Artificial Intelligence (AI) offers great potential for future aviation systems, yet its integration into safety‑critical applications must comply with the aviation sector’s stringent safety standards. The European Union Aviation Safety Agency (EASA) requires that the Operational Design Domain (ODD) and the data distributions used during development and verification be shown to be representative and complete. A systematic engineering process for defining target distributions and assessing their representativeness within an ODD, however, is still largely missing. This paper presents a method for assessing the representativeness of AI/ML constituent ODDs in the context of aviation safety assurance. Starting with a methodical identification of suitable target distributions, a process flow is proposed that guides developers from ODD definition and parameter distribution modelling to quantitative coverage assessment and interpretation, aligning with EASA’s learning‑assurance objectives. As quantitative measures, the $\\chi^2$ goodness‑of‑fit test is examined and found unsuitable for the large data sets typical of this domain, leading to the adoption of the Kullback‑Leibler divergence $D_{\\text{KL}}$ and Cramér’s $V$ for representativeness assessment. The method is demonstrated on AI‑based airborne collision‑avoidance systems using experimental data from previous Horizontal Collision Avoidance System (HCAS) and Vertical Collision Avoidance System (VCAS) simulations. The results illustrate how statistical distribution comparison can support representativeness assessment for safety‑critical AI applications and contribute to a systematic Safety‑by‑Design AI engineering process consistent with emerging EASA guidance.
Blogger's Review: This work provides a practical quantitative approach for verifying AI‑driven collision‑avoidance systems, highlighting the advantage of KL divergence over traditional chi‑square tests in large‑scale data scenarios.