Evaluation of interlayer bonding in layered composites based on non-destructive measurements and machine learning: Comparative analysis of selected learning algorithms.

In this work, the results of experimental research and numerical analyses considering using non-destructive testing and machine learning algorithms are presented. They were used to evaluate the interlayer bonding between the concrete repair mortar and repaired element. The analyses of using different machine learning algorithms for this purpose are presented and compared. Based on non-destructive and destructive measurements performed in 300 measuring places, the database was built. In order to predict the output, which was the pull-off adhesion, the artificial neural network, support vector machine and random forest algorithm were used. For simplifications of the numerical model as input only a few parameters were selected. Based on the numerical analyses, the most accurate algorithm for non-destructive evaluation of the pull-off adhesion in concrete repaired elements was selected. Its potential for using in construction practice was strengthened by performing experimental verification on separately made model element especially for this purpose. • The usefulness of machine learning algorithms for engineering purposes. • Evaluating pull-off adhesion by means of nondestructive measurements. • Comparative analyses of selected machine learning algorithm and NDT. • Simplified method for investigating the interlayer bonding. [ABSTRACT FROM AUTHOR]