Estimation of trapping efficiency of a vortex tube silt ejector

摘要

This paper investigates the potential of an adaptive neuro-fuzzy inference system (ANFIS), Gaussian Process Regression, M5P tree model, Random Forest (RF) and Multi-Linear regression approaches in estimating the trapping efficiency of a vortex tube ejector. As many as 144 data sets were obtained by conducting experiments with a vortex tube ejector. Out of 144 data sets, 100 randomly selected data were used for training and the remaining 44 were used for testing the models. The input data set consists of sediment size (mm), concentration of sediment (ppm), ratio of slit thickness, diameter of tube (t/d) and extraction ratio (%) whereas trapping efficiency (%) was considered as the output. Three membership functions, i.e. triangular, generalized bell-shaped and Gaussian were used with ANFIS. A comparison of results suggests that the Gaussian membership function (MF) of ANFIS was the better of the two MFs of ANFIS. The major conclusion was that RF works better than other approaches and it could be successfully used in prediction of trapping efficiency of a vortex tube ejector. Sensitivity analyses suggest that extraction ratio was the most important parameter in estimating trapping efficiency of a vortex tube ejector.