Sharpe, Kyal (2022) Development of Smart Sewer Pipeline Fault Detection Method using CCTV Inspection Data and Deep Learning. [USQ Project]
|
Text (Project)
Sharpe_K_Nguyen_Redacted.pdf Download (2MB) | Preview |
Abstract
Sewer networks are complex public infrastructure assets designed to deliver wastewater from property connection points to wastewater treatment facilities. Like all assets, sewer networks require ongoing inspection and rehabilitation to maintain an acceptable level of service which often requires generous resource allocation due to the complex infrastructure. For many municipalities and service authorities, large segments of existing sewer networks are approaching end of life, increasing the demand for maintenance and resources. Maintenance and rehabilitation of sewer networks are primarily founded on investigation programs that determine the existing asset condition, and consequently, the rehabilitation method. Current practice generally involves deploying a robotic closed circuit television video (CCTV) camera to inspect the network; this is an efficient process in isolation; however it commonly requires an extensive manual review process to determine faults within the network.
In this research, an automated fault detection model is developed to review and analyse CCTV inspection footage, and then locate and categorise faults within the data. The study utilises emerging deep transfer learning technology to locate and categorise abnormalities in the input data, based on a predetermined calibration dataset. The smart sewer detection model is adapted from the YOLOv2 object detection framework; twelve common convolutional neural networks were evaluated to determine the optimal feature extraction network, with Res-Net 101 determined to be the preferred network. Performance evaluation of the model included a mAP of 89.3%, and detection speed of 46.3fps which exceeds real-time capability. The success of this project will provide value in improved efficiencies, reliable and consistent fault detection and overall economic benefit to industry, and other users who may employ the findings of this research. Overall, the smart sewer detection model demonstrates capacity to significantly decrease the time required for data review and improve the overall accuracy of decision-making process by removing human error.
|
Statistics for this ePrint Item |
| Item Type: | USQ Project |
|---|---|
| Item Status: | Live Archive |
| Faculty/School / Institute/Centre: | Current – Faculty of Health, Engineering and Sciences - School of Engineering (1 Jan 2022 -) |
| Supervisors: | Nguyen, Andy |
| Qualification: | Bachelor of Engineering (Honours) (Civil) |
| Date Deposited: | 20 Jun 2023 00:04 |
| Last Modified: | 20 Jun 2023 00:04 |
| Uncontrolled Keywords: | Sewer Pipeline Faults; Deep Learning; Object Detection |
| URI: | https://sear.unisq.edu.au/id/eprint/51890 |
Actions (login required)
 |
Archive Repository Staff Only |
