Our approach in a nutshell

In this article, we describe our unique approach to address challenges commonly found in polymer extrusion and other continuous production processes. Additionally, our platform and use cases are introduced while expected process improvements are shown at the end.

Challenges in polymer extrusion and beyond

Operating continuous production lines, for instance polymer extrusion processes for pipes, profiles and foils, is a difficult endeavour due to their complexity and numerous influential factors that lead to increased process variance, quality deviations and unplanned downtime. Plant operators are usually required to manually adjust several settings to steer the process, react to emerging patterns and issues to keep the resulting products within their tight tolerances.

Additionally, to remain competitive, continuous production processes must be operated with an increasing level of efficiency in terms of material and energy consumption. However, as shown in Figure 1, manual adjustments of extrusion lines are limited by the sheer complexity of dozens of set points, hundreds of process measurements and delayed quality checks in the lab. As a consequence, the following challenges are commonly found in polymer extrusion lines:

• Reactive and stressful work environment with an incomprehensible complexity
• Process variance and quality deviations leading to fragile processes, scrap and downtime
• Increased material and energy consumption leading to competitive disadvantages

Importantly, due to increasing demands in the context of the circular economy, the share of recycled materials used in extrusion processes is expected to rise leading to additional challenges caused by impurities in the provided input material.

Figure 1: General setup of extrusion lines with dozens of set points and hundreds of process and lab measurements.

To address these challenges, state-of-the-art approaches are usually driven by data, physics or process expertise but neglect the potential of hybrid approaches:

• Approaches purely driven by data allow to gain insights into the process via visualizations, anomaly detections and black-box models but ignore the underlying process structure, causal dependencies and physical knowledge leading to fragile systems prone to false positives.
• Approaches purely driven by physics are well suited for designing new components and plants as well as for offline simulation and optimization of extrusion lines. However, due to computational demands and simplifying assumptions, applying these solutions for online optimization and process control is usually impossible.
• Approaches purely driven by process expertise of operators with decades of experience are perfect for troubleshooting anomalies and finding creative solutions to improve extrusion lines systematically. However, these experimental trial-and-error approaches do not scale well and are unsuited for continuous process optimization.

Our approach: Data + Physics + Expertise

At jora.tech we introduce a new approach to model, predict and optimize the dynamic behaviour of entire extrusion lines by fusing data, physics and process expertise into a holistic method. Relying on known or identified causal dependencies and physics-informed state propagation (dashed lines in Figure 2) significantly increases robustness and explainability while addressing spurious correlations in the process.

For this purpose, we identify the main process stages, set points and states, build targeted state-models and integrate them into an end-to-end process model via state propagation (see Figure 2). Additionally, we explicitly include physical knowledge directly via analytical equations or via guardrails during training the AI models whenever possible.

The resulting hybrid AI models are capable of predicting the future behaviour of entire extrusion lines for up to two hours in advance allowing to mitigate upcoming issues proactively while addressing unexpected disturbances quickly based on our real-time control platform.

Figure 2: Structured modelling approach integrating data, physics and process expertise.

An important aspect that differentiates our approach from other solutions is that we always consider the process from an end-to-end perspective including everything from the material supply to quality measurements at the end of the line and beyond. As a consequence, our solutions enable a holistic optimization of all relevant factors simultaneously and continuously.

Enabling real-time optimization via the jora.tech platform

Based on our unique approach to model continuous production processes, the jora.tech platform enables real-time optimization with several core use cases ranging from human-in-the-loop (Figure 3) to fully automated predictive control (Figure 4):

• What-If Analysis Plant scenarios can be tested virtually by adjusting set points, providing safe insights for faster decision-making. e.g. “How does the wall thickness change, if I increase heating zone 2 by 5°C?”
• Set Point Recommendation Desired outputs are specified, and the system recommends the optimal set points required to achieve them. e.g. “How do I need to set all plant parameters to achieve a wall thickness of 5mm?”

Figure 3: Human-in-the-loop use cases focused on what-if analysis and targeted recommendations.

• Anomaly Detection Processes are continuously monitored for deviations to enable efficient troubleshooting and targeted maintenance. e.g. “Which component behaves abnormally and may cause an incident soon?”
• Fully automated process control Optimal set points are continuously calculated and applied automatically to keep processes within specification, ensuring quality and reducing resource use. This approach represents the highest level of innovation, freeing the operators to focus on creative and systematic process improvements.

Figure 4: Fully automated closed-loop control via the jora.tech platform and underlying process models.

As a result of these use cases, jora.tech changes the way how production is done by moving from reactive to predictive optimizations. To ensure its applicability, the underlying platform is modular by design allowing tight integration in any existing infrastructure acknowledging the heterogenous reality of modern OT and IT environments.

Grounded in science, proven in practice

The underlying methods for modelling and controlling continuous production processes, most notably extrusion lines, are the result of several years of cooperative research among scientific and industrial partners. As a consequence, comprehensive industrial validation with actual extrusion lines revealed the following benefits of applying these methods:

• Predict the process behaviour for up to 2 hours in advance
• Reduce process variance by 60 %
• Reduce quality deviations by 80 %
• Increase output and process stability
• Detect anomalies faster to mitigate downtime and damages

As a result, depending on the complexity of the extrusion line, significant savings are expected when data, physics and process expertise are combined for AI-driven process control. Importantly, while the example above focuses on polymer extrusion, the underlying methods for modelling and optimization are applicable to continuous production processes in general.

Want to know more about our approach? Get in touch and let’s discuss on how jora.tech can help elevate your production to the next level!

Key references

• C. Abeykoon, “Single screw extrusion control: A comprehensive review and directions for improvements,” Control Engineering Practice, vol. 51, pp. 69–80, June 2016, doi: 10.1016/j.conengprac.2016.03.008.
• C. Rauwendaal. Understanding Extrusion. 3rd ed. Hanser Publications, Nov. 2018.
• M. del Pilar Noriega E. and C. Rauwendaal. Troubleshooting the extrusion process. 3rd ed. Hanser Publications, Oct. 2019.
• V. García, J. S. Sánchez, L. A. Rodríguez-Picón, L. C. Méndez-González, and H. de J. Ochoa-Domínguez, “Using regression models for predicting the product quality in a tubing extrusion process,” J Intell Manuf, vol. 30, no. 6, pp. 2535–2544, Mar. 2018, doi: 10.1007/s10845-018-1418-7.
• J. Grimard, L. Dewasme, and A. V. Wouwer, “Dynamic Model Reduction and Predictive Control of Hot-Melt Extrusion Applied to Drug Manufacturing,” IEEE Trans. Contr. Syst. Technol., vol. 29, no. 6, pp. 2366–2378, Nov. 2021, doi: 10.1109/tcst.2020.3038028.
• R. Hartner and V. Mezhuyev, “Time Series Based Forecasting Methods in Production Systems: A Systematic Literature Review,” Int J Ind Eng Manag, vol. 13, no. 2, pp. 119–134, June 2022, doi: 10.24867/ijiem-2022-2-306.
• R. Hartner, M. Kozek, and S. Jakubek, “Multi-task learning with state propagation for quality forecasts in polymer extrusion lines,” J Intell Manuf, May 2025, doi: 10.1007/s10845-025-02616-2.
• R. Hartner, M. Kozek, and S. Jakubek, “End-to-End Process Optimization in Polymer Extrusion Lines Using Model Predictive Control and Multi-Task Learning,” IEEE Access, vol. 13, pp. 168344–168360, 2025, doi: 10.1109/access.2025.3614061.