Runway Sign Classiﬁer - DAL C Certiﬁable Machine Learning System - Case Study

Sep 11, 2026

Uploaded by Jake

Machine learning advances are enabling new automation possibilities for aviation, yet ML-based airborne systems face incompatibilities with existing certification standards, including traceability, explainability, and inadequate coverage metrics. Certification under current standards is therefore problematic. This paper presents a case study of an airport runway sign detection and classification system using deep neural networks, extending prior work from Design Assurance Level “D” to the stricter Design Assurance Level “C”.

LanguageEnglish

CategoryPresentations

FormatPDF

Pages8

Reading Time3 min

AccessFree to download

Abstract

Introduction and Related Work

Certification challenges for ML-based airborne systems

Emerging efforts and guidance for ML certification

Proposed DAL C extension and architectural mitigation

Runway Sign Classiﬁer: A DAL C Certiﬁable Machine Learning System Konstantin Dmitriev Technical University of Munich Garching, Germany [konstantin.dmitriev@tum.de](konstantin.dmitriev@tum.de) Johann Schumann KBR, NASA Ames Research Center Moffett Field, CA [johann.m.schumann@nasa.gov](johann.m.schumann@nasa.gov) Islam Bostanov Technical University of Munich Garching, Germany [islam.bostanov@tum.de](islam.bostanov@tum.de) Mostafa Abdelhamid Technical University of Munich Garching, Germany mostafa.abdelhamid@tum.de Florian Holzapfel Technical University of Munich Garching, Germany ﬂ[orian.holzapfel@tum.de](orian.holzapfel@tum.de) Abstract—In recent years, the remarkable progress of Machine Learning (ML) technologies within the domain of Artiﬁcial Intelligence (AI) systems has presented unprecedented opportunities for the aviation industry, paving the way for further advancements in automation, including the potential for single pilot or fully autonomous operation of large commercial airplanes. However, ML technology faces major incompatibilities with existing airborne certiﬁcation standards, such as ML model traceability and explainability issues or the inadequacy of traditional coverage metrics. Certiﬁcation of ML-based airborne systems using current standards is problematic due to these challenges. This paper presents a case study of an airborne system utilizing a Deep Neural Network (DNN) for airport sign detection and classiﬁcation. Building upon our previous work, which demonstrates compliance with Design Assurance Level (DAL)”D”, we upgrade the system to meet the more stringent requirements of Design Assurance Level ”C”. To achieve DAL C, we employ an established architectural mitigation technique involving two redundant and dissimilar Deep Neural Networks. The application of novel ML-speciﬁc data management techniques further enhances this approach. This work is intended to illustrate how the certiﬁcation challenges of ML-based systems can be addressed for medium criticality airborne applications. I. INTRODUCTION AND RELATED WORK The remarkable progress of Machine Learning (ML) technologies in recent years has the potential to revolutionize aviation [1], [2] . Data-driven ML systems can implement highly complex cognitive functions such as vision and language processing that can enable single pilot or fully autonomous operation of large commercial airplanes, a level of automation not possible with traditional rule-based software systems [3] . However, ML technology encounters various inherent incompatibilities with existing airborne certiﬁcation standards. These incompatibilities encompass challenges related to explainability, traceability, and implementation coverage [4] . As a result, the utilization of ML-based applications within the framework of existing airborne certiﬁcation standards is currently impeded. The industry, aviation authorities, and academia are actively engaged in collaborative efforts to develop new certiﬁcation standards aimed at addressing the existing incompatibilities of ML technology with existing certiﬁcation practices. The development of a new standard for airborne machine learning (ML) certiﬁcation has been underway since 2019 through the collaborative efforts of the EUROCAE/SAE WG-1141/G-342 joint working group. The working group has published reports reﬂecting the intermediate results [4], [5] . The European Aviation Safety Agency (EASA) has released the guidance for ML applications [8], which includes anticipated objectivesand means of compliance for the certiﬁcation of safety-critical airborne systems based on machine learning. In collaboration with Daedalean AG, the U.S. Federal Aviation Administration (FAA) has released a research report on a neural network vision-based landing guidance system [6] . This report offers a practical assessment of the previously proposed W-shaped process for ML applications, as outlined in earlier Daedalean AG publications [1], [

FAQs

Why is certifying ML-based airborne systems difficult under existing standards?

Because ML models raise major issues such as traceability and explainability, and traditional coverage metrics do not adequately fit ML implementations.

What system does the case study focus on?

An airborne system for airport runway sign detection and classification using a deep neural network computer-vision approach.

How is the work extended from DAL “D” to DAL “C”?

The paper upgrades the prior system to meet DAL C by applying an architectural mitigation technique using two redundant and dissimilar deep neural networks, supported by ML-specific data management techniques.

Sep 11, 2026

Uploaded by Jake