Improving the Calibration of Surface Time Domain Reflectometry Sensors for Moisture Evaluation of Building Materials Using the Analysis of Covariance Method

Abstract

The paper presents the models for moisture evaluation using a set of the reflectometric sensors in some types of building materials. The readouts reveal the relationship between the building material moisture, being assessed gravimetrically and the apparent permittivity values obtained by the TDR (Time Domain Reflectometry) method and surface sensors. Based on the readouts, equations describing this relationship were derived. These types of equations function as calibration equations and are used to calibrate the sensors. Most of the equations used to describe the examined relationships are linear regression. These equations very often refer to specific materials and cannot be applied to others that differ in density or chemical composition, which is the cause of many incorrect measurements. In this article, we propose the use of the analysis of covariance method (ANCOVA) for the analysis of reflectometric data. Using this method, it will be possible to determine the moisture content of materials, regardless of their type and construction of the sensor, which can significantly improve moisture measurements using the reflectometric method. For comparative aims data achieved in conducted research were analyzed using both traditional linear regression models and using the analysis of covariance method (ANCOVA). Both types of fitting models are discussed and their quality was compared in terms of accuracy expressed by the Residual Standard Error (RSE), the Root Mean Square Error (RMSE) and the determination coefficient (R 2 ) values. The paper showed that the use of the ANCOVA method allows for improvement the fit of the model in terms of the determination coefficient by 0.0174. Moreover, the average RSE and RMSE value in the ANCOVA models are smaller about 1.24 vol.% and 1.25 vol.% than the ones in the regression model, respectively, which means that the models obtained using ANCOVA more accurately describe the examined relationship.