Eine Plattform für die Wissenschaft: Bauingenieurwesen, Architektur und Urbanistik
Eine Plattform für die Wissenschaft: Bauingenieurwesen, Architektur und Urbanistik
In North American, aging water infrastructure continually challenges communities’ safety. An efficient infrastructure management system is thus required to manage deteriorating pipeline systems. However, there are several challenges in the water infrastructure management. For instance, water systems are often buried underground, making regular inspection and asset condition evaluation difficult. Furthermore, pipe working environment is always complex and not well-understood. In this thesis, data-driven approaches were used to address these challenges. The first study examines the relationship between soil properties and cast iron water main deterioration condition. The effects of soil properties on pipe failure were visualized and analyzed. Additionally, a stacking ensemble based method was proposed to overcome the drawbacks of the individual models by optimally combining the predictive results from multiple learners. Using soil property data, a single-model and an ensemble-model were developed to predict the pipe condition. The prediction results demonstrated that the proposed ensemble method outperforms the existing single models. Next, to investigate the structural response of pipelines to varying soil movements, a machine learning based framework was proposed to predict pipe deformations. The critical predictors contributing to the pipe deformations were first identified by random forest-recursive feature elimination algorithm. Super learning based methods were then employed to predict pipe deformations considering the selected feature subsets. Based on the prediction performance, the scalability and superiority of super learning was validated. Finally, a novel risk analysis approach was investigated by developing both condition rating and consequence models. For condition evaluation, multiple regression analyses were conducted for pipe structural and operational condition rating, respectively. A geographical information system based method was then applied to determine the multi-variant weighting system criteria for economic, operational, environmental, and social impacts. Finally, a risk matrix was used to integrate the results of condition grades and failure consequence scores, allowing the high-risk areas to be identified. With the data-driven methodology employed in this research, a cloud-based infrastructure management system which can accommodate the data-driven analysis, was designed to support the decision making for pipe system inspection, maintenance, and rehabilitation. ; Applied Science, Faculty of ; Electrical and Computer Engineering, Department of ; Graduate
In North American, aging water infrastructure continually challenges communities’ safety. An efficient infrastructure management system is thus required to manage deteriorating pipeline systems. However, there are several challenges in the water infrastructure management. For instance, water systems are often buried underground, making regular inspection and asset condition evaluation difficult. Furthermore, pipe working environment is always complex and not well-understood. In this thesis, data-driven approaches were used to address these challenges. The first study examines the relationship between soil properties and cast iron water main deterioration condition. The effects of soil properties on pipe failure were visualized and analyzed. Additionally, a stacking ensemble based method was proposed to overcome the drawbacks of the individual models by optimally combining the predictive results from multiple learners. Using soil property data, a single-model and an ensemble-model were developed to predict the pipe condition. The prediction results demonstrated that the proposed ensemble method outperforms the existing single models. Next, to investigate the structural response of pipelines to varying soil movements, a machine learning based framework was proposed to predict pipe deformations. The critical predictors contributing to the pipe deformations were first identified by random forest-recursive feature elimination algorithm. Super learning based methods were then employed to predict pipe deformations considering the selected feature subsets. Based on the prediction performance, the scalability and superiority of super learning was validated. Finally, a novel risk analysis approach was investigated by developing both condition rating and consequence models. For condition evaluation, multiple regression analyses were conducted for pipe structural and operational condition rating, respectively. A geographical information system based method was then applied to determine the multi-variant weighting system criteria for economic, operational, environmental, and social impacts. Finally, a risk matrix was used to integrate the results of condition grades and failure consequence scores, allowing the high-risk areas to be identified. With the data-driven methodology employed in this research, a cloud-based infrastructure management system which can accommodate the data-driven analysis, was designed to support the decision making for pipe system inspection, maintenance, and rehabilitation. ; Applied Science, Faculty of ; Electrical and Computer Engineering, Department of ; Graduate
Network Analytics for Infrastructure Asset Management Systemic Risk Assessment
| ASCE | 2022
Condition Assessment of Water Supply Infrastructure
| Online Contents | 1997
Predictive analytics in facilities management
| Emerald Group Publishing | 2019
Subway Infrastructure Condition Assessment
| ASCE | 2015
Subway Infrastructure Condition Assessment
| Online Contents | 2015