Table './tdc/tbl_dc_meta_control_57' is marked as crashed and last (automatic?) repair failed
Tang, Ekuong. Application of artificial neural networks for predicting heat transfer performance of phase change material tube bank under electric field. Master's Degree(Energy Engineering). Chiang Mai University. Library. : Chiang Mai University, 2025.
Application of artificial neural networks for predicting heat transfer performance of phase change material tube bank under electric field
Abstract:
Numerous studies have explored the use of artificial neural networks (ANNs) to predict the heat transfer performance of tube bank heat exchangers subjected to electric fields. ANNs are instrumental in addressing the complexities of heat exchanger systems by efficiently managing nonlinear interactions and multiple variables, delivering accurate predictions where traditional methods often fall short. This research has significant practical relevance, particularly in HVAC systems, where enhanced heat transfer performance can lead to reduced energy consumption and lower operational costs. It also introduces innovative strategies for optimizing industrial heat exchanger designs. This study conducted an experimental analysis of the thermal performance of a tube bank configuration consisting of four rows and four columns of tubes, each with a diameter of 2 cm. The system maintained uniform airflow across the tube bank, with transverse and longitudinal pitch ratios of 2 cm and 4.5 cm, respectively. The paraffin phase change material (PCM) type RT31 was utilized during the discharge process under electrohydrodynamic (EHD) conditions. The effects of EHD voltages ranging from 0 to 17 kVDC and frequencies between 5 kHz and 100 kHz were investigated. The experimental results identified 30 kHz as the optimal frequency for high DC power supply of 017 kVDC to the PCM within the tube bank conditions under air velocity were varied between 0.2 and 1.2 m/s. The dataset consisted of 2,056 data points under the conditions such as inlet air temperature of 40°C, a duration ranging from 0 to 253 minutes, an air velocity between 0.2 and 1.2 m/s, and a high-voltage DC range of 717 kVDC, with outlet air temperature and RT31 temperatures serving as output variables. Various configurations of the artificial neural network (ANN) model were tested, including different activation functions and varying numbers of neurons in the first hidden layer. The results indicated that the sigmoid-purelin activation function combination was well-suited for heat transfer analysis. The optimal model performance was achieved with 10 neurons in the first hidden layer and a training epoch of 39. The selection of the optimal model was based on achieving the lowest mean squared error (MSE) of 10⁻⁸ across training, testing, and validation datasets. The ANN model exhibited strong predictive capabilities, closely matching experimental data and accurately predicting system performance under various conditions within the tube bank. Additionally, correlations were established to estimate outlet air temperatures for different maximum Reynolds numbers of the tube bundle, both with and without the application of an electric field. The study further demonstrated that electrohydrodynamic (EHD) technology reduced power consumption by up to 13% compared to non-EHD PCM tube bank conditions, highlighting its potential for energy savings in heat exchanger applications.