Machine learning for predicting the PCM melting process in a rectangular enclosure energy storage
Suping Shen, Chenlong Wu, Fei Duan
Abstract
This paper presents a novel approach to predicting the dynamic melting process of phase change material (PCM) in a rectangular enclosure with two fins using Artificial Neural Network (ANN). Numerical simulations are performed for three different fin lengths (12.5 mm, 25 mm, and 37.5 mm) to generate a comprehensive dataset for training the ANN model that captures the complex evolution of the melting front and temperature distribution. The trained ANN has high accuracy with low error metrics. For PCM melting front training, the mean absolute error (MAE), mean squared error (MSE), Pearson’s correlation coefficient (R) values are approximately 0.014, 0.03, and 0.99, respectively. Similarly, for PCM temperature distribution, these metrics are around 0.016, 0.03, and 0.99. The model is then employed to predict the melting behaviour for unseen fin lengths (20 mm and 30 mm). The predicted melting front and temperature distribution are closely aligned with numerical simulation results. The MAE, MSE, and R values for both the melting front and temperature distribution are approximately 0.02, 0.03, and 0.98, respectively. The study demonstrates the ANN model’s efficiency and accuracy in predicting the PCM melting for conditions beyond the training set, highlighting its potential to advance thermal management system designs. By reducing simulation time from around 6 days to just a few minutes with high accuracy, the ANN approach showcases remarkable efficiency. The findings show that ANNs not only save computational resources but also provide a highly effective method for modelling complex PCM melting processes.