A universal correlation for predicting two-phase frictional pressure drop in horizontal tubes based on machine learning. (March 2023)

• Record Type: Journal Article
• Authors: Nie, Feng; Yan, Shiqi; Wang, Haocheng; Zhao, Cong; Zhao, Yanxing; Gong, Maoqiong
• Is Part Of: International journal of multiphase flow. Volume 160(2023)
• Publication Date: 2023-03
• Subjects: Two-phase flow -- Frictional pressure drop -- Machine learning -- Correlation -- Feature importance; Multiphase flow -- Periodicals; Écoulement polyphasique -- Périodiques; Multiphase flow; Periodicals; 620.1064
• DOI: 10.1016/j.ijmultiphaseflow.2022.104377
• Languages: English
• ISSNs: 0301-9322

Abstract

Highlights: A database of frictional pressure drop data covering broad conditions is amassed. Two machine learning models are designed to predict frictional pressure drops. The parametric importance analysis is conducted to obtain key parameters. A new universal correlation is developed using the obtained key parameters. An excluded database is collected to verify the generality of the new correlation. Abstract: Accurate prediction of the two-phase frictional pressure drop (FPD) facilitates designing compact phase-change heat exchangers. This study implements ML methods and develops a new universal correlation for predicting FPDs in adiabatic and diabatic flow. A consolidated database with 8663 experimental samples from 64 published literature covering 25 fluids and broad operating conditions is amassed to serve this purpose. The designed ML models, based on artificial neural network (ANN) and extreme gradient boosting (XGBoost) theory, can predict the unknown data with the best mean absolute relative deviate (MARD) of 8.59% and the coefficient of determination ( R 2 ) of 0.988 or higher. With the aid of parametric importance analysis conducted by the trained ML models, the obtained key parameters, including vapor quality, dimensionless vapor velocity, Froude number, Bond number, and convection number, are adopted to formulate a new correlation. The new correlations can provide more reliable predictive performance than existing correlations for the consolidated database with a MARD of 24.84%. An excluded database is collected to verify the generality of the correlation, and a lower MARD of 27.47% than existing models is obtained. It demonstrates that ML methods can help develop the universal correlation with improved accuracy and universality.