A novel neural network and sensitivity analysis method for predicting the thermal resistance of heat pipes with nanofluids

Abstract

It is critical to accurately predict the heat transfer performance of heat pipes with nanofluids for the design of the thermal management system. The current prediction models for the thermal performance of heat pipe with nanofluids are mostly developed based solely on their experimental data, resulting in poor universality and low predictive performance. Moreover, there is a lack of quantitative analysis of the correlation degree of selected input parameters on the thermal performance of heat pipe. The innovation of this work lies in the development of a novel thermal resistance prediction model based on neural network and sensitivity analysis method, characterized by a broad range of inputs and high accuracy. Firstly, a neural network database was established based on our own experimental data and published literature. The effects of the learning rate and the number of nodes in the hidden layer on the prediction performance were investigated. Subsequently, the Garson algorithm was used for sensitivity analysis of the input parameters. The results showed that when the Adam adaptive optimization algorithm and 28 nodes in the hidden layer were used, the neural network model had the highest prediction accuracy, with a mean square error (MSE) of only 0.0020 and a mean absolute error (MAE) of 0.03 K/W, and a mean relative error (MRE) of 7.86 %, demonstrating that the developed model can be used for predicting the performance of heat pipes. The ranking of the influence of each input variable on the thermal resistance from high to low is as follows: heat load, confinement number, inclination angle, cold source temperature, filling ratio, thermal conductivity, and concentration. Moreover, the influence of concentration on the accuracy of the model cannot be ignored. This work provides a direct guidance for developing a performance prediction model for heat pipe with nanofluids with high universality and accuracy.