Detecting Delamination of Pavement Layers in Airfield Runways Using Local Vibration Testing and Machine Learning

Abstract

Delamination between the layers of asphalt pavement in airfield runways is studied by local vibration testing and machine learning. Debonding between pavement layers is modeled by placing sand or paper between the layers in the field experiment. The vibration testing is able to detect the delamination by the decreasing response of resonant frequency, however, a large variability was produced in the data taken during the summer season. This is likely due to the large variations of asphalt temperature, consequently changing the asphalt properties. For the specific case, machine learning is used to assist in capturing the debonding condition of the layers. The experimental data is first studied by supervised learning using a support vector machine (SVM); however, the damage was not detected well due to the lack of specific examples (i.e., delamination of the layers with exact depth) for input. Instead, machine learning using a principal component analysis (PCA) is applied where only standard data (i.e., without delamination) is given so that when an abnormality such as delamination of the layers is present, it is detected. Using this method, the presence of the delamination of the pavement layers was identified by an accuracy rate of around 90%. This indicates the proposed method, based on local vibration testing, can detect delamination of asphalt pavements with a high accuracy.