AI-driven Configuration for Filling Systems using Generative Adversarial Networks

Configuring filling systems in large-scale production, particularly when handling granular materials such as cement and construction materials, is often a manual and resource-intensive task. This process typically involves iterative prototyping based on material properties like density, granularity, and abrasiveness. Enabled by the automation and digitization of modern filling systems we present a concept for a deep learning-based Configuration of such systems. The core of our concept is a conditional generative adversarial network (cGAN), which generates optimized machine parameters based on multi type material and bag properties. The model captures the complex, non-linear relationships between these inputs and machine performance and generates distributions of parameter settings. It also incorporates continual learning capabilities, allowing the model to adapt to changes in product properties. The concept forms the basis for future research activities in which we will realize an automated Configuration system for running production. We present a novel use-case that has not yet been explored in present literature.