An integrated machine-learning platform for assessing various dynamic responses of steel beams

Abstract

Quantitative monitoring of structural damage progression is crucial for ensuring the safety of civil infrastructure. This study presents a novel machine-learning (ML)-based platform designed to predict damage states in steel beams. The proposed platform incorporates eight distinct ML algorithms, typically employed for training on the dynamic responses of steel beam structures. Then, the prediction performance of each algorithm was evaluated to identify the most effective model. A new filtering method was developed to improve the efficiency in processing, training, and evaluating of the data in a timely fashion and integrated into the platform. Additionally, the platform was augmented with the Shapley additive explanation, which precisely link input variables with specific damage types. This integrated platform confirmed a high accuracy (∼ 97%) in classifying damage types and identifying the best ML model for specific inputs, demonstrating the benefits of the proposed method in enhancing overall assessment capabilities.