To content
Department of Mechanical Engineering
Our research plan

Goals and Directions

Control systems are getting more and more complex and solving control tasks increasingly relies on interdisciplinary approaches combining elements from mechatronics and cybernetics. Against this background, we aim for tailored control schemes that address the special needs of such cyberphysical systems. In particular, we design networked and autonomous control systems involving numerical optimization, machine learning, and cybersecurity.

Secure control for networked systems

Cloud-computing and distributed computing are becoming ubiquitous in modern control systems such as smart grids, building automation, robot swarms, or intelligent transportation systems. While cloud-based and distributed control schemes offer many opportunities, they also increase the risk of cyberattacks. In fact, the involved communication and evaluation of sensitive data via public networks and on thirdparty platforms promote eavesdropping and manipulation of data. Future control schemes should counteract those threats and ensure confidentiality and integrity of the involved data. We address these needs with tailored control schemes that combine methods from cryptography, numerical optimization, and control.

Related projects

Resilience meets secure networked control (Rescuenet)

Applied encrypted control for critical infra­structure (Aperitif)

Encrypted optimization-based control for networked systems (EpicNets)

Illustration of a cloud-based controller with encrypted communication links © MSD ​/​ RCS
Illustration of an enrypted cloud-based controller © MSD ​/​ RCS
Encrypted control aims for encrypted data troughout the control-loop. This requires tailored controller reformulations (in contrast to encrypted communications as in the figure on top).

Explainability in Data-Driven Control

Data-driven predictive control (DPC) is a method that uses collected trajectory data to make predictions, offering an alternative to traditional model predictive control (MPC). Instead of relying on an explicit system model, DPC uses linear combinations of recorded data to predict future system behavior.

However, the absence of an explicit prediction model can make it difficult to understand the effects of commonly used regularization terms and the resulting predictions. This lack of transparency can lead to challenges, such as the need for empirical tuning of regularization parameters and potentially misleading interpretations of norm-based regularizations. Our research aims to address these issues by analyzing the structure of the underlying optimal control problem in DPC, enhancing the explainability and reliability of DPC.

Related publications

M. Klädtke and M. Schulze Darup, Implicit predictors in regularized data-driven predictive controlIEEE Control Systems Letters, vol. 7, pp. 2479-2484, 2023. DOI: 10.1007/LCSYS.2023.3285104, Preprint: arXiv:2307.10750

M. Klädtke and M. Schulze Darup, Towards explainable data-driven predictive control with regularizationsat - Automatisierungstechnik, 2025. Accepted January 17, 2025. Preprint: arXiv:2503.21952

A diagram illustrating two approaches to data-driven control design: 1) An indirect approach involving "Model Identification" from "Data" followed by "Model-Based Design" to "Control". 2) A "Direct Data-Driven Design" path bypassing explicit model identification. The concept of "Implicit Predictors" provides a model-based interpretation of predictions made by direct data-driven control schemes. © MKL​/​RCS
Implicit predictors provide an indirect (i.e., model-based) view on predictions made by direct data-driven control schemes

Tailored learning-based control

Details follow.

Related publications

N. Schlü­ter and M. Schulze Darup, Novel convex decomposition of piecewise affine functions, in Proc. of the IFAC World Congress, 2020.

M. Schulze Darup, Exact representation of piecewise affine functions via neural networks, in Proc. of the 2020 European Control Conference (ECC), pp. 1073-1078, 2020. DOI: 10.23919/ECC51009.2020.9143957

ConvexControllerDecomposition © NSC ​/​ RCS

Predictive control for geothermal systems

We pioneer advanced control and estimation techniques for complex cyber-physical systems with an emphasis on improving sustainability and efficiency of geothermal systems. For instance, our recent work on aquifer thermal energy storages (ATES) leverages sophisticated model predictive control with mixed-integer programming to achieve sustainable and economically viable operations. Furthermore, we design robust and computationally efficient state estimators to accurately determine critical states in hybrid systems where direct measurements are scarce. These developments showcase our dedication to advancing optimization-based control and hybrid systems analysis for impactful real-world solutions. Moving forward, we will apply and expand these powerful techniques to address emerging challenges in smart grids, autonomous robotics, and sustainable manufacturing, while also harnessing the potential of artificial intelligence to create more adaptive and intelligent systems. Our ultimate aim is to continuously advance the theory and practice of control systems for a more efficient and resilient future.

Related publications
J. van Randenborgh and M. Schulze Darup, MPC using mixed-integer programming for aquifer thermal energy storages, Proc. of the 8th IFAC Conference on Nonlinear Model Predictive Control (NMPC), vol. 58(18), 2024. DOI: doi.org/10.1016/j.ifacol.2024.09.004

Illustration of an Aquifer thermal energy storage system with building in heating mode © JVR ​/​ RCS
Aquifer thermal energy storage system with building in heating mode.

New approaches to model predictive control

Model predictive control (MPC) is one of the most powerful, widely-used, and well-studied control schemes. Its underlying idea can be traced back to the 60s. Since then, countless successful applications and theoretical improvements have been reported.  Nevertheless, the theory for MPC is far from complete and novel variants and implementations pop up regurlarly. We contribute structural insights on linear and nonlinear MPC, fast real-time implementations, and secure controller evaluations.

Related publications

M. Schulze Darup, M. Klädtke, and M. Mönnigmann. Exact solution to a special class of nonlinear MPC problems, in Proc. of the 7th IFAC Conference on Nonlinear Model Predictive Control (NMPC), pp. 294-299, 2021. DOI: 10.1016/j.ifacol.2021.08.559.

M. Schulze Darup, G. Book, and P. Giselsson, Towards real-time ADMM for linear MPC, in Proc. of the 2019 European Control Conference, pp. 4276-4282, 2019. DOI: 10.23919/ECC.2019.8796239

M. Schulze Darup and M. Cannon, Some observations on the activity of terminal constraints in linear MPC, in Proc. of the 2016 European Control Conference, pp. 770-775, 2016. DOI: 10.1109/ECC.2016.7810382

 

 

Sets associated with exact solution to bilinear MPC © MKL ​/​ RCS
Feasible set of a predictive controller for a bilinear system and underlying structure resulting from exact linearization.