Newsletter April 2022

Parametrizing the controllers

Using the calculated system properties such as PTx behavior and the dead time, a workflow task determines the parameters for PI controllers using various methods and validates them on the model. All results of the workflow are summarized in a PowerPoint report.

Further innovations in MoBA Automation Version 3.0

In addition to control oriented system analysis, MoBA Automation Version 3.0 also offers other add-on libraries for further use cases, such as model reduction or machine learning, model regression testing, model validation, optimization, and measurement data evaluation. In addition, the new release contains numerous ergonomic improvements, such as an accelerated start of the task or better assistance when parameterizing complex workflows. Licensees will soon be receiving detailed information about the new version. A list of all new features of version 3.0 can also be found in the release notes.

Customized solutions for process automation

In addition to the standard tasks and the add-on libraries for special applications, we also offer our customers customized solutions on request, which can be used for individual automation tasks. Please contact us if you need support for the implementation of your workflows.

Your Feedback

We thank you for your valuable feedback, which has contributed significantly to the further development of MoBA. Of course, we will continue to work on improving the ergonomics of MoBA Automation with great enthusiasm. If you have any suggestions, please do not hesitate to contact us.

Contact Control Oriented System Analysis: Dr.-Ing. Andreas Varchmin

Contact MoBA Automation: Dr.-Ing. Philipp Ebeling

As part of the funded VEOTOP R&D project, which was completed in autumn 2021, TLK, together with Technische Universität Braunschweig, developed software tools for topology optimization of complex thermal energy systems.

The Topology of Thermal Energy Systems

Thermal energy systems must perform an ever-increasing number of thermal processes: cooling, heating, deep-freezing, air-conditioning, dehumidifying, etc. - and all under changing operating conditions! For the topology of the energy systems, i.e., the selection and combination of the technical components, there are an unmanageable number of options. To evaluate these topologies, concepts for operating strategy and control are also necessary.

VEOTOP

To address this need, TLK developed methods and software tools in the VEOTOP project, together with the Technische Universität Braunschweig (Labs for thermodynamics, control engineering and mathematical optimization), to design and optimize topologies and to evaluate them including operating and control strategies.

The newly created methods and tools make it possible to generate several topologies of a thermal system based on user specifications and to optimally design these topologies taking dynamic boundary conditions into account. The developed systems can then be automatically simulated and visualized.

We would like to thank the Federal Ministry of Education and Research for funding and supporting the project.

The use of neural networks to create computationally efficient surrogate models can produce high quality approximations to physics-based models. Through their use in the control of a CO2 refrigeration cycle, TLK was able to achieve COP-optimized operation as well as an improvement in system behavior during load changes.

The energy-efficient operation of thermal management systems through advanced control concepts is an important element of the vehicle of the future. As part of the PHyMoS (Proper Hybrid Models for Smarter Vehicles) research project, TLK is investigating the use of machine learning models for model-based control of thermal management systems for passenger cars. Compared to physical-mechanistic models, these models can be characterised by a significantly lower computational load for model evaluation, which is why their use is particularly advantageous when there is little available computing capacity.

TLK has investigated the use of neural networks to determine optimal process parameters and for model-based feedforward control of a CO2 refrigeration cycle.

Determination of optimal process parameters

The determination of optimal process parameters is a significant challenge for complex refrigeration cycles. Due to their high complexity, detailed physical-mechanistic models of refrigeration cycles are only suitable for online optimization to a very limited extent. Thus, optimization calculations are usually carried out separately and the results are stored - e.g., in the form of a characteristic diagram - for further use. The use of machine learning methods, on the other hand, enables online optimization calculations to be carried out on the basis of a reduced, approximate model of the refrigeration cycle. For this purpose, a neural network is trained to map the correlations between the process boundary conditions, the process parameter to be determined and the variable(s) that are relevant for the optimization task. This method, shown in Figure 1, was used to determine the optimal high pressure of a switchable CO2 refrigeration cycle with parallel evaporators. With the help of the neural network, the maximum possible COP (Coefficient of Performance) was achieved within a mean relative deviation of approximately 0.4 % compared to the simulatively determined maximum COP (tested for 80 random boundary conditions).

