Estimating Stripping of Asphalt Coating Using k-Means Clustering and Machine Learning–Based Classification

Abstract

Stripping is a primary form of moisture-related damage in hot mix asphalt, which mainly results from a loss of bond between the asphalt cement and aggregate. Static immersion and boiling water tests are common methods to estimate stripping of bituminous cover from the aggregate surfaces in loose mixtures, but the accuracy of the assessment depends on the skill and experience of the technician; therefore, alternatives to subjective visual assessments were sought. Image processing methods are known to be reliable means for quality control in different areas and are able to address the inconsistency issues noted with manual assessment. This paper presents an automated image processing system developed to assess the stripping of asphalt coating from aggregate surfaces using the static immersion test. This framework uses a set of preprocessing methods to improve the contrast and lighting condition of the samples. Then it uses k-means algorithm to segment pixels with similar values on the surface of aggregates. Finally, two machine-learning methods were used to classify whether the resulting clusters represent an asphalt coated or uncoated area on the aggregate surfaces in a loose mixture image. This system was evaluated using 159 test samples and demonstrated promising performance, with a mean difference of 4.91% from the technician assessments and standard deviation of 6.50%.