To accomplish the EU's climate objectives by 2050, an increase in energy efficiency and ubiquitous utilization of renewable energy sources are imperative. At DLR Institute of Engineering Thermodynamics, gas-solid reactions are studied for various applications in energy technology ranging from thermal energy storage to air conditioning systems. Reactor design for such kind of processes is crucial but challenging due to interdependency of transport phenomena and reaction kinetics. To design reactors optimally in terms of weight and power, 2D models have been used so far, in addition to analytically simplified model equations. 

With the advancing of computational methods, it became very attractive to define new pathways for the optimization of reactor design parameters under different operational constraints, potentially making such effort computationally inexpensive. However, it is important to evaluate how well such optimization algorithms work for this class of problems. Solution of this task will undoubtedly contribute to better understanding of possible optimization methodologies and, consequently, make future energy applications more optimal and attractive for their users. 

The present thesis aims at the implementation of machine learning algorithms to perform the optimization task for a gas-solid reactor to be designed for energy applications. An existing 2D model will be used to create an extensive data set that will be used to train an algorithm. For the evaluation of the performance a comparison to an existing simplified 0D model can be used.

Your tasks will be: 

• understanding and using an existing model in COMSOL for 2D reactor design based on fundamental physical equations 
• production of data sets based on the 2D software to train a machine learning algorithm 
• applying the algorithm to find the optimal configuration 
• evaluation of machine learning algorithm by comparison with an analytically simplified model 
• preparation of final thesis report