AI-Assisted Attribute-Based Screening Framework for Ground-Subsidence Risk Using Underground Utility Information

Ground subsidence in urban environments is closely linked to the deterioration of underground utilities. Conventional NDT techniques are effective for detecting structural defects but are limited by accessibility and inspection coverage. This study introduces an AI-assisted Virtual NDT screening approach that predicts subsidence risk using existing utility attribute data, including material, diameter, installation year, burial depth, and maintenance history. Machine-learning models were trained using standard validation metrics, and results showed that attribute-based prediction offers a meaningful early-warning capability. The proposed framework supports efficient allocation of field NDT resources by narrowing inspection targets.