A Symmetry-Based Hybrid Model of Computational Fluid Dynamics and Machine Learning for Cold Storage Temperature Management

Abstract

Cold chain temperature management is crucial for preserving product quality and safety across various industries. While Computational Fluid Dynamics (CFD) provides detailed insights into thermal analysis and fluid dynamics, its computational intensity limits practical applications. This study presents a novel hybrid approach combining CFD and machine learning to enhance both computational efficiency and prediction accuracy in cold storage temperature management. A validated 3D CFD model was developed to analyze temperature distribution and airflow patterns in a refrigerated container with multiple storage boxes. Taking advantage of the cold room’s symmetrical design along its longitudinal axis significantly reduced computational requirements while maintaining model accuracy. Over 200 cases were simulated by varying key process parameters to generate training data for machine learning models. Random Forest (RF) and Neural Network (NN) models were developed and compared, with RF demonstrating consistently superior performance across all storage locations. Feature importance analysis revealed cold air temperature as the dominant control variable, while SHAP analysis identified optimal operational ranges for air velocity and heat transfer coefficients that balance product quality with energy efficiency. This research work also revealed distinct patterns in the influence of process parameters, with cold air and ambient temperatures showing hierarchical impacts on system performance. The hybrid methodology successfully addresses the computational limitations of traditional CFD approaches while maintaining high prediction accuracy, offering a practical solution for sustainable temperature management in cold storage applications. Finally, this research provides valuable insights for optimizing cold chain operations and demonstrates the potential of hybrid modeling approaches in thermal management systems.