The study of the resistance of plain concrete to high temperatures is a current topic across the field of civil engineering diagnostics. It is a type of damage that affects all components in a complex way, and there are many ways to describe and diagnose this degradation process and the resulting condition of the concrete. With regard to resistance to high temperatures, phenomena such as explosive spalling or partial creep of the material may occur. The resulting condition of thermally degraded concrete can be assessed by a number of destructive and nondestructive methods based on either physical or chemical principles. The aim of this paper is to present a comparison of nondestructive testing of selected concrete mixtures and the subsequent classification of the condition after thermal degradation. In this sense, a classification model based on supervised machine learning principles is proposed, in which the thermal degradation of the selected test specimens are known classes. The whole test set was divided into five mixtures, each with seven temperature classes in 200 °C steps from 200 °C up to 1200 °C. The output of the paper is a comparison of the different settings of the classification model and validation algorithm in relation to the observed parameters and the resulting model accuracy. The classification is done by using parameters obtained by the acoustic NDT Impact-Echo method and image-processing tools.
Keywords: Impact-Echo; concrete; high temperatures; image analysis; machine learning; nondestructive testing; resonance method.