Back to search

University of British Columbia

Enhancing sewer asset management using machine learning algorithms

Abstract

dc:description

Closed-circuit television (CCTV) is widely employed for assessing defects in sewers. This thesis: i) develops a number of deep convolutional neural network (CNN)-based image classification models for identifying sewer defects in CCTV videos, ii) develops a number of Machine Learning (ML)-based models for predicting the likely occurrence of specific defects in sewer networks, as well as identifying the factors that contribute to these defects, and iii) analyses the model performances with data from multiple utilities. The image classification models are trained and tested using images extracted from CCTV inspection videos for three utilities. They are analysed for serving two purposes: conducting a quality assurance and quality control (QA/QC) assessment of manual inspection reports of previously inspected videos, and identifying defects in images from videos that have not yet been manually inspected. Models are developed based on three pre-trained CNN backbone architectures: MobileNet V3 Large, ResNet50, and ResNet101; two model configurations: one- and two-stage models; and grouped classes. Two-stage models result in higher performance compared with alternative approaches. For QA/QC, on average, the ResNet models outperform MobileNet V3 Large where model accuracies for Utilities A, B, and C, are 0.95, 0.70, and 0.98, respectively. For recognizing defects in videos not previously viewed, the optimal model for each of the three utilities does not perform well. However, all models possess the ability to distinguish between non-defective and defective images with accuracies higher than 0.50, where the Utility A model achieves an accuracy of 0.80. Multiple decision tree-based ML models that predict the likely occurrence of infiltration and structural defects are developed using data from CCTV inspections coupled with additional pipe information and Geographical Information System files for two utilities. The performance of the models is assessed using the area under the Receiver Operating Characteristics (AUC-ROC) and Precision Recall (AUC-PR) curves. LightGBM-based models with cost-sensitive learning show the best performance overall when predicting infiltration and structural defects. The best performing model achieves an AUC-ROC of 0.79 and an AUC-PR of 0.62. For these utilities, an application of SHapley Additive exPlanations (SHAP) shows that the most important factors are “pipe location” and “pipe age”.

Degree

thesis:*
Name thesis:degree_name
Master of Applied Science - MASc
Level thesis:degree_level
master's
Discipline thesis:degree_discipline
Civil Engineering
Grantor dc:publisher
University of British Columbia
Year dc:date
2024

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Seng, Vannary
Contributors dc:contributor
  • Lence, Barbara J.

Rights

dc:rights
Statement dc:rights
  • Attribution-NonCommercial-NoDerivatives 4.0 International
Language dc:language
eng

Identifiers

dc:identifier.*
Handle dc:identifier
http://hdl.handle.net/2429/87751

Chain of custody

source
Harvested from
University of British Columbia
Base URL
circle.library.ubc.ca/oai/request
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
related terms
citation

Seng, Vannary. Enhancing sewer asset management using machine learning algorithms. master's thesis, University of British Columbia, 2024. http://hdl.handle.net/2429/87751