Analysis of Airworthiness Requirements for Application of Machine Learning in Airborne Systems

Machine Learning(ML) technology has the characteristics of a data-driven, inexplainable and complex model. On the one hand, computer vision, autonomous flight, preventive maintenance and other applications which are based on ML, bring huge opportunities for aviation development. On the other hand, the trustworthiness problem also brings unprecedented impact to aviation safety and an airworthiness certification, which seriously restricts its application in airborne systems. This paper establishes airworthiness certification framework and analyzes its coverage relationship with current airworthiness regulations. The specific challenges caused by ML application to airworthiness certification are reviewed, and the key airworthiness compliance technologies and methods are proposed by integrating the classical methods such as SAE ARP4754A and ARP4761 with the technical characteristics of ML, forming the certification elements so as to support industry's understanding of airworthiness requirements.