Assessment of thermal conductivity of rocks using regression analyses and artificial neural networks.

This study investigated the thermal conductivity of natural stones (k) through regression analyses and artificial neural networks (ANN). In order to gather a sizable number of datasets for the aforementioned analytic methodologies, a thorough literature review was carried out. Based on different physicomechanical rock characteristics, like dry density (pd), effective porosity (ne), uniaxial compressive strength (UCS), and pulse wave velocity (Vp), seven estimated models (M1-M7) were created for the evaluation of k. The regression-based models (M1-M5) demonstrated that the considered rock properties influence the k of natural stones at different degrees. Notably, the ne and Vp were found to be highly correlative parameters for estimating the k of natural stones. A number of statistical indicators were used to assess the performance of the developed models. The statistical evaluations indicated that the ANN-based models (M6, M7) provided more consistent results than the M1-M5 models. In addition, the mathematical expressions for ANN-based models were also given in the present study to let users carry out them more efficiently. In this case, this study is thought to ensure applicable and comprehensible information on the heat conduction of natural stones and can be described as a research study on how to model the k of natural stones as a factor of various rock characteristics. [ABSTRACT FROM AUTHOR]