An Automated 3D Crack Severity Assessment Using Surface Data for Improving Flexible Pavement Maintenance Strategies

Evaluation of crack severity in flexible pavements predominantly centers around the analysis of cracks’ surface characteristics. However, this study highlights the critical importance of 3D crack parameters, including volume and depth, for comprehensive assessment. The objective here is to develop an autonomous crack severity assessment, by predicting the vertical parameters of cracks exclusively from their surface properties. To achieve this, a dataset of 3D parameters comprising 200 cracks from eight flexible pavements was acquired, and both linear and nonlinear correlations were conducted among these 3D parameters. Subsequently, five single-output and one multi-output machine learning models were developed to explore the potential of utilizing surface parameters to predict the vertical parameters of cracks. The outcomes validated the effectiveness of two specific methods, namely, Artificial Neural Network and Extreme Gradient Boosting models, in predicting crack volume based on surface parameters, with R2 scores of 0.832 and 0.748, respectively. Additionally, the multi-output machine learning model we developed achieved classification prediction of the crack damage penetration depth using surface parameters, yielding optimal precision, recall, and F1 scores of 0.790, 0.779, and 0.761, respectively. This study has introduced a crack damage evaluation index, based on a 3D assessment, that relates crack depth classification to severity. We provide suggestions that could pave the way for informed decision-making on maintenance strategies that could be adopted to extend asset life cycle.