Model-based feed forward control

A significant improvement of the control behaviour can be achieved by a model-based determination of the controllers’ actuating variables. Typically, simplifying model assumptions are made when creating surrogate models required for such a model-based feedforward control, which means that nonlinearities in the correlation between actuating variables and boundary conditions are unlikely to be represented adequately.

For this reason, TLK has tested the suitability of neural networks as approximate surrogate models for use in model-based feedforward control. A neural network was used for the feedforward control of the compressor speed of a CO2 thermal management system. This resulted in a significant improvement of the control behaviour compared to a simulatively designed PI controller. As a result, the maximum control error of the supply air temperature was reduced by 50 % when the air mass flow rate in the air conditioning unit was changed.

Provision of neural networks

TLK can provide the generated neural networks either as code in the programming language C or as a DLL. The neural networks can thus be integrated into both Modelica and Matlab/Simulink models or compiled and executed on any system.

Future research on machine learning methods

TLK is constantly researching and developing new ways of using various machine learning methods to solve control engineering problems. In particular, the determination of trajectories to improve control behaviour and the use for icing prevention are to be investigated more intensively in the future.

Contact Machine-Learning: M.Sc. Henrik Schatz
Contact Control: Dr.-Ing. Andreas Varchmin

For efficient operation and to promote long lifetimes of PEM fuel cells, a good water balance is crucial. To optimize this, membrane humidifiers are used in PEM fuel cell systems. TLK is investigating such membrane humidifiers on a complex test rig.

The thermodynamic losses of a fuel cell are significantly influenced by moisture-dependent proton conductivity. Water content also affects fuel cell degradation. To minimize thermodynamic losses as well as slow the degradation process, so-called membrane humidifiers are used.

In membrane humidifiers, water contained in the air is driven from the wet exhaust air to the dry supply air of the fuel cell by the moisture concentration gradient. The humidifier membrane used for this purpose is highly permeable to water, but blocks the transport of other gases.

In TLK's laboratory, the investigation of heat and mass transfer in membrane humidifiers under various boundary conditions - e.g., air mass flow rates, relative humidities, temperatures and pressures - is possible. The greatest challenge is to precisely adjust the mass flows and at the same time accurately measure the power. Even low quantities of only a few watts are relevant. And when it comes to balancing, every drop of water, no matter how small, counts.

In addition to individual investigations for our customers, TLK also carries out its own measurements as part of the publicly funded AUTO-GEN project. AUTO-GEN stands for "Partially automated generation of physics-based and data-based thermodynamic models for the design and operation of complex systems in e-mobility". In AUTO-GEN, among other things, humidifier models for fuel cell systems in commercial vehicles are created in collaboration with the Technische Universität Braunschweig (Markus Pollak, m.pollak@tu-braunschweig.de). The models are then validated with measurement data obtained from the test bench.

These humidification experimental investigations are just one example of the interaction between experiment and simulation at TLK. A wide variety of experimental investigations are carried out in labspace of around 1000m², both for our customers and for the optimization of our model libraries.

Contact: Dr.-Ing. Nicholas Lemke

Add-On Adsorption

The new add-on Adsorption enables the simulation of gas purification and gas drying plants. The library contains a universal basic model for gas mixtures and adsorption equilibria. With the help of the modular models, complex adsorption plants and processes - such as PSA plants (Pressure Swing Adsorption), TSA plants (Temperature Swing Adsorption) or Direct Air Capture processes - can be modelled. These models are designed to analyze breakthrough curves and adsorption kinetics and to draw conclusions for optimization (e.g., of cycle times).

The substance database of available adsorbents can be easily expanded.

