A Scalable and Modular Architecture for Intelligent Scheduling

Sustainability and digitalization fundamentally change how production is planned and controlled. One of the central parts of production control is job scheduling, which has the potential to include vast amounts of data to improve the quality of solutions. However, solving the scheduling problem is not limited to financial or operational aspects anymore; solutions must also include social and environmental aspects. While this issue is increasingly recognized among a plethora of other aspects in literature, many approaches still only consider small subproblems leading to a strong fragmentation. This study, therefore, develops a holistic framework for intelligent scheduling to integrate sustainability and social responsibility objectives while simultaneously offering a solution to bringing together the multitude of influencing factors in the domain of scheduling. The development of the concept is based on a structured literature review and an expert questionnaire. The Scalable Modular Architecture for Scheduling is a module-based software solution that connects to the real production system based on asset administration shells and graph-based problem modeling. The graph is transformed into a vectorized representation using a heterogenous graph neural network and combined with energy and maintenance-related information to be fed to a scheduling agent. This module utilizes a proximal policy optimization architecture to select the next best action based on mutliple criteria regarding economical and sustainable aspects and feed this information back to the real system, closing the loop. The framework is validated on benchmark data sets.