Interpretable Machine Learning-Based Prediction and Analysis on Flow and Heat Transfer Characteristics of Phase Change Material Slurry in Helically Coiled Tube with Twisted Tape Insert

Abstract

With growing environmental risks from nonrenewable energy and refrigerants, ice slurry provides a sustainable cooling solution. Enhancing ice slurry’s heat transfer through helically coiled tube with twisted tape insert (HCTTT) offers a promising method for improving energy efficiency in cooling systems. This study employs the Euler-Euler model to numerically investigate the thermal-hydraulic properties and heat transfer enhancement mechanism of ice slurry in HCTTT. The influence of twisted tape width to tube diameter ratio (Wt/d), twist ratio (y), coil diameter to tube diameter ratio (D/d), Reynolds number (Re), inlet ice concentration (αin), and ice particle diameter (ds) on flow and heat transfer is systematically assessed. The findings show that twisted tape in HCTTT induces vortex flows, which intensify secondary flow and enhance heat exchange performance. Both the Nusselt number (Nu) and resistance coefficient (f) increase with decreasing y and increasing Wt/d. To address phase-change complexities, six machine learning (ML) models, namely BPNN, ELM, SVM, LSSVM, XGBoost, and RF, are employed to predict Nu and f. A novel Crested Porcupine Optimizer (CPO) algorithm, integrated with 5-fold cross-validation, is implemented for hyperparameter tuning. Among the six ML models, the CPO-ELM and CPO-SVM exhibited superior predictive capabilities for Nu and f, respectively. A comparative analysis between five empirical correlations and ML models demonstrates a 30%-78% improvement in prediction accuracy achieved by ML approaches, highlighting that ML is effective for predicting the flow and heat transfer of ice slurry. Finally, new correlations are developed to predict both Nu and f for ice slurry in HCTTT.

Keywords: ice slurry, Twisted tape insert, Machine learning, Crested Porcupine Optimizer, SHapley Additive exPlanations.