Innovative Practical Crack Propagation Measurement of Asphalt Concrete Specimens
Advisor: Professor Imad L. Al-Qadi
Abstract
Approximately 95 percent of paved roads in the United States are surfaced with asphalt concrete (AC).
Cracking is a common failure mode in pavements. The cracking potential of AC significantly affects
pavement durability and serviceability. Numerous tests have been developed and employed to predict AC
cracking potential. Accurate crack measurement during testing is crucial. However, there is a lack of an
efficient and accurate crack propagation measurement technique.
This dissertation aimed to develop an automated crack measurement technique that can efficiently
deliver accurate results for AC cracking tests. To achieve this goal, a generalized crack detection
framework was developed using fundamental fracture mechanics theory and digital image correlation
(DIC). Multi-seed incremental reliability-guided DIC analysis was proposed to solve the decorrelation issue
due to large deformation and discontinuities. A robust method was developed to detect cracks based on
displacement fields. It uses critical crack tip opening displacement (š¯›æō€Æ– ) to define the onset of cleavage
fracture. The proposed threshold š¯›æō€Æ– has a physical meaning and can be easily determined from DIC
measurement.
To enable automated crack propagation measurement in AC cracking tests, a deep neural network,
CrackPropNet, was trained. An image library representing the diversified cracking behavior of AC was built
for supervised learning. CrackPropNet could accurately and efficiently measure crack propagation with an
F-1 of 0.781 at a running speed of 26 frame-per-second. The model showed promising generalization on
fundamentally different images.
An accurate measurement can only be achieved when the cameraā€™s principal axis is perpendicular to the
specimen surface. However, this requirement may not be met during testing due to device constraints. A
simple and reliable method was proposed to correct errors induced by non-perpendicularity. The method
is based on image feature matching and rectification. A theoretical analysis was performed to quantify
the effect of a non-perpendicular camera alignment on measurement accuracy. The proposed method
showed satisfactory accuracy in compensating errors induced by non-perpendicularity. It was verified as
a valid approach assisting the CrackPropNet in measuring crack propagation with a non-perpendicular
camera alignment.
Engineers and practitioners could use smartphones to monitor crack development under complex imaging
environments as a part of AC mix design and quality control/quality assurance. In addition, this technique
may assist researchers in characterizing cracking phenomena, evaluating AC cracking potential, validating
test protocols, and verifying theoretical models.