Abstract
The United States has many aging bridges, mostly designed with outdated 1950s standards and now facing overloading issues due to heavier modern vehicles. Assessing these bridges' conditions and updating safe load capacities are essential. This paper proposes using acoustic emission (AE) to predict vehicle loads on prestressed concrete girder bridges, potentially supplementing traditional weigh-in-motion (WIM) sensor methods. Three enhanced machine learning techniques: balanced training artificial neural network (BT-ANN), balanced training random forest (BT-RF), and balanced training AdaBoost (BT- AdaBoost) were developed for AE data analysis, addressing training imbalances with an ensemble strategy. The study involved a full-scale flexural test on nine prestressed concrete girders, treating load determination as a classification task. AE signals were categorized into corresponding load steps. The results demonstrate that the proposed BT-RF performs well in classifying AE signals into their respective load steps. Additionally, this paper investigates and discusses the robustness of the BT-RF model.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call