Estimation of stripping by static immersion test using image processing and machine learning

Abstract

Hot Mixed Asphalt (HMA) is one of the most common types of pavement, which exists on the surface of the roads, inside and outside of cities. One of the main destresses in HMA is moisture-related damage, which mainly occurs in the form of stripping. The process of losing adhesion and cohesion of asphalt cement due to the presence of moisture and cyclic loads is called “stripping”. Several test procedures have been designed and conducted on different types of asphaltic mixtures to identify and measure moisture damages, especially stripping. Stripping evaluations could be divided into two classes: tests on compacted mixtures and tests on loose mixtures. Test procedures for loose mixture have been adopted by different highway agencies, such as the Ministry of Transportation Ontario (MTO), and pavement industries, because they are easy to perform, cost-effective, and do not require complex equipment. But since stripping estimation is based on visual assessment, the results could be inconsistent when they are estimated by inexperienced operators. One of the most common tests on loose mixtures is static immersion test, and a modified version of the static immersion test has been used by MTO, listed as LS-285 R29. To evaluate stripping in this test procedure, 104g of loose asphaltic mixture should be immersed inside water for 24 hours and then the retained coating areas should be measured by a skilled technician as a percentage of the total surface area. Image processing methods are proper examples of using smart agents in visual assessment problems, such as object detection and pattern recognition. In this research, a vision-based algorithm and a low-cost light improvement system were developed as an alternative for manual judgment. The system receives images of samples captured in a controlled lighting condition, which is called illumination box, and then it applies Contrast Limited Adaptive Histogram Equalization to enhance contrast intensity of the image. In addition, the system uses inpainting to reconstruct specular highlights in the image, and then classifies the regions on the image, i.e. coated and stripped areas, using combinations of K-means clustering and K-Nearest Neighbors and Support Vector Machines classifiers. The developed system is able to overcome most of the shortcomings of prior methods, such as evaluation of the stripping on mixtures with dark-colour aggregates and processing test images without alteration of the test samples. The differences of the results in the best configuration of classifiers from manual estimations had the mean of 4.8 % and the standard deviation of 5.2 %. Moreover, application of illumination box and contrast enhancement module proved to be effective to improve the performance of this system.