstrate or validate new operational concepts. In addition, corner cases can be configured to accurately represent system behavior in non-nominal situations.
Conclusion
With AI generated simulations, truly random unpredictable events can be created while using realistic constraints to create a true-to-life experience of a difficult air traffic situation to stress the performance and metrics of a HITL experiment. These scenarios can also be recorded and used for fast time simulations to identify the most efficient outcomes to inform future policy or playbook coordination.
This new capability is made possible by synthetic data. This emerging technology enables new practices that reduce data cycle costs, enhances data privacy, and allows data scientists to generate and test scenario hypotheses accurately.
Would you like to learn more?
Learn more about synthetic data here. Do you have a specific need in aviation data analytics? Contact Us, we’ll be happy to help!
