Using ANNs in calibrating the measurements of a simplified hot-plate method

A simplified hot-plate method (SHOP) using single constant high temperature region in measuring materials' thermal conductivity, k, has been developed. This method is characterized by its simplicity that it can be used to fabricate portable devices for in-situ measurements and its ability to perform steady-state and non-destructive measurement of composite materials. However, the systematic error introduced by the one-dimensional approximation extended to two-dimensional problems requires a calibration method for the measurement using the SHOP. In this paper, an artificial neural network (ANN) for improving the accuracy of the SHOP is presented. It was devised using the data of 720 measurement cases performed with an experimentally verified numerical model. The experiments of measurement using a prototype instrument fabricated according to the SHOP were carried out. The k values were predicted with the ANN and compared with those using the Hot Disk^® apparatus to verify the ANN. The results showed that the ANN improved the accuracy of the SHOP and resolved the composite materials that could not be resolved by the Hot Disk^® apparatus. The coefficients of k as linear functions of temperature can also be obtained by two independent measurements with two high temperature setups corresponding to a fixed ambient temperature. The rules of setting two high temperatures are also given so that the coefficients can be determined within 1% of relative error.