Machine learning analysis of thermal separation of a ranque hilsch vortex tube with linear, kNN, SVM, and RF regression models

This study consists of modeling studies for thermal separation of a Ranque Hilsch Vortex Tube (RHVT) by using four different machine learning methods. Compressed air used RHVT, the data obtained as a result of experiments with different nozzles were modeled with linear, k-Nearest Neighbor (kNN), Random Forest (RF), and Support Vector Machine (SVM) regression methods to compare each other. Nozzle properties and inlet pressure were used as input parameters, and the total temperature gradient Delta T was examined as the output. Experiment results were handled in two groups as training and test groups at different percentages. Delta T calculated by modeling hot and cold output test data, and Delta T calculated directly with the experiments were modeled and compared. According to the obtained results, the highest percentage of accuracy value of 97.58% was obtained with the SVM method, and this value was obtained with the set in which 90%-10% of the experimental results were used as the training and test data, respectively. The accuracy ratios calculated with RF, kNN, and linear regression models under the same conditions are 93.99, 88.49, and 78.97, respectively.