Trusting Machine-Learning Applications in Aeronautics

Abstract

A general recommendation from the French office for aeronautical and space standardization (BNAE) is being drawn up by experts from Onera, Thales, Dassault and Safran, with the collaboration of Airbus, MBDA and ADP, the main French aeronautical companies. This document is based on mathematical and statistical elements which are reintroduced within a system and software development process considering the specificities of algorithms based on learning methods from data sets or data generators. For each activity in this development process, whether it is data capitalization or the use of artificial intelligence, risks are identified, and mitigation methods proposed. A few application cases are included in the document to illustrate the particularities of certain types of algorithms. Methods of estimation, classification, categorization or even reinforcement learning are mentioned. This paper gives a summary in English of the general recommendation.