Active subspaces for the preliminary fluid dynamic design of unconventional turbomachinery

Abstract

The fluid dynamic preliminary design of unconventional turbomachinery is customary done with meanline design procedures coupled with gradient-free optimizers. This method features various drawbacks, since it might become computationally expensive, and it does not provide design insights or guidelines to the designer. This work proposes a strategy to abate this disadvantages, namely, the construction of a reduced-order model by means of active sub-spaces, and the use of the surrogate combined with a gradient-based optimizer. The case study is the design optimization of a Organic Rankine Cycle radial inflow turbine. The results show that active subspaces exist for this application, and that it is possible to construct a surrogate with an approximate error of ±1% for the total-to-static efficiency. Additionally, the optimization using the surrogate leads to accurate results and a computational cost at least four times faster. Furthermore, the results reveal that the models for unconventional turbomachinery feature multiple regions containing constrained optima. Active subspace methods thus prove to be a promising alternative for optimization of unconventional turbomachinery.