Estimation of the thermal conductivity of cement composites using bayesian statistical approach

Abstract

In cold regions, black ice was a big problem because it caused severe life losses. Already the pavement heating systems were studied by many researchers. However, it was still a big problem to take a long melting time. Therefore, enhancing the thermal conductivity was the clear way to reduce the melting time. In addition, the proportion of research using simulation increased in construction research. Therefore, it is necessary to study in connection with predicting the thermal conductivity of materials through simulation. In this study, as an initial step of study on this, an equation for predicting the thermal conductivity of cement composites was proposed. In addition, along with the proposed equation, the Gaussian process regression (GPR) was performed to estimate the thermal conductivity of the cement composites. The used data was collected by experiment, and the lengthscale hyperparameter was calculated through the data to apply to a kernel. The kernel was chosen as the Matèrn kernel of k5/2 because this kernel could reflect the lengthscale. With the k5/2, the estimation of thermal conductivity through the fitted GPR and proposed equation showed a high accuracy with R2 = 0.9986 in the test group of this study. Furthermore, this process showed the accurate estimation with R2 = 0.9443 of the thermal conductivity in the existing studies even the studies including the geopolymer case. Therefore, it was demonstrated that this work could be applied to existing cases, and the accuracy was high.