A MACHINE LEARNING-DRIVEN COMPUTATIONAL MODEL FOR PREDICTION THE RELIABILITY OF REINFORCED CONCRETE BRIDGE DECK UNDER FATIGUE

Abstract

This study presents a hybrid finite element–probabilistic–machine learning framework for fatigue assessment of reinforced concrete (RC) bridge decks subjected to cyclic traffic loading. A three-dimensional finite element model of a 400-ft bridge deck, discretized into five 80-ft segments, was developed in CSI Bridge using the AASHTO HL-93 (2020) traffic loading model. Stress responses at top and bottom fibres were extracted at fatigue-critical locations and used to compute stress ranges, mean stresses, fatigue life, and cumulative fatigue damage based on classical S–N relationships and Miner’s rule. To enhance prediction accuracy, a Random Forest machine learning model was trained using stress-based features and spatial parameters along the bridge length. To better understand the behaviour of each configuration, the stress, fatigue and damage data were further processed in Python using tools such as Pandas for organizing the datasets, NumPy for computing stress fatigue envelopes, and Matplotlib to visualize the results. The machine learning model demonstrated excellent predictive performance, achieving an R² value of 0.999 and a low RMSE, indicating strong agreement with physics-based fatigue damage estimates. The hybrid approach effectively captures nonlinear stress–fatigue relationships and spatial variability in fatigue demand along the bridge deck. The results confirm that integrating finite element analysis, probabilistic fatigue modelling, and machine learning provides a robust and reliable tool for bridge fatigue assessment and maintenance planning.

KEYWORDS: Fatigue reliability; Reinforced concrete bridge deck; CSI Bridge; AASHTO HL-93; Machine learning; Random Forest; Miner’s rule; Hybrid fatigue modelling