|
Title |
Development of an Integrated Short- and Long-Term Deterioration Prediction Model for Bridge Substructures Using LightGBM and Weibull Distribution
|
|
Authors |
양지수(Ji-Su Yang) ; 김우혁(Woo-Hyeok Kim) ; 민근형(Geun-Hyeong Min) ; 김우석(Woo Seok Kim) |
|
DOI |
https://doi.org/10.4334/JKCI.2026.38.5.693 |
|
Keywords |
머신러닝; 교량 하부구조 열화; 결함도지수(DI); 생존분석; 통합 예측 모델 machine learning; bridge substructure deterioration; defect index (DI); survival analysis; integrated prediction model |
|
Abstract |
Bridge substructures have received limited attention in deterioration prediction research, with existing models facing two key limitations: non-physical deterioration patterns over long-term extrapolation intervals, and an inability to reflect structure-specific initial conditions. We developed an integrated short- and long-term deterioration prediction framework for bridge substructures that incorporates environmental and regional characteristics. Eight machine learning and deep learning algorithms were applied for short-term prediction (1-5 years), while four survival analysis models were employed for long-term prediction (beyond 5 years). The key contribution lies in linking short-term prediction results to the survival analysis models as bridge-specific initial conditions, enabling structure-level deterioration trajectory estimation. Light Gradient Boosting Machine (mean absolute error: 0.0276; mean absolute percentage error: 11.09 %) and the Weibull model (concordance index: 0.9916; integrated Brier score: 0.0037) were identified as optimal models, with generalization performance validated using independent field inspection data.
|