B1817
Title: Predicting bridge condition ratings
Authors: Iryna Okhrin - Dresden University for Technology (Germany) [presenting]
Rene Jaekel - ScaDS AI TU Dresden (Germany)
Pramod Baddam - ZIH TU Dresden (Germany)
Mariela Rossana Sanchez Figueroa - ZIH TU Dresden (Germany)
Abstract: The focus is on modelling and predicting bridge conditions, a critical aspect of ensuring safe transportation for logistic movements, including passengers and goods. Maintaining high-quality roadways, particularly the condition of bridges, is paramount to safe and efficient transportation. However, these inspections can be costly, sometimes conducted partially, and reliant on inspectors' subjective expertise. Furthermore, the increasing traffic density and the impact of natural disasters on bridge materials add complexity to the challenge. The primary objective is to develop models for predicting bridge condition ratings and examining their dependence on various bridge parameters. Multiple machine learning approaches are employed for this predictive task, encompassing k-nearest neighbors, random forest, XGBoost, support vector machine, and deep learning techniques like convolutional neural networks, known for their ability to capture spatial and temporal patterns in time-dependent data. The study leverages the national bridge inventory (NBI) data, containing detailed information on highway bridges across all U.S. states since 1972 and pertinent weather data. The outcomes can significantly enhance bridge management practices, enabling proactive decision-making regarding maintenance and repair.
