A Comparative Study of Component Surrogate Model Approximation for Holistic, Multi-Disciplinary Optimization of High Temperature Heat Pumps

Kurzfassung

Fast and reliable design methods are essential to reach the highest possible efficiency of high temperature heat pumps and enable their full potential. The heat pumps performance is strongly dependent on the efficiency of its components like turbomachines and heat exchangers. Conventional design is a sequential procedure starting with the cycle conceptualization. Component performance is initially based on assumptions and their design is optimized in subsequent steps with increasing level of detail. This sequential aspect makes it impossible to find the overall optimal heat pump configuration. To overcome this, holistic design strategies optimize the cycle parameters simultaneously with detailed geometric component design. This concept leads to significantly improved heat pump performance, saves time and reduces uncertainty. Multi-disciplinary optimizations are complex problems with a high number of design variables. This paper addresses two aspects of holistic heat pump design: A collaborative design architecture is introduced as a multilevel approach. Multiple components are designed in subproblems, which leads to a high number of required simulations. To mitigate high computational effort, the second focus is on approximation of the component analyses with the use of computationally inexpensive surrogate models. It is found that Gaussian process regression is most accurate and outperforms both linear regression and radial basis functions. In comparison to previous studies, the number of iterations for holistic design is drastically reduced to only 500. The presented optimization architecture also accelerates the process producing results in 35 CPU hours. The overall number of function evaluations for complex compressor design can be kept below 1000 simulations. In conclusion, the proposed strategy is very promising for application in future heat pump design with high potential for further research.