Predicting thermal contact resistance from infrared temperature fields using machine learning

Thermal contact resistance (TCR) plays a crucial role in determining the overall heat transfer across interfaces in applications such as aerospace structures, electronics cooling, and battery packs. Accurate TCR prediction is therefore important in the design and optimization of thermal management systems. Because TCR is influenced by many factors, predicting it is a complex nonlinear problem, however, the effects of those factors are already embedded in the temperature distributions, which can be used to estimate the TCR directly. To address this issue, in the current study, machine learning (ML) is combined with infrared thermography to capture the complex nonlinear relationship between the temperature distributions and the corresponding TCR. A laser is used to slightly heat the top surface near the interface between two materials, and an infrared camera captures the resulting temperature distribution, which is subsequently processed with ML. Because capturing the thermal images for a large set of materials experimentally will be time-consuming and expensive, therefore, to test this hypothesis, COMSOL Multiphysics software is utilized taking advantage of its ability to generate large datasets for the ML algorithms. The ML algorithms used are based on a Convolutional Neural Network (CNN) combined with five other regressors, and they predicted TCR with high accuracy. The best models were the CNN with extra trees (ET) and k-nearest neighbor (KNN). Additionally, two parametric studies were conducted for the unknown surface reflectivity and the convective heat transfer coefficient, the KNN model was used and demonstrated high precision for predicting TCR. A nondimensional approach of recovering TCR from a dimensionless variable was also compared to the direct method of predicting TCR.