Bridge deterioration prediction models using artificial intelligence in a missing data scenario

Abstract

Bridge infrastructure is crucial for global transportation, but its deterioration due to environmental factors and continuous use presents significant safety and maintenance challenges. This study introduces a methodology for predicting bridge deterioration using artificial intelligence (AI) in data-limited scenarios, integrating real inspection data with simulated data. Artificial Neural network models, particularly the Multi-Layer Perceptron, outperformed conventional methods such as deterministic and probabilistic models in terms of both accuracy and applicability. The methodology progressed from deterministic models based on third-order polynomial functions to probabilistic models using Markov matrices, ultimately culminating in neural networks. This approach overcame data limitations by combining real and simulated data, resulting in a comprehensive database. The AI models effectively captured complex interactions between key variables like bridge age, traffic volume, and environmental conditions, leading to more accurate predictions. Applied in both aggressive and non-aggressive environments, the AI models consistently outperformed traditional methods, achieving a coefficient of determination (R²) of 0.84 and a mean absolute error (MAE) of 0.33 in non-aggressive environments, and an R² of 0.81 with an MAE of 0.34 in aggressive environments. Bridges in aggressive settings showed critical deterioration approximately 10 years earlier than those in non-aggressive environments. These results emphasize the potential of AI to enhance deterioration prediction, improving infrastructure management and maintenance.