Sunday 16 March 2025
The development of machine learning (ML) has led to significant advancements in various fields, including aviation. However, the integration of ML-based systems into aircraft poses new challenges for certification and safety assurance. A recent paper proposes a semi-automated approach to certify low-criticality ML-enabled airborne applications.
The authors recognize that traditional software certification standards, such as DO-178C, are not tailored to address the unique characteristics of ML systems. These systems learn from data and adapt in response to new information, making it challenging to ensure their reliability and performance over time. The paper suggests a structured approach to certify ML-based systems by focusing on model consistency, data quality management, and resilience in varied operational contexts.
The authors propose an Assurance Profile for ML systems, which provides a comprehensive evaluation of the system’s certification readiness. This profile includes metrics such as dataset quality, model documentation, integration documentation, and usability assurance. The authors also highlight the importance of integrating human oversight into the certification process to ensure that complex decisions are made by experts.
The paper presents a case study on an object detection system designed for reconnaissance and surveillance aircraft. The system uses YOLOv8, a popular ML algorithm, to classify military and civilian vehicles in real-time. The authors demonstrate their semi-automated certification approach using this system as a test bed.
The results show that the ML-based object detection system meets the compliance criteria for DO-178C Level D criticality, which is suitable for applications with lower safety risks. The paper also highlights the need for ongoing monitoring and drift checks to ensure the system’s performance remains consistent over time.
The authors emphasize the importance of developing certification standards specifically tailored to ML systems. They argue that traditional software certification standards are not sufficient to address the unique challenges posed by ML systems. The proposed semi-automated approach provides a framework for certifying low-criticality ML-enabled airborne applications, which is essential for ensuring safety and reliability in these systems.
Overall, this paper presents a significant step forward in addressing the certification challenges associated with ML-based systems in aviation. The proposed approach offers a structured framework for evaluating the certification readiness of ML systems and highlights the importance of integrating human oversight into the certification process. As ML continues to play an increasingly important role in various fields, including aviation, it is essential to develop certification standards that are tailored to these systems’ unique characteristics.
Cite this article: “Semi-Automated Certification Approach for Machine Learning-Based Systems in Aviation”, The Science Archive, 2025.
Machine Learning, Aviation, Certification, Safety Assurance, Software Certification Standards, Do-178C, Object Detection System, Yolov8, Semi-Automated Approach, Assurance Profile.
