Experiment on radial inflow turbines and performance prediction using deep neural network for the organic Rankine cycle

• Published in: Applied thermal engineering Vol. 149; pp. 633 - 643
• Main Authors: Kim, Jun-Seong, Kim, Do-Yeop, Kim, You-Taek
• Format: Journal Article
• Language: English
• Published: Oxford Elsevier Ltd 25.02.2019; Elsevier BV
• Subjects: Artificial neural networks; Deep neural network; Economic conditions; Energy recovery; Geothermal power; Heat recovery; Heat recovery systems; Inflow; Neural networks; Organic Rankine cycle; Performance prediction; Preliminary design program; Preliminary designs; Radial inflow turbine; Rankine cycle; Studies; Turbines; Deep neural network; Preliminary design program; Organic Rankine cycle; Radial inflow turbine
• ISSN: 1359-4311; 1873-5606
• DOI: 10.1016/j.applthermaleng.2018.12.084

•Experimental equipment was constructed based on the thermodynamic cycle.•Each turbine was manufactured by a self-developed program and a commercial program.•A turbine designed by a self-developed program showed somewhat better performance.•A prediction model was studied using experimental data and deep neural networks.•The deep neural network was able to predict the trends of the experimental results. The organic Rankine cycle makes it possible to accomplish energy recovery from a low-temperature heat source, which is typically not recovered for economic reasons. As the expander for the organic Rankine cycle, the radial turbine is easy to manufacture and has advantages in terms of size and efficiency. The radial turbine design modeler (RTDM), which was developed from in-house code, is a preliminary design program for radial inflow turbines and is different from the commercially available program RITAL. In this study, an experiment on radial inflow turbines is performed using both RTDM and RITAL. As a result, the output and efficiency of the RTDM and RITAL turbines are 36.04 kW, 80.03% and 35.03 kW, 76.01%, respectively. Experimental results demonstrate that the performance of the RTDM turbine is almost similar to the RITAL turbine. We also perform analysis on performance prediction utilizing a deep neural network with two hidden layers based on the experimental data. As a result, the minimum root mean squared errors of the RTDM turbine and RITAL turbine are estimated to be approximately 1.81 and 1.65, respectively. The deep neural network is able to predict the trends of the experiment for the organic Rankine cycle.