Scholarlink

<p>This study investigates the application of vertical acceleration patterns from motion sensors mounted on vehicle axles to predict early failures in the sub-base and sub-grade layers of pavement structures. By leveraging real-time acceleration data and supervised machine learning algorithms, including Convolutional Neural Networks (CNNs), a predictive model was developed to identify potential failure points. Laboratory tests on soil samples collected from selected sites were conducted to correlate sensor data outputs with soil properties, enhancing the model's robustness. The model achieved a prediction accuracy of 92%, with a precision of 88% and recall of 85%, effectively identifying failure patterns. Compared to conventional methods like the Falling Weight Deflectometer (FWD), a developed model enables real-time, uninterrupted assessment over long stretches of road with significantly reduced disruption and improved early detection capability. These findings suggest significant implications for proactive infrastructure maintenance, enabling timely interventions, reducing repair costs, and improving road durability. The integration of sensor-driven data analytics and machine learning provides a scalable, cost-efficient tool for enhancing pavement management systems and ensuring sustainable infrastructure development.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Machine Learning-Driven Prediction of Pavement Sub-Grade Failures Using Vehicle-Mounted Acceleration Sensors

  • 期刊论文

  • 发表日期：

    2025年8月27日

查看全文
  • Yogesh Bafna,
  • Jigisha Vashi,
  • Santosh Bothe

摘要

This study investigates the application of vertical acceleration patterns from motion sensors mounted on vehicle axles to predict early failures in the sub-base and sub-grade layers of pavement structures. By leveraging real-time acceleration data and supervised machine learning algorithms, including Convolutional Neural Networks (CNNs), a predictive model was developed to identify potential failure points. Laboratory tests on soil samples collected from selected sites were conducted to correlate sensor data outputs with soil properties, enhancing the model's robustness. The model achieved a prediction accuracy of 92%, with a precision of 88% and recall of 85%, effectively identifying failure patterns. Compared to conventional methods like the Falling Weight Deflectometer (FWD), a developed model enables real-time, uninterrupted assessment over long stretches of road with significantly reduced disruption and improved early detection capability. These findings suggest significant implications for proactive infrastructure maintenance, enabling timely interventions, reducing repair costs, and improving road durability. The integration of sensor-driven data analytics and machine learning provides a scalable, cost-efficient tool for enhancing pavement management systems and ensuring sustainable infrastructure development.