Development of an automatic detection system for measuring pavement crack depth on Florida roadways

Description

Cracking has an adverse effect on pavement performance, and hence it is an importantcriterion for maintenance intervention. However, statistically reliable detection of theextent of cracking can also be one of the major difficulties encountered whenimplementing a pavement management system. The main difficulty is that there is noreliable way to directly obtain the information on pavement crack depth withoutdestructing the pavement structure. This report summarizes a research project sponsoredby Florida Department of Transportation (FDOT) to develop a system that canautomatically and dynamically measure pavement cracks and estimate crack depthwithout destructing the pavement structure. The principle used in the project was to usehigh-accuracy laser sensors to measure the crack opening geometric including crackwidth, crack edge slopes, and measurable crack depth. With these obtained data and aneural network model developed in the project, the depth of the crack can be statisticallyestimated. Based on the evaluation results, it was found that the system developed in theproject can detect pavement crack depth with a statistically reliable accuracy.To detect crack depth, two steps are needed: crack identification and crack depthestimation. At the early stage of the project, several approaches were proposed. Theseproposed approaches included static ultrasonic method, dynamic ultrasonic sensormethod, radar sensor method, ground penetrating radar method, and so on. Based onpreliminary laboratory experiments and literature search and review, it was concludedthat these methods were not able to detect crack depth with a statistically reliableaccuracy or practical applicability. However, it was found that a combination of high-accuracy laser sensors and estimating models could produce satisfactory results.To dynamically identify a crack, two laser sensors were used to minimize the detectionerrors. An algorithm called Partial Cross Correlation (PCC) was developed to enhancethe crack detection ability. This algorithm was evaluated through field experiments andproven that the detection performance of PCC was much better than the approaches usedin past research studies for identifying pavement cracks.To automatically measure the crack opening geometric, one of the key elements is themeasurement of longitudinal displacement of the measuring system. A distance sensorwas used to measure the longitudinal displacement. However, field tests found that thesampling rate of the distance sensor had certain effect on the measurement accuracy.Thus, a scan-rate-effect-canceling model was developed based on field experimental dataand modeling results. With the model, the accuracy of the longitudinal displacementmeasurement was significantly improved.With the obtained crack opening geometric characteristics and the information on thepavement section such as the average daily traffic, pavement life cycle, pavement age,and other pavement related information, a neural network model was used to estimate thecrack depth. The database used for the neural network model development wascomprised of two parts: one was the distance sensor reading including the geometriccharacteristics of the crack; the other included pavement related variables. The crackinformation data were obtained from 95 pavement sections with the system developed inthe project and a static ultrasonic measuring system which can statically measure crackdepth with a reliable accuracy. The pavement related information data were obtainedfrom a database provided by the FDOT. In the model development, different networkarchitectures and different training algorithms were investigated and tested. An optimalarchitecture was determined based on the tests of different model architectures.The system was implemented and installed in a manually operated push-car. This systemnow can measure crack depth at a walking speed. The current system consists of twohigh-accuracy laser sensors, a longitudinal distance measuring sensor, a portablecomputer with an interface to communicate with the sensors, and a comprehensive modelsoftware used to estimate pavement crack depth using the readings from these sensorsand the pavement section related information. The system can only be used to measuretransverse crack depth since the developed neural network model was based on field dataof transverse cracks. To measure longitudinal crack depth, longitudinal crack data areneeded for modeling purpose.Although the system is now operated at walking speed, it can be operated in a muchfaster speed close to 55 mph if adequate modifications are implemented to increase thesampling interval and the sampling rate by using a more powerful computer such as anindustry computer.