Automatic detection of pavement surface crack depth on Florida roadways.

摘要

Cracking has an adverse effect on pavement performance, and hence it is an important criterion for maintenance intervention. However, accurate detection of the extent of cracking can also be one of the major difficulties encountered when implementing a Pavement Management System (PMS). This dissertation is based on a project sponsored by the Florida Department of Transportation (FDOT). It presents a computerized system that realizes automatic detection of pavement surface crack depth.; The crack depth detection can be divided into two parts: crack detection and crack depth estimation. Several plans were proposed based on the analysis of many developed systems in the first stage of this study. Preliminary laboratory experiment results indicated that laser technology could be used to detect pavement cracks. With the most advanced laser displacement sensors, the system realized non-contact and dynamic pavement-crack detection with high resolution. It can also automatically record the pavement microscopic profile with high accuracy. Developed software is used for data processing and management. To cancel the effect of the scan rate on the system performance, a scan-rate-effect-canceling model was developed. As the key point of the data processing program, various developed crack detection algorithms have also been studied. A new Partial Cross Correlation (PCC) algorithm was developed to enhance the crack detection ability. Also, performance analysis is done through field tests.; A neural network model was developed to estimate the actual crack depth. The database used for model development was comprised of two parts: one was the instrument reading including the geometric characteristics of the crack; the other included pavement related variables. The crack information data covered ninety-five pavement sections, which were scattered within five counties in Florida. In the model development, different network architectures and different training algorithms were investigated. An optimal architecture was determined. Early stopping was applied to gain good generalization and avoid overfitting. Through the performance evaluation and validation, it was proved that the developed system realized the automatic pavement surface crack depth detection in a non-destructive mode. With the reliable results, the system is ready to be used in the field application.