Files
Abstract
This research represents a comprehensive exploration of the utilization of Ground Penetrating Radar (GPR) technology and advanced deep learning models in the assessment of civil engineering infrastructure, with a particular focus on road construction and pavement maintenance. The research comprises two distinct yet interrelated components: the development of the GPR-Density model for subgrade layer evaluation and the introduction of the GPR-YOLOR model for subsurface distress detection. Together, these components provide novel and practical solutions to long-standing challenges in infrastructure assessment.The GPR-Density model offers a non-invasive means to assess subgrade layer physical parameters, notably dry density, contributing to more accurate and efficient construction quality control. The model derives from a robust mathematical framework, involving the relationship between the dielectric constant of soil and its density, among other properties. Calibration in laboratory settings, combined with field test validation, demonstrates the model's potential for accurately estimating soil properties, particularly for fine-grained soil types. The research highlights the feasibility of using GPR to predict subgrade dry density and paves the way for future investigations into network-level road assessment.
In parallel, the GPR-YOLOR model, integrated with various image enhancement techniques, revolutionizes pavement assessment by enhancing the reliability and speed of subsurface distress detection. This deep learning framework, known for its speed and accuracy, utilizes high-resolution GPR images to detect and classify pavement cracks and abnormalities. The model's real-world implementation is underscored by extensive field tests and validation, ensuring its applicability and effectiveness in practical scenarios. Furthermore, the study explores the potential integration of GPR data with other non-destructive testing technologies, offering a more comprehensive assessment of infrastructure health.
While this research marks significant advancements in GPR-based infrastructure assessment, it acknowledges limitations, including the need for expanded datasets, refined image enhancement techniques, and a focus on the mechanical properties of pavement layers. These challenges provide fertile ground for future research, ensuring the continuous evolution of non-destructive testing methodologies in civil engineering. Integrating remote sensing and satellite imagery analysis techniques with GPR reflection calculations is also proposed as an innovative avenue for large-scale infrastructure monitoring and management.
In summary, this research not only contributes to the field of civil engineering and pavement infrastructure assessment but also serves as a testament to the potential of emerging technologies and methodologies in addressing critical challenges in maintaining and managing vital infrastructure assets.