Underground Utility Mapping

AI-Powered Probabilistic Framework for Automated Subsurface Infrastructure Detection Using Fuzzy Logic and Computer Vision

Overview

Accidental strikes of underground utilities during construction and excavation activities remain a critical safety and economic concern. This project presents a novel probabilistic framework that fuses AI-generated utility maps with existing as-built records to create confidence-aware underground infrastructure maps. The approach addresses the fundamental challenges of data interpretation costs, large-scale mapping, and the inherent inaccuracies in utility records.

Framework for probabilistic underground utility mapping using AI-generated initial maps and data fusion.

Motivation & Background

The Problem

Underground utility infrastructure represents a hidden and critical component of modern society. However:

  • Utility as-built plans are notoriously inaccurate and often become outdated as infrastructure changes
  • Accidental utility strikes cause serious injuries, fatalities, and substantial economic losses
  • Traditional investigation methods (GPR, electromagnetic surveys) are expensive and labor-intensive
  • The One-Call system often yields incomplete information, excluding service lines and out-of-service utilities

Our Solution

This research introduces a probabilistic data fusion framework that:

  1. Automatically generates initial utility maps by detecting aboveground appurtenances (manholes, meters, valve covers) and applying hierarchical inference rules
  2. Fuses AI-generated maps with as-built records to create probabilistic utility maps with confidence bands
  3. Identifies low-confidence zones for targeted ground-truth investigations
  4. Dynamically updates as new information becomes available

Methodology

AI-Generated Utility Map Creation

The framework employs fuzzy logic approximate reasoning to infer underground utility locations from visual observations of aboveground features:

Utility Classes

  • Water
  • Electric
  • Sanitary
  • Stormwater
  • Gas
  • Telecom

Detected Appurtenances

  • Gas Meters, Water Meters
  • Valve Covers, Fire Hydrants
  • Storm Traps/Outfalls, Cleanouts
  • Electric Meters, Transformers
  • Control Boxes, Manholes
  • Pedestals, Light Poles

Hierarchical Connection Rules

Utility connections are inferred using deterministic logic rules organized by:

Rule Category Description
Proximity Rules Connect close appurtenances of similar utility class
Roadway Right-of-Way Establish utility connection orientation
Downstream Endpoints Determine service line direction from main utility

Probabilistic Data Fusion

The fusion process combines multiple data sources using probability distributions:

\[P(utility|x,y) = \sum_{i=1}^{n} w_i \cdot N(\mu_i, \sigma_i^2)\]

Where:

  • \(w_i\) = weight assigned to data source \(i\) (expertise, recency, accuracy)
  • \(\mu_i\) = utility location from source \(i\)
  • \(\sigma_i\) = standard deviation reflecting confidence level

Confidence Factors

Factor Impact on Confidence
Data Source Age Older sources → Lower weight
Ground-Truth Data Excavation/GPR verified → High weight
As-Built vs As-Planned As-built more reliable
Sensor Limitations GPR/EM interference → Higher uncertainty
Construction Variance Non-compliance → Increased σ

Web Application: U-Map

A Django-based web application was developed to operationalize the framework, enabling users to interactively map and manage underground utilities.

Key Features

🗺️ Interactive Mapping

Google Maps integration with region-of-interest selection and asset placement on satellite imagery

📍 Asset Detection

Mark and catalog aboveground appurtenances with utility class classification

🤖 AI Map Generation

Apply hierarchical rules to automatically generate utility connection maps

Demonstration Video

Watch the presentation explaining the web application workflow and system architecture:

Video presentation of the U-Map web application flowchart and system architecture.

Case Study: Ames, Iowa

The framework was validated using field data from Iowa State University’s Research Park:

  • Site selected for low traffic volume and availability of utility as-built maps
  • As-built map contained water, storm sewer, and sanitary utilities
  • AI-generated map created from manually surveyed aboveground appurtenances

Results

The probabilistic fusion revealed:

  1. High-confidence zones where AI-generated and as-built maps aligned
  2. Low-confidence areas requiring targeted GPR investigation
  3. Significant cost savings by focusing detailed investigations on uncertain regions

Impact & Applications

Application Benefit
Construction Safety Reduce accidental utility strikes during excavation
Infrastructure Planning Rapid preliminary utility mapping for large projects
Asset Management Dynamic updating of utility records
Cost Reduction Focus expensive investigations on low-confidence zones

Publication

This research was published in Sensors (MDPI):

Oguntoye, K.S., Laflamme, S., Sturgill, R., Martinez, D.A.S., Eisenmann, D.J., & Kimber, A. (2024). Probabilistic Method to Fuse Artificial Intelligence-Generated Underground Utility Mapping. Sensors, 24(11), 3559. https://doi.org/10.3390/s24113559

Funding: Iowa Energy Center under Grant #21-IEC-006

Source Code

GitHub Repository

Future Work

  • Automated appurtenance detection using computer vision (YOLO, CNN)
  • Drone-based data collection for rapid surveying
  • 3D probabilistic mapping including utility depth estimation
  • Cross-utility correlation integration
  • Machine learning weight optimization from historical data

Contact

For questions about this research or collaboration opportunities: