An ensemble model of explainable soft computing for failure mode identification in reinforced concrete shear walls

Abstract

Shear walls are critical structural components that resist lateral forces. The failure mode of a shear wall directly impacts the overall structural behavior. Therefore, accurate identification of the failure mode of shear walls is essential for the design, construction, and maintenance of safe and reliable structures. This paper presents an ensemble learning (EML) approach that integrates various different types of models to predict the failure mode of shear walls. Utilizing 393 shear wall samples gathered from the literature, a new EML model with better generalization performance was constructed using random forest (RF), support vector machine (SVM), light gradient boosting machine (LGBM), and logistic regression (LR) models, and these sub-models were further optimized using bayesian optimization (BO). The results demonstrate that the EML model achieved the highest recognition accuracy at 92.37 % in eight models, and the prediction performance of sub-models significantly improved after optimization. Multiple model interpretation methods revealed that variables such as aspect ratio, wall length-to-thickness ratio, cross-sectional shape, and boundary element reinforcement indices play a particularly influential role in predicting failure mode. This study provides an in-depth explanation of the identified failure modes from both an overall and sample-specific perspective and constructs a graphical user interface (GUI) for demonstrating the recognition of failure modes. Overall, as a study of ensemble learning using different types of models to recognize the failure mode of shear walls, the EML model established in this paper can effectively predict the failure mode of reinforced concrete shear walls.