Residual‐based damage evaluation of RC columns using feature classification and model optimization

Abstract

SummaryA residual‐based damage evaluation framework with physical feature classification is proposed for categorizing RC columns regarding engineering parameters. To realize a real‐time damage assessment, increasing the damage model's feasibility is of interest. Considering the physical diversities of different RC columns, a damage model with a constant parameter setup may become unreliable in practice. To address this issue, a self‐organizing map (SOM) is utilized to classify RC columns in terms of their engineering parameters. Such classified RC columns are associated with similar physical attributes. For RC columns sorted in the same cluster, a damage model parameter optimization procedure is implemented. One hundred and twenty RC columns are selected from the Pacific Earthquake Engineering Research (PEER) center. In specific, using the selected RC columns, a comparative study has been implemented to show the feasibility of the proposed damage model. The SOM is then trained to classify 100 RC columns associated with a model optimization procedure. Such well‐trained SOM is utilized for pattern recognition, and the optimized damage model is used to evaluate the damage status of 20 columns in the validation data pool. In the end, a parametric study is accomplished to extend the reliability of the proposed damage evaluation framework.