The central challenge in optimizing production paths in an SFG is the high complexity and dynamics of production. Different MMUs with specific capabilities must be flexibly available for different production orders. As requirements and system states change frequently, the production paths must be adapted at runtime to minimize failures and delays. In addition, meeting variable production targets - such as energy efficiency, production speed and on-time delivery - is a challenge.

Another difficulty is the precise capability assessment of each MMU depending on factors such as production duration, transportation times and resource availability. The use of digital twins enables real-time monitoring and adjustment of production, but the integration of different data sources and the handling of complex production scenarios remains challenging. Adapting paths to short-term changes such as machine or transport failures also requires stable and flexible control methods.

Our approach investigates the use of genetic algorithms and reinforcement learning to efficiently identify and control optimal production paths in a smart factory grid. In this scheduling problem, both machining and transportation processes are considered. Each Micro-Manufacturing Unit (MMU) and its system state is considered as part of this model, so that the production steps and transports are mapped in the overall system.

For the capability assessment, we use digital twins that collect and analyze relevant real-time and historical data on the MMUs. A combination of simulation-based and data-driven methods is used to develop adaptive control of the matrix production systems, which can also adapt to complex boundary conditions such as variable transport times, deadlines and failures. The implementation of advanced optimization methods ensures that the paths flexibly meet the production requirements.