2022 IEEE/AIAA 41st Digital Avionics Systems Conference (DASC) · 2022 · 13 citations · 0 references
ComputingAerospace EngineeringComputer ScienceDesign AutomationFormal MethodsMachine LearningDeductive VerificationVerificationAutomated Machine LearningAutonomous OperationsCertification ComplianceCertification StandardsDesign AssuranceMitigationSoftware Certification
Exceptional advances of Machine Learning (ML) technologies in recent years have opened up opportunities for next level of automation in aviation systems, such as single pilot or fully autonomous operation of large commercial airplanes. But there are several essential incompatibilities of Machine Learning technology with existing airborne certification standards, such as traceability and coverage issues. These incompatibilities prevent approval of ML-based applications using current certification standards. In this paper, we study the combination of architectural mitigation technique with several ML-specific verification methods to achieve compliance with Design Assurance Level (DAL) C. This approach proposes incremental evolution of existing assurance practices and extends the custom ML workflow for DAL D systems presented in our previous works [1], [2].