📘 Resource Overview & Documentation Details

You are viewing the technical directory listing for Rolling Mill Digital Twin: From Sensors to Machine Learning. This engineering resource is curated under our Cyber Physical Production System division to support automation specialists, control systems engineers, and manufacturing technicians looking for verified reference materials.

🔍 Contextual Abstract & Technical Scope

Share Summary:LinkedInEmail

A rolling mill digital twin can help engineers connect machine measurements with process knowledge and data-based recommendations. Transforming a stand-alone mill into a cyber physical production system (CPPS) involves more than adding sensors: measurements must be calibrated, collected, organized and made useful to operators and analysts. A project at the Smart Forming Lab at MontanuniversitÀt Leoben demonstrates this progression, from sensor retrofitting to a machine-learning tool for rolling schedule recommendations.

From stand-alone equipment to a connected system

The project began with an existing rolling mill aggregate and developed it into an integrated digital environment. The physical machine remained central, but its measurements were brought into a wider architecture that included data acquisition, processing, storage and user-facing visualization. The resulting system was organized as a six-layer digitalization architecture hosted on an internal university server.

This structure matters because sensor readings alone do not create a usable digital representation of a process. Signals need to be converted into physical quantities and connected with the process context in which they were captured. Once integrated, the data can support both real-time awareness of the rolling operation and later analysis of process behavior.

Rolling mill sensor retrofitting and measurement

The mill’s measurement setup was upgraded by replacing existing load cells and adding inductive and magnetic displacement sensors. The sensors were calibrated and integrated with the system’s data acquisition and control hardware. This allowed machine conditions and process variables to be represented digitally rather than remaining available only through the original stand-alone setup.

For industrial teams considering a similar retrofit, the sequence is important: identify the quantities needed for process understanding, select suitable measurement devices, calibrate them, and establish a reliable path from signal to usable data. Integration with the broader architecture then makes it possible to present those measurements through an operator-facing interface.

One of the two front-end interfaces was designed for condition monitoring during rolling. Its purpose is to make relevant machine information visible during operation. This provides a practical bridge between the physical equipment and the digital system, helping users observe the process through a consistent interface.

Using machine learning to support rolling schedules

The second interface presents results from a Python-based machine-learning algorithm. Developed using data from more than 1,900 rolling steps, the model considers variations in roll-gap height, sheet width and friction conditions. It can interpolate and extrapolate across these parameters and different initial sheet thicknesses, supporting data-based recommendations for changes between rolling process steps.

Rather than replacing process judgment, the algorithm serves as a decision-support tool. It predicts and adapts a rolling schedule for a defined metal sheet, while the interface helps users see how parameter changes relate to the predicted outcome. The underlying approach is described as a black-box model: it learns relationships from the available rolling data rather than relying on a fully specified physical model of every process interaction.

An important feature is the ability to incorporate additional rolling schedules as they are carried out. This means the model can learn from new process data, extending the basis for recommendations. The digital twin is therefore tied to both the physical rolling process and an evolving dataset, rather than being only a static visualization of machine geometry.

Connecting process data with material behavior

Rolling outcomes are not only about machine settings. The project also examined rolled specimens to investigate how the plastic strain path affected mechanical properties, including material strength and anisotropy. Tensile testing and microstructure observations provided ways to assess changes associated with the rolling process and specimen orientation.

This materials-focused work points toward a further development of the digital environment: incorporating materials-science data to help predict how different materials may respond to rolling. The initial machine-learning tool focuses on process variables and rolling results; linking it with material information could broaden the questions the system is able to address. The project presents this as a basis for further integration, not as a completed capability.

What the architecture enables

Combining measurement, interfaces and analytics creates value across several activities:

  • Operational visibility: A condition-monitoring interface makes selected machine information accessible during rolling.
  • Process analysis: Collected data can be used to examine relationships between settings and rolling outcomes.
  • Schedule support: The algorithm offers data-based recommendations for changes between process steps.
  • Learning and training: The server-hosted visualization provides students and other interested users with an environment for exploring rolling behavior and machine-learning fundamentals.
  • Future development: The architecture provides a foundation for bringing material properties into data-based process predictions.

These functions depend on the complete data path, from calibrated sensors to processing and visualization. A model’s recommendations are only meaningful within the variables and data represented in its training and use. For that reason, integrating measurement and process context is as important as selecting the machine-learning method.

Practical implications for industrial teams

The rolling mill transformation illustrates a staged route to digitalization: retrofit measurement where needed, connect signals to a structured architecture, provide interfaces for monitoring and interpretation, and then apply analytics to a defined process question. This approach can help teams make existing equipment more data-accessible without treating a new algorithm as a substitute for instrumentation or engineering understanding.

For plant managers, system integrators and automation engineers, the key takeaway is to plan the digital system as an integrated whole. Sensor calibration, data handling, operator access and model scope all shape whether analytics can support practical decisions. A rolling mill digital twin built on those foundations can connect process observations with schedule recommendations, while leaving room to incorporate richer materials data as the digital environment develops.

Access Official Technical PDF