Session: 37-01 Machine Learning and Optimization

Paper Number: 82590

82590 - Detailed Design and Optimization of the First Stage of an Axial Supercritical CO2 Compressor 

The detailed design of the first stage of a 3-stage supercritical carbon-dioxide (S-CO2) compressor rig is presented.  The plan will be to test this single stage at the University of Notre Dame Turbomachinery Lab followed by the 3-stage rig.  The power required for the 3-stage rig at full power is expected to be 9 MW at 19,800 rpm.  This will be the scaled first stages of a 9-stage compressor that would be used for an energy storage application.

The thermodynamic properties of S-CO2 allow for high thermal efficiency and large power. A detailed design optimization methodology was developed and implemented in Python to navigate the parametric design space of compressor blades in S-CO2. This approach first optimizes a model geometry in quasi-3D CFD at 5 sections. The quasi-3D optimized blade is then the initial geometry for the subsequent full 3D CFD optimization.

The blade geometry is generated using the open-source program T-Blade3 developed at the University of Cincinnati. The airfoil blade performance is calculated using MISES, a quasi-3D blade-to-blade solver developed and owned by MIT. For full 3D CFD, the FineTurbo by Cadence is utilized. The optimization driver is the OpenMDAO framework developed by NASA written in Python3.

Static structures, modal analysis, and hot to cold transformation will also be presented showing the detailed process of creating solid geometry with Engineering Sketch Pad (ESP) and the Finite Element Code, ANSYS.

For the quasi-3D fidelity, the gradient based optimizer in OpenMDAO is utilized to communicate and drive T-Blade3 and MISES. Parameters that control the blade shape are modified by OpenMDAO using finite difference steps. Each airfoil shape is processed through MISES and an objective function that combines on and off design incidence is returned to OpenMDAO to improve blade efficiency as well as operational range.

The full 3D optimization uses a generic algorithm methodology in OpenMDAO to navigate the design space which is also modeled by a multi-objective function that includes on and off design efficiency. 

The performance prediction is then presented in detail.

 

Keywords: Compressor, Optimization, S-CO2

 

For the Vanguard Chair:  This paper has been submitted to the Design Methods and CFD track for the Turbomachinery Committee.  It could also be of interest for Axial Compressor track and for the Supercritical CO2 committee.

Presenting Author: Matthew Ha University of Cincinnati

Presenting Author Biography: PhD student student studying turbomachinery design using multi-disciplinary analysis and optimization methods.<br/><br/>Currently a graduate researcher in the Gas Turbine Simulation Laboratory at the University of Cincinnati.<br/><br/>Interests include: Turbomachinery design, CFD, MDAO, FEA, applied machine learning and AI, numerical methods, and SCO2.

Authors:

Matthew Ha University of Cincinnati
Justin Holder University of Cincinnati
Saugat Ghimire University of Cincinnati
Adam Ringheisen University of Cincinnati
Mark Turner University of Cincinnati