Conference Paper · 2026
On the Applicability of Safety Nets: A Safety-By-Design Solution for Certifying Neural Networks
35th Congress of the International Council of the Aeronautical Sciences (ICAS)
The Short Version
A systematic study of Safety Nets explores the trade-off between neural-network size and corrective lookup tables for HCAS and VCAS, reducing storage while preserving correct outputs.
Abstract
The integration of Artificial Intelligence (AI) in safety-critical aviation systems presents significant challenges for certification and deployment. Aviation, often regarded as the safest form of transportation, relies on numerous safety-critical systems, from flight control to collision avoidance. For future safety-critical AI-based systems, EASA requires a Safety-by-Design approach, which can be achieved by using Safety Nets that combine neural network compression with lookup tables to ensure 100 % correct runtime behavior. Although Safety Nets have been studied, no comprehensive study of their performance characteristics and system design trade-offs has been conducted, leaving critical questions unanswered. This work presents the first systematic analysis of the trade-off between neural network and lookup table size in Safety Nets for next-generation collision-avoidance systems. By systematically comparing neural networks with diverse architectures—including varying activation functions (ReLU, LeakyReLU, GELU), hidden-layer configurations, and encoding strategies—this study identifies optimal design parameters that minimize overall storage and memory requirements while maintaining certification compliance. Results demonstrate that architectures with 3 to 5 hidden layers, each with approximately 50 to 100 nodes, combined with one-hot encoding, achieve the best balance. In these configurations, neural networks accurately represent at least 97 % of the data, while compact lookup tables handle the remaining errors. The resulting Safety Nets reduce the system size by one to two orders of magnitude, making deployment on current avionics hardware feasible while guaranteeing100 % correct outputs as required by EASA guidelines. This work provides the first-ever open-source implementation of Safety Nets for HCAS and VCAS with replicable results, demonstrating a practical pathway toward certifiable AI-based systems in aviation and establishing Safety Nets as a viable Safety-by-Design solution for safety-critical applications.
Reference
Cite This Work
Johann Maximilian Christensen, Thomas Stefani, Elena Hoemann, Frank Köster, Sven Hallerbach. "On the Applicability of Safety Nets: A Safety-By-Design Solution for Certifying Neural Networks". 35th Congress of the International Council of the Aeronautical Sciences (ICAS), 2026.