AI-based assembly in the production of aircraft shells using reinforcement learning

Digitalization in combination with artificial intelligence methods promises the automation of complex manufacturing tasks, which are usually only economically feasible with high volumes using classical methods. The AGREED project aims to create cost- and quality-related competitive advantages through the development of (partially) automated and digitally supported processes in the field of aircraft shell assembly. In this context, the Institute for TMDT is conducting research in collaboration with Premium Aerotec GmbH in two use cases on the introduction of AI-supported control of assembly robots and the use of artificial intelligence methods to create adaptive processes.

The goal of the first use case is to fully automate the assembly of components using robots in the area of partial aircraft shells and to avoid time-consuming teach-in procedures. For this purpose, reinforcement learning methods are being researched and developed to train intelligent and self-learning agents that independently perform an automated programming to control gripping and joining processes based on acquired sensor data. To avoid a high training effort in complex real environments, transfer learning solutions are developed to efficiently and cost-effectively train the agents from simulated environments.

The second use case focuses on the automated creation of work plans for the collaboration of humans and robots in the final shell assembly. Specifically, after the complete digitalization of the previously paper-based work planning, reinforcement learning methods are developed for an automated analysis of process data. The goal is the realization of an intelligent, dynamic and adaptive planning and distribution of work steps. Here, emerging resource bottlenecks or machine failures are compensated by a targeted rescheduling and a resource-efficient final assembly with lower throughput times and higher volumes achieved by the dynamic distribution between employees and robots.