Optimization and thermal-performance deep learning on carbon/epoxy composite panels with microchannel structure for battery cooling

Yong Chen,Dongming Li,Shuai Feng,Qiong Huang,Zheming Chen,Dongwei Shu

DOI: https://doi.org/10.1016/j.applthermaleng.2022.119162

IF: 6.4

2022-11-25

Applied Thermal Engineering

Abstract:Carbon/epoxy composite with microchannel structure possesses potential commercial application in battery thermal management systems. Herein the thermal performance of the carbon fiber/epoxy composite panels with parallel microchannels was investigated by numerical simulation with experimental verification. Combining numerical simulation, design of experiment (DOE), and genetic aggregation (GA) algorithm, the response surface methodology (RSM) was performed to optimize the microchannel structure accounting for lower maximum surface temperature (Tmax). Optimally inclined microchannel structures showed that the branch channels near inlet and outlet was conducive to reduce Tmax and standard deviation of surface temperature (σT) on the composite panel. As channel diameter increased from 0.50 mm to 0.86 mm, little difference (less than0.6°C) occurred on maximum surface temperature while great reduction (greater than100 kPa) on inlet pressure took place. Flow rate greatly affected the panel’s cooling performance, and there existed an optimal flow rate to improve the cooling efficiency of the microchannel structure. Consolidating diverse inputs of microchannel-structure parameters, flow rates and channel diameters, and outputs of both Tmax and σT, a fully connected deep learning neural network model using two algorithms of Levenberg-Marquardt and Bayesian regularization was setup. Results showed that the neural network model predicted better output responses than RSM with higher R values. Bayesian regularization offered better accuracy and higher iteration efficiency on neural network training for Tmax and σT than Levenberg-Marquardt algorithm.

energy & fuels,engineering, mechanical,thermodynamics,mechanics