Machine-learning based thermal conductivity prediction of propylene glycol solutions: Real time heat propagation approach

Abstract

The objective of this paper is to evaluate the capability of an ANN to classify the thermal conductivity of water-glycol mixture in various concentrations. Massive training/validation/test temperature data were created by using a COMSOL model for geometry including a micropipette thermal sensor in an infinite media (i.e., water-glycol mixture) where a 500 ?s laser pulse is irradiated at the tip. The randomly generated temporal profile of the temperature dataset was then fed into a trained ANN to classify the thermal conductivity of the mixtures, whose value would be used to distinguish the glycol concentration at a sensitivity of 0.2% concentration with an accuracy of 96.5%. Training of the ANN yielded an overall classification accuracy of 99.99% after 108 epochs.