The University of Sheffield
Browse

CCWI2017: F47 'ONE-CLASS SVM – LEAK DETECTION IN WATER DISTRIBUTION SYSTEMS'

Download (308.28 kB)
journal contribution
posted on 2017-09-01, 15:06 authored by Roya Cody, Sriram Narasimhan, Bryan Tolson
Acoustic leak detection in water distributions systems has been reviewed and validated for decades in various laboratory and field settings. However, the existing systems rely heavily on detailed knowledge of the pipe system, an assumption of ideal conditions, as well as direct access to infrastructure pipelines. This paper presents an experimental investigation that addresses the need of minimally invasive water distribution monitoring in cold climates. Monitoring in cold climates is achieved with a permanent dry barrel hydrant mounted passive sensor system. The sensor system sits within the water column while still being accessible via the hydrant. Lab tests utilize a retrofitted hydrant and pipe system. Experiments show the effectiveness of using fire hydrant mounted sensors in leak detection. Acoustic signals due to simulated leaks are measured, and a one-class support vector machine (OCSVM) classification methodology is applied. Results showed that a simulated leak can be detected with a 97% classification accuracy.

History

Usage metrics

    Department of Civil and Structural Engineering

    Licence

    Exports

    RefWorks
    BibTeX
    Ref. manager
    Endnote
    DataCite
    NLM
    DC