Integrating Machine Learning and GIS for Sewer Condition Assessment and Visualization
Doctoral thesis
Permanent lenke
https://hdl.handle.net/11250/3150381Utgivelsesdato
2024Metadata
Vis full innførselSamlinger
Sammendrag
Sewer network, including wastewater and stormwater pipelines, is among the critical infrastructures that can be seen as one kind of national asset. These asset-related issues can cause serious consequences affecting people and the environment. As with other infrastructures, the sewer network deteriorates over time, and continuous adjustments are required. Consequently, rehabilitation and maintenance activities are needed to ensure its engineered functions work properly, reduce risks, optimize performance, and minimize costs.
One of the effective ways of the predictive maintenance strategy is to estimate the condition of sewer pipelines. In general, the sewer condition assessment model is a tool that provides decision-makers valuable information on not only the current state but also the future state of the sewers and support for prioritization of inspection, reparation, or renewal of sewer pipes. However, the change in sewer condition significantly depends on input factors. Therefore, the quality of sewer condition models is influenced both by the input factors used and the methods employed. Although many different methods and techniques have been proposed and implemented for sewer condition analysis, no agreement has been reached regarding the best method for sewer condition assessment.
In this thesis, a methodology for modeling sewer conditions has been successfully developed and applied for Ålesund city, Norway by employing state-of-the-art machine learning (ML) and deep learning. In addition, various feature selection methods have been investigated to assess the importance of physical factors and environmental factors. In this thesis, Geographic Information System (GIS) was used as a main tool to analyze, store, and visualize the results. The primary purpose of the developed methodology is to partially support local water agencies to control and operate the wastewater/stormwater system more effectively and partly optimize predictive maintenance strategies.
The condition of sewer pipes can be defined based on damage score (regression problem) or damage class (classification problem). The performance of sewer condition assessment models using these outputs has not been assessed and compared. This thesis addresses the above statement by developing ML models for sewer condition assessment. The performance of the models was compared using the popular assessment criteria for the regression problem and the
classification problem.
The thesis evaluates the potential application of ML algorithms for predicting the damage scores of sewer pipelines. The performance of the developed models was compared using the popular statistical metrics for the regression problem. By transforming damage scores to damage classes, the prediction performance of ML models has been improved significantly. The assessment criteria for the classification problem between models are more stable compared to regression models. The efficiency of hybrid ML models in predicting sewer conditions was tested. The results show that hybrid models have better performance, even with a multiclassification problem. Finally, a sewer condition assessment map was prepared that provides useful information for supporting predictive maintenance strategies.
The thesis also introduces an integrated framework as a combination between GIS, 3D-creation platform, augmented reality techniques, and ML algorithms for the dynamic visualization of the condition of sewer networks.
Består av
Paper 1: Nguyen Van, Lam; Razak, Seidu. Application of Regression-Based Machine Learning Algorithms in Sewer Condition Assessment for Ålesund City, Norway. Water 2022 ;Volum 14.(24) s. – Published by MDPI. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license. Available at: https://doi.org/10.3390/w14243993Paper 2: Nguyen Van, Lam; Tien Bui, Dieu; Seidu, Razak. Comparison of Machine Learning Techniques for Structural Condition Assessment of Sewer Network. IEEE Access 2022 ;Volum 10. s. 124238-124258. Published by IEEE. Open Access. This work is licensed under a Creative Commons Attribution 4.0 License CC BY. Available at: http://dx.doi.org/10.1109/ACCESS.2022.3222823
Paper 3: Nguyen Van, Lam; Razak, Seidu. Predicting sewer structural condition using hybrid machine learning algorithms. Urban Water Journal 2023 ;Volum 20.(7) s. 882-896. Published by Informa UK Limited, trading as Taylor & Francis Group. This is an Open Access article distributed under the terms of the Creative Commons Attribution-NonCommercial-NoDerivatives License CC BY NC-ND. Available at: https://doi.org/10.1080/1573062X.2023.2217430
Paper 4: Nguyen Van, Lam; Tien Bui, Dieu; Seidu, Razak. Utilization of Augmented Reality Technique for Sewer Condition Visualization. Water 2023 ;Volum 15.(24) s. Published by MDPI. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license. Available at: http://dx.doi.org/10.3390/w15244232