Separator performance modeling and analysis using artificial neural network and response surface method

Abstract

The performance of swirl-vane separator is mainly obtained by experiments or numerical simulation, which is expensive and time-consuming. ANN models based on a back-propagation neural network (BPNN), a generalized regression neural network (GRNN), and an Elman neural network (Elman NN) have been developed. 127 experimentally obtained separator performance data points are used for model training and verification. The results show that GRNN model demonstrated the highest prediction accuracy. Furthermore, parametric analysis for the gas–liquid separator is carried out by GRNN model incorporated with response surface method (RSM) which shows advantage to reflect the multivariate interaction of separator. It is found that three items (Rewater, Regas × d and Rewater × d) play the key roles in affecting separation efficiency. Quadratic polynomial correlations for separation performance have been developed by RSM and have discrepancies less than 7.5% compared with the experimental results, which show good performance to quickly and accurately predict the performance of separators.