Machine Learning Models for Bedrock Condition Classification in Pavement Structure Evaluation: A Comparative Study

Abstract

Pavement performance evaluation based on modulus is crucial for controlling the overall performance of pavements and decisions making throughout the pavement’s life cycle. Falling weight deflectometer (FWD) tests are commonly employed to collect deflection data, which is subsequently back-calculated to get each layer’s modulus. However, existing studies lack a complete framework for incorporating the bedrock condition in the back-calculation process. Here, an integrated process of pavement performance evaluation utilizing FWD tests is proposed, and the focus is on the classification of bedrock condition by modern classification algorithms (BPNN, MLP, SVM, and RF) to determine the presence or absence of bedrock and its depth range. The implementation of classification process allows for the inclusion of bedrock influence in the back-calculation process, thereby improving the accuracy of modulus results. Results from the four classification algorithms reveals that RF is the most suitable for classifying bedrock depth, exhibiting superior overall performance. The proposed integrated back-calculation process enables a comprehensive and objective evaluation of pavement structural performance, providing a valuable framework for informed decisions making.