A Bayesian model updating approach for detection-related problems in water distribution networks. (May 2019)

• Record Type: Journal Article
• Authors: Jensen, H.A.; Jerez, D.J.
• Is Part Of: Reliability engineering & system safety. Volume 185(2019)
• Page Start: 100
• Page End: 112
• Publication Date: 2019-05
• Subjects: Bayesian model updating -- Large scale water distribution systems -- Leakage detection -- Model class selection; Reliability (Engineering) -- Periodicals; System safety -- Periodicals; Industrial safety -- Periodicals; Fiabilité -- Périodiques; Sécurité des systèmes -- Périodiques; Sécurité du travail -- Périodiques; 620.00452
• DOI: 10.1016/j.ress.2018.12.014
• Languages: English
• ISSNs: 0951-8320

Abstract

Highlights: A Bayesian model updating approach for complex utility networks has been presented. Transitional Markov chain Monte Carlo facilitates model parameter estimation. The identification scheme is somewhat robust to model and measurement errors. The proposed framework can handle networks with thousands of nodes and links. Leakage events are properly identified. Abstract: A Bayesian system identification technique is proposed for detection-related problems in the context of water distribution networks. The identification process is combined with a hydraulic simulation model for performing steady and unsteady analyses. In particular, leakage detection in water pipe networks is examined. A number of hydraulic model classes are defined as potential leakage events. Based on information from flow rates in the pipes, the implemented simulation-based Bayesian model updating technique provides estimates of the most probable leakage scenarios. Such scenarios correspond to the model classes that maximize their posterior probabilities. The effectiveness of the proposed framework is illustrated by applying the leakage detection approach to a complex water distribution system. Issues such as the effect of model and measurement errors, number of flow tests at each monitoring location, and sensor number and location on the performance of the detection methodology are addressed.