Performance modeling of parallel-connected ranque-hilsch vortex tubes using a generalizable and robust ann

Abstract

In this study, the performance patt ern of air-driven two parallel-connected Ranque-Hilsch vortex tubes (RHVT) by using artificial neural network (ANN) is considered. Different parameters such as vortex tube inlet parameters, type of working fluid, nozzle material, and nozzle number affect the temperature separation in vortex tubes. In this context, overall temperature difference (?T), which is also known as the effectivity indicator of vortex tubes, was modeled according to the aforementioned parameters which were obtained from experiments. A novel framework is presented to make the ANN model generalizable and robust. The ?T quantity was selected as an output parameter and obtained with the well-trained ANN structure according to nozzle material (thermal conductivity), nozzle number, and inlet pressure. The coefficient of determination (R2), post error ratio (C), and the mean absolute percentage error (MAPE) of the proposed ANN model have been calculated as 0.9878, 0.19, and 0.0671, respectively. To model an experimental process, shorten the time, and save costs, a decision-support system was designed with three types of input parameters that are heat transfer coefficient of nozzle material, inlet pressure, and nozzle number. Thus, the system easily calculating the ?T value by the generalizable and robust ANN model, which is the first trial for a parallel-connected system allowing the decision-maker to use different parameter values and different materials, is constituted. © 2020 Elsevier B.V., All rights reserved.