Applying Model-Based System Engineering and DEVOPS on the implementation of an AI-based collision avoidance system

Abstract

The implementation of Artificial Intelligence (AI) in the field of aviation promises significant improvements from assistance functions to a higher level of autonomy. The new Airborne Collision Avoidance System (ACAS X) is one example. However, as with any other technology in aviation, the implementation of artificial intelligence also needs to be strictly regulated to ensure the highest possible level of safety and security. The European Union Aviation Safety Agency (EASA) set the first regulatory framework, suggesting a W-model, which is placed on top of the well-known V-model, covering the dedicated requirements for the AI-system throughout the engineering process. Yet, this approach does not coincide fully with the established DevOps cycle from software engineering, which is increasingly being used for the development of AI-based systems. The early combination of development and operation stands in contrast to the linear approach of the W-model. To investigate AI-Engineering methodologies and the potential harmonization of both approaches, the example use case of collision avoidance is chosen. In this work, the AI-based ACAS XA and ACAS XU are implemented into a simulation environment and tested in scenarios avoiding multiple intruders. One horizontal and one vertical multi-collision avoidance scenario is executed demonstrating the implementation of the system-under-test as well as the successful collision avoidance. Throughout the engineering process from defining requirements down to the sub-system level, Model-Based System Engineering (MBSE) is used as an efficient way of handling the ever-increasing complexity of the engineering process and improving communication and understanding for all involved stakeholders.