ABSTRACT
California Bearing Ratio and Modulus of Subgrade Reaction are key parameters for pavement evaluation, often estimated through empirical equations. While convenient, these methods overlook soil variability. This study explores Ultrasonic Pulse Velocity (UPV) testing as an alternative to predict subgrade failures in pavement structures, avoiding time-intensive traditional tests. Acceleration data were collected using Arduino Uno-based sensors mounted on vehicles across selected sites, and UPV tests were conducted on soil samples from these locations. Results showed poor soil quality, indicated by low primary and shear wave velocities. Additional parameters such as Poisson’s ratio, modulus of elasticity, and bulk modulus further confirmed reduced soil stability. A supervised machine learning model using a Convolutional Neural Network (CNN) was developed based on the sensor data, achieving 92% accuracy, 88% precision, and 85% recall. The study demonstrates the effectiveness of combining UPV testing with real-time data and AI to improve road maintenance, reduce costs, and ensure infrastructure reliability.
