Industrial systems often consist of intricate networks of machines that must operate in harmony. These networks typically evolve over time, incorporating both legacy and modern equipment. Creating a digital twin of such systems for optimal control is highly resource-intensive and, in practice, often infeasible due to numerous variants, aging components, and production inconsistencies. This challenge is particularly pronounced for older machines with limited sensor data. Atlas Copco encounters similar difficulties with its central controller—a device designed to optimize compressor and air utility networks to minimize energy consumption. In this talk, you get insight into how these challenges can be tackled by using a combination of approaches going from industrial control approaches and optimization to machine learning.