PREDICTIVE MODELING OF BRIDGE LOAD CAPACITY USING MACHINE LEARNING AND REAL-TIME SENSOR DATA

Abstract

Bridge load capacity, defined as the maximum load a bridge can sustain without structural failure, is a critical determinant of infrastructure safety and operational reliability. Traditional assessment methods, including analytical modeling, static load testing, and empirical estimation, often fail to capture the dynamic, nonlinear, and time-dependent behavior of bridges subjected to evolving environmental and operational conditions. This study presents a predictive modeling framework that integrates machine learning (ML) techniques with real-time sensor data to enhance the accuracy and adaptability of load capacity prediction. The research utilized continuous data streams collected from strain gauges, accelerometers, displacement sensors, and fiber Bragg gratings, complemented by environmental and traffic datasets capturing temperature, humidity, axle loads, and vehicle volumes. Data preprocessing involved noise reduction, normalization, dimensionality reduction, and feature selection using principal component analysis (PCA) and LASSO regression. Multiple machine learning algorithms, Random Forest, Gradient Boosting, Support Vector Regression, and Deep Neural Networks, were developed and evaluated against traditional regression models. Descriptive and correlational analyses revealed that strain, displacement, and axle load exhibited the strongest positive relationships with load capacity, while temperature and humidity had significant negative effects. The regression model achieved high explanatory power (R² = 0.86), with ensemble learning methods further improving predictive accuracy (R² = 0.91–0.92) and reducing error metrics (RMSE ≈ 27 kN, MAE ≈ 20 kN). Reliability analyses demonstrated high internal consistency (Cronbach’s α ≥ .85) and measurement stability, while validity tests confirmed strong convergence with established load rating standards. Collinearity diagnostics and regularization techniques effectively mitigated redundancy among predictors, enhancing model interpretability. Findings indicate that integrating real-time sensor data with ML enables continuous, adaptive modeling of structural behavior under varying loads and environmental influences. This approach captures complex interactions that static models cannot, providing early indicators of structural deterioration and supporting predictive maintenance strategies. The study concludes that data-driven predictive modeling represents a paradigm shift in bridge engineering, offering superior precision, scalability, and operational insight for infrastructure management. It recommends expanded sensor deployment, hybrid physics-informed modeling, and interdisciplinary collaboration among civil engineers, data scientists, and transportation authorities to advance resilience, safety, and sustainability in bridge infrastructure systems.