Logo image
Back
Noninvasive identification of lead water service lines using hammer-induced stress waves
Journal article   Open access   Peer reviewed

Noninvasive identification of lead water service lines using hammer-induced stress waves

K. I. M. Iqbal, Kurt Sjoblom, Charles N. Haas, Ivan Bartoli and Arvin Ebrahimkhanlou
Smart materials and structures, v 35(7), 075007
01 Jul 2026
DOI: https://doi.org/10.1088/1361-665X/ae7cfb
url
https://doi.org/10.1088/1361-665X/ae7cfbView
Published, Version of Record (VoR) Open

Abstract

Instruments & Instrumentation Materials Science, Multidisciplinary Science & Technology Materials Science Technology
Identifying lead water service lines (LSLs) without excavation remains a significant challenge for utilities across the United States, as existing noninvasive methods are limited in their practical use and scalability. This study introduces a stress wave-based technique for detecting LSLs. This method is applicable to both the utility and customer-owned sides of service lines (SLs) without requiring access into homes, making it suitable for large-scale field deployment. Stress waves are generated by striking an extension rod at the curb-stop valve, while surface-mounted accelerometers capture the resulting signals. The method was field-tested on more than 300 SLs across 17 U.S. cities. All recorded signals were processed using wavelet transform techniques to extract robust time-frequency features. These features were then used as inputs to three deep learning models: 2D-convolutional neural network (CNN), CNN-BiLSTM, and a TimesFormer-based transformer. Each model was trained to perform binary classification between lead and non-lead SL materials. Among the models evaluated, the TimeSformer demonstrated the highest performance, achieving an accuracy of 80% on a held-out test set and 82% in a blind-tested set of 66 SLs from new locations. These results suggest that the stress wave technology proposed in this study, combined with wavelet-based signal processing and deep learning models offers a promising noninvasive solution for identifying LSLs under diverse field conditions.

Metrics

1 Record Views

Details

Logo image