Predicting thermal conductivity of granite subjected to high temperature using machine learning techniques

Abstract

The evolution and accurate prediction of thermal conductivity (TC) of granite subjected to high temperature is of great significance for many geological and underground engineering. In this study, the relationship between TC and temperature, mineral composition, porosity and density after high temperature was studied experimentally. Subsequently, totally 229 measurements containing four input variables (i.e., temperature, porosity, density and quartz content) were collected, and a new prediction model for granite TC was proposed using back propagation neural network (BPNN-TCPM). The results indicate that the TC of granite is strongly dependent on temperature and decreases with the increase of temperature. The TC is inversely proportional to porosity and positively related to density, the effect of temperature on the mineral content can be ignored, but the damage of mineral structure can significantly affect the heat conduction capacity of granite, which also demonstrate that the initiation and propagation of thermally-induced cracks in granite during thermal treatment is the main reason for the deterioration of TC. More importantly, machine learning (ML) techniques could prove to be highly accurate and efficient new methods for predicting the TC of granite. The prediction results on the testing data set show that the average absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R²) of the BPNN-TCPM are 0.0286, 0.0765 and 0.9785, respectively, and the prediction accuracy is better than the other 7 ML models and 8 temperature-dependent empirical models of rock TC. This also means that considering the coupled effects of multiple factors can help improve the accuracy of granite TC prediction. In addition, a graphical user interface (GUI) is developed for practical application, which can obtain single or batch TC data by directly inputting variables.