Machine learning models applied to moisture assessment in building materials

Abstract

Moisture-related defects hinder long-term building durability and must be prevented. Non-destructive techniques that measure the surface temperature of building materials have good potential for moisture analysis. This article presents machine learning models to predict the moisture content of materials according to their surface temperature using as input the material and the environmental conditions. Results showed that neural network models present a coefficient of determination higher than 0.96 and an error of less than 5%. The application of the models demonstrated that in brick and granite the model presented excellent results, but in limestone and concrete it enabled the identification of moisture only in the near-surface zone.