Abstract
The interfacial debonding as an invisible damage significantly undermines the bearing capacity and durability of steel-concrete composite structures (SCCS). Although the percussion method has been widely utilized in practical applications, the single damage index (DI) extracted in the analysis process tends to be invalid to abnormalities and leads to misjudgment. The accurate damage prediction requires extracting multiple indicators strongly correlated to the interfacial debonding. Accordingly, this paper proposed a novel debonding detection methodology for a multi-indicator system by integrating the percussion method with machine learning. The classification framework includes two parts: (i) 79 DIs are calculated to represent the interface characteristics comprehensively, and the fused feature selection algorithms are investigated to determine the optimal DI subset. (ii) Five classical machine learning approaches after hyperparameter tuning are evaluated with the final objective to compare the prediction accuracy of each model. Aided by the experimental measurements and correlation analyses, 20 key DIs are filtered out as model input. Then, the evaluation performance reveals that the KNN classifier achieves the superlative identification results on the experimental dataset, presenting a prediction accuracy of 99.6 %. Finally, a new dataset confirms the accuracy and effectiveness of the proposed method in the form of 2D damage imaging. The research findings of this study are beneficial for improving the detection accuracy and provide significant guidance on the percussion method-based interfacial debonding detection for the SCCS.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have