Add-On Hydrogen-Energy-Systems

Hydrogen-based energy systems can be modelled with the already existing add-on Hydrogen-Energy-Systems (referred to as "TIL H2"). What is new is that the PEM fuel cell model can be discretized along the gas channels. Models for describing the water balance have also been added. The system example with anode and cathode supply has been extended to include humidifier bypass control. A jet pump model is now available for recirculation on the anode side. More types of stack models have been added. A reversible SOC (solid oxide cell) for operation as SOFC (fuel cell) or SOEC (electrolyser) can be flexibly parameterized and simulated for both pure hydrogen operation and operation with methane. Examples for the operation of a SOFC with methane steam reformer and a SOEC with reformer for methanation were created.

As an additional example, hydrogen refuelling using a tube trailer is modelled. An adaptation of individual boundary conditions for the subsequent simulation is possible via various settings.

Add-On Automotive

Thermal management systems in vehicles can be modelled with the already existing add-on Automotive. One of the innovations is that the refrigerant R-1234yf is used instead of R-134a in the examples for refrigeration circuits in automotive applications, as this corresponds to current practice. In addition to reciprocating compressors, scroll compressors are also used in the refrigeration circuits. The detailed heat exchanger with MPET geometry "Detailed MPET", which was previously only available for VLEFluid/Moist-Air, is now also available for Liquid/Moist-Air. It includes all thermal masses for headers as well as several airflow layers.

Add-On HeatStorage

Thermal energy storage can be modelled with the already existing add-on HeatStorage. One of the innovations is that an internal heat exchanger can be optionally activated for the hot water storage model with temperature stratification. The pipe coil through which an incompressible fluid flows can be individually parameterized with geometry and heat transfer data and connected to the outside via liquid ports.

TIL Basic and TILMedia

In the new release of TIL Suite 3.12.0, two examples for vehicle air conditioning, among others, are adapted to the refrigerant R-1234yf, according to the innovation in the add-on Automotive. A pressure loss model with nominal point for mass flow and density is now available for all refrigerant heat exchangers. With the help of the parameters for nominal values and exponents of mass flow and density, fitting the model to measured data is usually very easy.

With the new release of TILMedia 3.12.0 methanol as a VLEFluid and updated TILMediaSpline media are available. Further fast-calculating TILMediaSplines are available on request.

Contact NTU: Dr.-Ing. Nicolas Fidorra

Contact Hydrogen: Dr. rer. nat. André Thüring

Contact TIL Suite: Dipl.-Ing. Ingo Frohböse

We cordially invite you to our second Modelica and FMI Users' Meeting "ThermoSim 2022 - Modeling and Simulation of Thermal Systems" on October 4 and 5, 2022 in Braunschweig, Germany. Please submit your abstract for a presentation or poster by May 31, 2022.

We offer our proven training courses for Modelica and TIL both online and in person. We also offer individual training courses on vehicle air conditioning and our products upon request.

Together with our partner TLK-Energy, we will again offer training courses for Modelica and TIL from April 2022. Online and face-to-face events are planned. The exact dates and additional information can be found here.

In addition, we would like to point out our individual course offering on the subject of vehicle air conditioning. In a one or two-day training course, in addition to the basics of thermodynamics, heat transfer and climate physiology, the special requirements in passenger compartments and various refrigerants are discussed. The different air-conditioning requirements of conventional and electrified passenger cars and other means of transport are also addressed. It is possible to adapt the course to your individual requirements and questions. For those interested in this topic, we would also like to draw your attention to the Vehicle Air Conditioning Conference (HdT and TAE) on 27 and 28 April 2022 at the Haus der Technik in Essen.

We offer further training options for our products in the areas of simulation technology, visualisation, optimisation, and automation on request. If you have any suggestions or questions, please do not hesitate to contact us.

Contact:

Dr.-Ing. Wilhelm Tegethoff

Dr.-Ing. Sven Försterling