Process Dynamics and Operations

Methods for process design and operations

Objective of this research area is the development and application of methods for the operation and design of processes. This entails the incorporation of model uncertainty in optimization tasks, the design of frameworks for parameter estimation, data reconciliation, dynamic real-time optimization as well as the decomposition and reformulation of models to improve their convergence behavior.

Research Approach

  • Optimal synthesis of chemical processes
  • Optimizing operation of process plants
  • Online application of optimal design of experiments and parameter estimation
  • Consideration of uncertainties in optimization problems, for example by probability restriction

Contact person

Room KWT-A 108
Office hoursThursdays, 8-10 am, online - link upon request
Room KWT-A 110

Optimal process design

Process synthesis and process design

In process synthesis tasks, there are often several process alternatives that have to be decided between. This can quickly result in countless alternatives that can no longer be investigated manually. In mixed-integer optimization, various modeling and solution approaches are available to solve such problems efficiently.

Design with consideration of dynamics

Processes are increasingly subject to dynamic disturbances. For this reason, the dynamics should already be taken into account in the plant design.

Process optimization

Dynamic operation optimization and optimizing control

The start-up and shut-down of plants as well as operating point changes are often still controlled by manually created recipes. Energy and resources can be saved during operation by determining optimal operation trajectories. The continuous reoptimization of plants is also an important current field of research.

Interaction with plants

Modern methods of optimizing control still fail due to the complexity of real processes. Our focus here is, among others, on the communication of process control technology and optimization tools, on the formulation of robust solution algorithms as well as on the extension of existing methods for optimizing control, state estimation and much more.

Design of experiments and parameter estimation

The determination of model parameters is a frequently recurring task for all new reactions, separation operations and systems. Here, the performance of the real experiment is often the biggest hurdle. Using methods of experimental design, model-based optimal experiments can be planned, carried out and parameter values can be estimated in parallel to the experiment.

Optimization using dynamic surrogate models

Optimization methods should often be solved in real time on real plants. Due to model complexity, this is often not possible directly. With the help of surrogate models of plants, for example, this can be implemented in practice. These models have to be trained before application in order to be able to cover all operating ranges in question. The efficient design and use of such surrogate models for optimization are open research questions.

Consideration of uncertainties

Quantification of uncertainties

Models are always only an abstraction of reality. Accordingly, all models are always subject to errors. The quantification of such errors in the form of probability density functions of model parameters or specific error models for entire models are essential for the evaluation of optimization results but also for the robust optimization of systems.

Probability constraints

In reality, many constraints are soft and do not always have to be met with absolute certainty. With the help of probability constraints, such soft constraints can be integrated and solved in optimization problems in a mathematically quantifiable way.

Numerical solution methods

Initialization of nonlinear models

Solving nonlinear models in process engineering is often non-trivial and usually requires time-consuming, manual initialization of variables. We investigate advanced mathematical methods for interval arithmetic in order to make the initialization more problem-independent and to be able to do without manual initialization as far as possible.