A Geometry‐Aware Multimodal Fusion Transformer for Pipeline Leak Localization From Time‐Lapse Resistivity and Permittivity Sections

Abstract Restricted survey access at industrial sites complicates the localization of buried wastewater pipeline leaks. We develop a geometry‐aware spatiotemporal multimodal fusion transformer (STMFT) that combines resistivity and permittivity features at matching locations and times, models temporal changes, and integrates information across survey directions. Candidate coordinates and distances to monitoring sections guide feature aggregation for evaluating possible leak locations. Tests use a site‐informed synthetic benchmark comprising complete 180‐day time‐lapse sequences of transport‐derived property sections with prescribed anisotropic smoothing and noise. For source locations excluded from training under a fixed site configuration, STMFT selected the correct cell among 154 pipeline cells on a 1‐m grid in 90.68% of samples and included it among the three highest‐ranked candidates in 99.69%. Joint‐property input and explicit distance bias improved identification over single‐property and no‐distance‐bias configurations, respectively. This approach illustrates how relationships among observations and candidate‐source geometry can guide multimodal fusion for pipeline inspection prioritization.

Authors

Institutions

Publication Details

Journal
Geophysical Research Letters
Published
2026-11-14
DOI
https://doi.org/10.1029/2026gl127099
Primary Topic
Geophysical and Geoelectrical Methods
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

A Geometry‐Aware Multimodal Fusion Transformer for Pipeline Leak Localization From Time‐Lapse Resistivity and Permittivity Sections

Zhenxue Dai, Liang Shen, Jichun Wu, Mohamad Reza Soltanian et al.
Geophysical Research Letters
Geophysical and Geoelectrical Methods
article

A Geometry‐Aware Multimodal Fusion Transformer for Pipeline Leak Localization From Time‐Lapse Resistivity and Permittivity Sections

Zhenxue Dai, Liang Shen, Jichun Wu, Mohamad Reza Soltanian, Chuanjun Zhan, Kenneth C. Carroll, Xiaoying Zhang, Lizhi Du, Yuancang Li, Junhao Zheng
article en

Abstract

Abstract Restricted survey access at industrial sites complicates the localization of buried wastewater pipeline leaks. We develop a geometry‐aware spatiotemporal multimodal fusion transformer (STMFT) that combines resistivity and permittivity features at matching locations and times, models temporal changes, and integrates information across survey directions. Candidate coordinates and distances to monitoring sections guide feature aggregation for evaluating possible leak locations. Tests use a site‐informed synthetic benchmark comprising complete 180‐day time‐lapse sequences of transport‐derived property sections with prescribed anisotropic smoothing and noise. For source locations excluded from training under a fixed site configuration, STMFT selected the correct cell among 154 pipeline cells on a 1‐m grid in 90.68% of samples and included it among the three highest‐ranked candidates in 99.69%. Joint‐property input and explicit distance bias improved identification over single‐property and no‐distance‐bias configurations, respectively. This approach illustrates how relationships among observations and candidate‐source geometry can guide multimodal fusion for pipeline inspection prioritization.

Geophysical Research LettersVol. 53(22)
New Mexico State University (US), Sinopec (China) (CN), Jilin University (CN), State Key Laboratory on Integrated Optoelectronics (CN), SINOPEC Research Institute of Safety Engineering Co., Ltd. (CN), Qingdao University of Technology (CN), University of Cincinnati (US), Nanjing University (CN)
Geophysical and Geoelectrical Methods
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon