Artificial Neural Network Modeling of Liquid Thermal Conductivity for alkanes, ketones and silanes

The values of thermal conductivity λ at different temperatures for organic and inorganic compounds in the liquid phase is essential in the study of numerous processes, but experimental data are frequently not available with acceptable reliability or not available at all, since rigorous theoretical or semi-theoretical models of the liquid state are usually of poor practical use for engineering purposes. The Artificial Neural Network (ANN) approach is a very powerful tool and it can be a good indicator of the lowest limit achievable with a selected database and with a selected set of inputs. This study investigates the applicability of the ANN as an efficient tool for the prediction of pure organic compounds' thermal conductivity of three important families such as alkanes, ketones and silanes, for a wide range of temperatures. The families of n-alkanes, ketones and silanes were chosen to verify the general reliability of the proposed method when used in large temperature ranges for very different organic families, above all the silanes (compounds containing silicon), whose liquid thermal conductivity is experimentally investigated in very few cases. This method appears to be successful: in all reduced temperature range, along or near the saturation line, the average absolute deviations between calculated and experimental thermal conductivity data are 0.19% and the maximum absolute ones 2.44%