Why does this project matter?

Advanced process control is a key enabler of process tolerance. Even with well-designed unit operations, maintaining consistent product quality under uncertain and varying conditions requires intelligent real-time control strategies. Model Predictive Control (MPC) is uniquely suited for this: it anticipates future process behavior, accounts for constraints, and optimizes control actions over a prediction horizon. Robust nonlinear MPC (NMPC) extends this to handle model uncertainties and disturbances but current implementations are computationally demanding and do not provide systematic insight into how much uncertainty a process can actually tolerate while still meeting its constraints. P8 develops a new generation of adaptive robust MPC methods to close this gap, demonstrated on real TALENT processes.

What are we aiming to achieve?

P8 pursues two complementary objectives: (1) develop adaptive robust MPC schemes that exploit real-time measurements to dynamically adjust to model uncertainty and process input variations (feedstock composition, fluctuating energy supply), ensuring optimal performance under changing conditions; and (2) develop a systematic methodology for quantifying process tolerance as the maximum admissible uncertainty in system inputs and process models that still allows safe and efficient operation. P9 provides an actionable design guidance across the TALENT process chain.

What will you work on as a PhD researcher?

As doctoral researcher in P8, you develop the adaptive robust MPC methodology demonstrated on the TALENT fermentation process (P2). You begin by building a hybrid process model combining first-principles mass and energy balances with data-driven components, implemented in Python using the do-mpc open-source framework developed by the Lucia group. The core methodological contribution is a principled uncertainty quantification approach using conformalized quantile regression: you train a quantile regression model on experimental process data, apply conformal prediction to obtain statistically valid prediction intervals, and develop methods to ensure physical consistency of the uncertainty bounds (satisfaction of mass and energy balances).

The quantified uncertainty and expected input variations are embedded in a scenario-tree-based robust NMPC formulation. You implement a real-time measurement update loop in which new sensor data continuously recalibrate the uncertainty estimates, then evaluate closed-loop performance in simulation before connecting to experimental setups in P2, P3 or P7 via control interfaces, which is giving you hands-on experience with real-time digital control of physical laboratory equipment.

Skills and methods you will develop during your doctorate:

• Mathematical process modeling: first-principles and data-driven hybrid models for fermentation

• Machine learning for uncertainty quantification: quantile regression, conformal prediction, neural networks

• Robust model predictive control: scenario-tree NMPC

• Scientific computing: do-mpc framework, numerical optimization

• Real-time control implementation: digital interfaces to lab equipment

• Closed-loop simulation and control performance evaluation

• Process systems engineering: connecting models, optimization, control, and experiments in an integrated workflow

Who will you work with and where?

The Lucia group is a leading center for advanced process control, with internationally recognized contributions to robust nonlinear MPC, uncertainty quantification, and deep learning-based control law approximation. The group has developed the do-mpc open-source software framework for nonlinear and robust MPC, which is actively used by research groups and industrial partners. Computational resources include high-performance workstations, a fully established Python-based control and optimization software stack, and interfaces for real-time digital control of laboratory equipment. The group is embedded in the strong data science environment at the TU Dortmund University, including connections with the Lamarr Institute for Machine Learning and Artificial Intelligence. Embedded in RTG TALENT, you gain access to a structured qualification program that combines advanced scientific training with transferable skills development, active exchange with academic and industrial collaboration partners, and tailored career support including the opportunity for a three-month placement in research, industry, or a start-up aligned with your career goals.