Non-Invasive Detection of Lead Connection Pipe with Machine Learning Model

Area of Science:

• Environmental Engineering
• Data Science
• Materials Science

Background:

• The U.S. Environmental Protection Agency mandates lead service line replacement within 10 years.
• Accurate identification of buried lead pipes is a significant challenge for water utilities.
• Current detection methods are often costly, disruptive, or unreliable due to environmental interference.

Purpose of the Study:

• To develop a noninvasive, efficient, and cost-effective method for detecting buried lead pipes.
• To integrate physics-based modeling with machine learning for improved pipe material classification.
• To provide a scalable solution for utilities to comply with regulatory requirements.

Main Methods:

• Developed a physics-based finite element analysis (FEA) surrogate model to simulate the dynamic behavior of buried pipes.
• Incorporated realistic loading conditions, such as stop-valve openings, into the FEA model.
• Generated over 13,000 synthetic acceleration signals, simulating real-world noise and signal limitations.
• Trained seven machine learning (ML) models, including K-nearest neighbor (KNN) and Extreme Gradient Boosting (XGBoost), on the simulated data.

Main Results:

• The integrated FEA-ML framework achieved high accuracy in classifying pipe materials.
• K-nearest neighbor (KNN) and Extreme Gradient Boosting (XGBoost) models demonstrated 99.9% classification accuracy.
• The method proved effective in identifying pipe acceleration under various conditions and noise levels.

Conclusions:

• The developed noninvasive approach offers a scalable and cost-effective solution for lead pipe detection.
• This framework enables utilities to efficiently locate and replace lead pipes, ensuring regulatory compliance.
• The method minimizes operational disruptions and resource expenditure compared to traditional techniques.