Companies often need more advanced automation to increase system performance or achieve a higher robustness. In CoMoDO we will help them build controllers relying on estimates of models updated during operation, which, to date, is more efficient than upcoming data-driven and AI methods. Having models or model-based estimators can help the controller in many ways: models and their estimates provide the controller insight in how to exploit the system behavior fully, how to compensate non-linearities or coordinate multiple DOFs. They can further predict outcomes of candidate control actions, such that the best action can be chosen. Or they can be used to estimate unmeasured variables, such as system loads, ambient conditions, or the behavior of third-party units, allowing the controller to compensate for those influences effectively.

Despite the possible benefits of control using models updated during operation, and despite plenty of existing SOTA, it is not widespread in the SOTP. This results from (i) companies not having the types of models needed for the SOTA and not knowing which option to pursue when making a new one, and (ii) the SOTA methods not being sufficiently application-oriented in dealing with typical, imperfect data, models and usage scenario’s.

In this project we aim to tackle these problems by developing tools to help companies realize control based on models updated during operation. We will build on previous projects to develop methods and decision logic to assist companies in selecting and setting up appropriate control models, estimators and control logic, without reverting to MPC, as it is a computationally expensive, complex method that is cumbersome to implement reliably and not widely adopted in SOTP. Since we target models being updated during operation, not only during engineering, we will also focus on reliability in practical situations, researching methods to handle industrial challenges like sensor noise, drift, different operating modes, varying conditions, machine degradation, and sensor failures. This should enable companies to start using models updated during operation in their control logic, while feeling confident about the reliability for both short and long-term usage.