The aviation industry with its many subsectors and systems cannot escape the digital transformation. Accordingly, the areas of application for machine learning (ML) and artificial intelligence (AI) occurring in aviation are exceedingly diverse. In addition to comparatively uncritical applications, such as in business analytics and logistics, there are extremely safety-critical fields of application in terms of reliability and security, such as in aircraft and their production. In the latter case, therefore, very demanding requirements exist regarding the usability and certification of ML and AI. In other areas of aviation, ML and AI methods and tools are already being tested and used more or less successfully today. The lecture series provides insights into a wide variety of research topics at aircraft manufacturers and their suppliers.
Data is key for machine learning training and validation. A high quality and representative dataset is required for high performance deep learning models. In the case of vision based avionic systems, for example for autonomous landing or obstacle avoidance, while it is possible to use real images captured with aircraft onboard camera, the amount of data will be insufficient and will not enable the coverage of all the combinations of conditions (dangerous corner cases, weather conditions, unexpected obstacles…). The use of additional synthetic data is required to cover all the possible scenarios. For that, we must simulate the environment (here a digital twin of the airfield) and model the sensor. Such a simulation also accelerates the validation of the ML algorithm through its use in virtual closed-loop tests. The lecture will present the main challenges of generating and using synthetic data for vision based avionic systems: photorealism, accuracy, scalability, multi-domain, labeling, segmentation, massive generation, assessment of the image quality, synth/real ratio in the dataset.
Using AI/ML is becoming more and more a reality in all industry sectors and our everyday life, but aviation and in particular the aircraft cabin has still many “white spots”. With the evolution of cabin systems and the standardized communication between them the aircraft cabin is developing into an intelligent system itself, which gives numerous opportunities for improvement of existing business services, creation of new business processes and the experience of a seamless transition between the outside world and the aircraft cabin. In this lecture an overview, from IFE perspective, shall be given what AI/ML and supporting hardware- and software platforms can contribute to achieve an “intelligent & smart” aircraft cabin.
If a customer wants to order an aircraft today, a comprehensive process is initiated in which the individual wishes are recorded at the beginning, the feasibility is checked in the next step, the customer version is specified, production is initiated, customer acceptance is carried out when the aircraft is handed over and customer feedback is obtained after delivery. The process flow contains several quality gates in combination with different measurement factors. If a measurement factor moves outside the specified boundaries, measures must be taken so that the process can be adhered to in terms of time and quality specifications. This lecture deals with the possibility of digitizing the customization process using a model-based approach and the ability of automatic measurement factors monitoring. Furthermore, it is discussed how necessary measures can be supported with by means of machine learning in order to ensure the efficient flow of the E2E process.