Leak detection in a water supply system using supervised machine learning models | |
| Author | May Thet Htar |
| Call Number | AIT Thesis no.WM-23-17 |
| Subject(s) | Machine learning--Mathematical models Water-supply--Vietnam |
| Note | A thesis submitted in partial fulfillment of the requirements for the degree of Master of Engineering in Water Engineering and Management |
| Publisher | Asian Institute of Technology |
| Abstract | In Vietnam, high water loss rate (averagely 23% of total water supply) becomes additional issues for water demand. One of the major cities of Vietnam, Hai Phong city’s water supply system need to adapt and increase efficiency in accordance with demographic growth and urbanization process and non-revenue water loss is 33% in 2012. Therefore, leakage in the water supply system of Do Son District, Hai Phong City is detected by three machine learning models and one artificial neural network. The main focus of this study is to detect leakage in the water supply network by changing pipe materials that if changing pipe materials can affect the performance of machine learning models. Roughness coefficient is the significant factor for changing old pipe materials (U-PVC) to new pipe materials (M-PVC). Leakages in the study network are simulated by changing emitter exponent and emitter coefficient values and for this, three scenarios will be implemented by three previous research papers and not only individual node scenarios but also combined leaking scenarios are discussed. First of all, non-leaking scenario for EPANET hydraulic model is simulated and datasets are taken from twelve monitoring nodes. After that, individual leaking scenarios are simulated. There are four leaking nodes in the system that individual nodes are simulated with several leaking scenarios and this is for first objective. For second objective, roughness coefficient of pipes in specified zones are updated from 120 to 140 and then the procedures are same as previous objective; simulate non-leaking scenarios and after that, leaking scenarios are simulated for individual nodes. For final objective, leaking nodes in the systems are increasing one by one that there will be four scenarios. Results show that if small emitter coefficient values are used, there is no much pressure differences for leaking scenario and performances of machine learning models are quite low. For higher emitter coefficient values with varying emitter exponent values will show better pressure difference compared with scenario one and therefore, the accuracy become higher. While default emitter exponent values and varying emitter coefficient values are used for scenario three, it shows the best performance among three scenarios. From point of changing pipe materials, accuracy of new pipe system is higher than accuracy of old pipe system. In all machine learning models, ANN shows the best accuracy. |
| Year | 2023 |
| Type | Thesis |
| School | School of Engineering and Technology |
| Department | Department of Civil and Infrastucture Engineering (DCIE) |
| Academic Program/FoS | Water Engineering and Management (WM) |
| Chairperson(s) | Ho Huu Loc |
| Examination Committee(s) | Babel, Mukand Singh;Shrestha, Sangam;Shanmugam, Mohana Sundaram |
| Scholarship Donor(s) | Thai Pipe Scholarship;Asian Institute of Technology Scholarships |
| Degree | Thesis (M. Eng.) - Asian Institute of Technology, 2023 |