AI-based production planning for small batch manufacturing in the aerospace industry using reinforcement learning

Digital transformation is reshaping the entire industry in many different aspects. A major goal of this transformation is the realization of efficient production even with small batch sizes. A major challenge here is that classical economies of scale cannot be realized to meet the highest quality demands with simultaneously small batch sizes. The automation and digitalized planning of process with common methods from line production as they are often used in the automotive industry are not profitable.

As a result, the organization, planning and control of manufacturing processes and individual part production is often paper-based and relies on manual processes and procedures. Information and key figures are manually recorded in production and used in regular meetings to control production. Proactive, future-oriented actions are only possible to a limited extent. A major challenge in controlling production processes is to achieve the trade-off between component quality, productivity and reliability. Especially for the reliable estimation of delivery dates for specific orders, the predictive analysis of the data and information generated in the individual steps of the process chain is necessary. Machine learning methods and their use for the predictive analysis of production data are very promising for this purpose.

The aim of the project is to develop an automated and adaptive production control and production planning system for the production of small batches that contributes to a demonstrable and sustainable reduction of control and planning efforts in production. State-of-the-art technologies from the field of Deep Reinforcement Learning are used for this purpose, in which agents learn independently to develop the best approach for a defined planning goal.