PDP 기반 xAI 접근법을 활용한 철근콘크리트 기둥의 데이터 기반 내폭 보강 효과 분석

초록

This study introduces an integrated framework for the rapid evaluation of retrofit levels (RL) in reinforced concrete (RC) columns subjected to blast loading, employing machine learning and explainable artificial intelligence (xAI) techniques. A multi-stage machine learning approach was developed to classify failure modes and predict retrofit levels. Partial dependence plot (PDP) analysis was subsequently applied to extract data-driven insights on retrofit effectiveness. The proposed framework comprises two major components: (1) blast performance assessment employing a multi-stage ML model for failure mode classification and RL prediction, and (2) PDP-based analysis for systematic evaluation of input variable effects. The RL prediction models were trained on a blast damage dataset and validated across three damage conditions (severe, moderate, and minor) for both flexural and shear failure modes. PDP-based analysis effectively identified feasible and infeasible reinforcement ranges for both longitudinal and transverse reinforcement ratios, providing clear guidance for optimal retrofit strategies. The proposed framework offers a practical tool for blast-resistant design that enables rapid and informed retrofit decisions. It supports efficient evaluation of retrofit demands and goal-oriented reinforcement planning based on column detailing and specific blast load scenarios.

키워드