Machine learning-based prediction of underground utility counts using electrical resistance numerical data

Understanding subsurface conditions is essential for mitigating unexpected hazards during excavation, particularly in areas with underground utilities. Electrical resistance values play a crucial role for predicting these conditions. This study develops a non-destructive, cost-effective framework to predict the number of buried pipelines using machine learning applied to numerical electrical resistance data. The resistance data was generated by applying a numerical electrical resistance model, developed using generalized mesh techniques, based on electrode and structural geometric parameters. A total of 87,572 data samples, comprising 56 electrodes across 667 cases and 90 electrodes across 558 cases, were used. Various machine learning techniques, including Support Vector Machine, Random Forest, and Extreme Gradient Boosting, were employed to classify underground utility counts. Additionally, a deep learning method, specifically Convolutional Neural Network, transformed the resistivity data into a 2D matrix format for analysis. The results indicate that the data provides sufficient information to accurately determine the number of buried pipelines, demonstrating the potential of these models for underground utility prediction. This work integrates numerical simulations with machine learning to develop a model capable of underground utility prediction. Given the significant challenges associated with collecting and processing real-world data for such applications, utilizing simulation data is essential to demonstrate the feasibility of these models.