LINEAR, kNN, SVM, AND RF REGRESSION APPLICATIONS FOR TEMPERATURE SEPARATION PERFORMANCE OF A RANQUE-HILSCH VORTEX TUBE USING AIR AND O2

Abstract

The performance of a counterflow Ranque-Hilsch vortex tube was modeled by applying different prediction models using the data obtained via the experiments with compressed air and oxygen. The data obtained using nozzles made of different materials (polyamide, aluminum, steel, and brass) were analyzed with four different machine learning methods, in which nozzle and fluid properties were used as input parameters. The performance parameter temperature gradient (Delta T) is used as the output parameter. Linear, random forest (RF), k-nearest neighbor (kNN), and support vector machine (SVM) regression models were used for vortex tube performance estimation, and Delta T parameter behavior was modeled in two different ways as calculated and predicted. In addition, two different datasets were used and precision percentages were calculated for each method. The data were divided into 80%-20% and 90%-10% training and testing datasets and calculations were performed. The highest accuracy ratio was obtained with the SVM regression method as 0.9554, followed by the ratios of RF, kNN, and linear regression models, respectively.