GPR method for estimation of unbound load-bearing layers’ parameters using machine learning and interactive modelling

Abstract

The demand for nondestructive testing techniques (NDTs) that can be implemented continuously and cost-effectively for large-scale study purposes is increasing. Ground penetrating radar (GPR) as an NDT method permits the estimation of pavement materials characteristics without disrupting the serviceability of the system. This research used typical pavement materials for constructing load-bearing layers (base and subbase) for the GPR laboratory tests. A 2 GHz GPR antenna was applied to execute the tests by changing three essential variables of the material: water content, compaction and clay content. Machine learning and interactive methods were innovatively utilised to model the collected data. As a result, a multivariate non-linear empirical function is proposed by the interactive procedure. Furthermore, the machine learning modelling (SVM method) with R2 of 0.91 indicates promising results to evaluate the pavement layers’ properties. Machine learning can enhance the speed and accuracy of analysis when faced with multi-variables and extensive data.