| dc.description.abstract | Asphalt pavement’s density measurement is an important step in the quality control of asphalt road construction. It is usually achieved by applying the coring method (CM), nuclear density gauge (NDG), and electromagnetic density gauge (EDG). CM is the most accurate method, but it is a destructive method because the pavement is damaged when the cores are taken. NDG and EDG are nondestructive methods with high efficiency, but their measurement accuracy is poorer than that of CM. An EDG commonly used in density measurement is named pavement quality indicator (PQI). A novel method named density profiling system (DPS) is also based on the potential EDG. However, it was not applied to this research because more tests are required to verify its accuracy. This paper presents an approach to improve the accuracy of the nondestructive methods with NDG and PQI. It is based on the artificial neural network (ANN), which processes the raw data got from NDG and PQI and produces the predicted asphalt density as the output. The density measured in CM was used as the target density and the error between ANN-predicted density and target density was computed. To minimize this error, various ANN architectures and learning algorithms were tried in the ANN training process. Each established ANN model makes a substantial improvement in the performance of NDG or PQI in asphalt density measurement. This research was initiated by Fulton Hogan (FH) Limited, a large road construction and maintenance company in New Zealand. FH lab teams are responsible for asphalt road density measurement in FH’s road projects. One of the main method they use is to measure the densities of the cores taken from asphalt pavements (coring method). It is quite accurate but destructive and very time-consuming. They also use NDG or PQI, which are highly efficient nondestructive measurement devices. However, their measurement accuracy is poorer than that of the coring method. FH lab teams wanted to have a new density measurement method that is both accurate and efficient. An ANN-based approach is presented in this paper to address the issues faced by the FH lab teams. Densities collected with coring methods, NDG, and PQI were used to train and validate the ANN models. The results from the ANNs show substantial improvements of the measurement accuracy and efficiency. The proposed approach has been presented to the FH lab teams, who are impressed with its performance and plan to implement it in their projects. | |
