data analysis-Based Failure Mode Prediction and Interpretation for Concrete-Filled Steel Tube Shear Walls

Literature Overview

The study by Ma Gao and Wang Yao, published in Earthquake Engineering and Engineering Dynamics in 2022, develops a data-driven approach for predicting the failure modes of concrete-filled steel tube (CFST) shear walls and provides interpretability analysis of the prediction results. Funded by the National Natural Science Foundation of China (51878268) and the Hunan Youth Talent Plan (2021RC3041), this research addresses the practical challenge of classifying shear wall failure modes—bending failure, shear failure, and bending-shear failure—based on structural parameters and loading conditions. The work is conducted at Hunan University, leveraging the Engineering Structural Damage Diagnosis Hunan Provincial Key Laboratory.

Core Technical Content

The research employs multiple data analysis algorithms to classify CFST shear wall failure modes, with the SMOTE (Synthetic Minority Over-sampling Technique) algorithm used to address class imbalance in the training dataset. The SHAP (SHapley Additive exPlanations) method is applied to interpret the model predictions and identify the relative influence of each feature parameter on the predicted failure mode. The study examines the effects of shear span ratio (λ), horizontal reinforcement index in the web, longitudinal reinforcement index, steel tube reinforcement index in edge members, and axial compression ratio (n) on failure mode classification.

SMOTE Algorithm and Class Imbalance

In experimental datasets for CFST shear walls, certain failure modes may be underrepresented, leading to biased model training and poor generalization performance. The SMOTE algorithm generates synthetic minority class samples by interpolating between existing minority samples and their nearest neighbors, thereby balancing the class distribution without simply duplicating existing samples. This approach is particularly important when the dataset contains a disproportionate number of bending failure specimens compared to shear or bending-shear failure specimens, which is common in experimental databases.

SHAP-Based Interpretability Analysis

The SHAP method provides a unified framework for interpreting the output of various data analysis models by attributing the prediction value to each input feature. In the context of CFST shear wall failure mode prediction, SHAP values reveal which structural parameters are most influential in determining the failure mode and how their values shift the prediction toward or away from each failure category. This interpretability is crucial for engineering applications, as it provides physical insight into the structural behavior rather than treating the prediction model as a black box.

Key Findings and Technical Parameters

The findings confirm well-established structural engineering principles: a higher shear span ratio promotes flexural behavior, while a lower shear span ratio promotes shear behavior. The bending-shear failure mode, which represents a transition state between pure bending and pure shear, is most likely when the shear span ratio is moderate. The reinforcement indices and axial compression ratio influence the relative contribution of flexural and shear capacity, thereby affecting the failure mode classification.

Engineering Practice Integration

The prediction model developed in this study can serve as a preliminary design tool for CFST shear wall systems, enabling engineers to estimate the likely failure mode before detailed structural analysis or experimental testing. This is particularly valuable in the early stages of structural design, where rapid assessment of failure mechanisms informs the selection of appropriate design parameters and detailing requirements. The interpretability provided by SHAP analysis ensures that the model predictions can be cross-validated against engineering judgment and existing design codes such as GB 50011-2010, JGJ/T 191-2010, and ACI 318.

From a fabrication and quality control perspective, understanding the predicted failure mode informs the welding and assembly specifications for the CFST shear wall. For bending-dominated walls, the critical regions are the plastic hinge zones near the wall base, where cyclic plastic deformation leads to steel tube buckling and concrete crushing. Weld quality in these regions must be verified through non-destructive testing (NDT) methods such as ultrasonic testing (UT) and magnetic particle testing (MT) to ensure that weld defects do not initiate premature failure. For shear-dominated walls, the critical regions are the web areas between the edge members, where diagonal cracking and shear buckling govern the failure behavior. The concrete infill quality and steel tube-to-concrete bond strength are particularly important in these regions.

The identification of bending-shear failure as the most difficult to predict is a significant finding, as this failure mode represents a complex interaction between flexural and shear mechanisms. Engineers should exercise additional caution when designing CFST shear walls with moderate shear span ratios, where the failure mode is less predictable and the structural response is more sensitive to parameter variations.

Key Reflections and Insights

The integration of data-driven methods with structural engineering principles represents a promising direction for advancing the design and assessment of CFST shear wall systems. The interpretability provided by SHAP analysis bridges the gap between computational prediction and engineering understanding, ensuring that the model outputs can be meaningfully interpreted and validated by structural engineers. However, the accuracy and reliability of the prediction model depend on the quality and representativeness of the training dataset, which should encompass a wide range of structural configurations, material properties, and loading conditions.

The finding that the shear span ratio is the dominant factor in failure mode classification aligns with fundamental structural mechanics and reinforces the importance of proper geometric proportioning in CFST shear wall design. The secondary influence of reinforcement indices and axial compression ratio highlights the role of material and detailing parameters in modifying the failure mechanism, which is consistent with the design philosophy of capacity-based seismic design where ductile flexural yielding is preferred over brittle shear failure.

In conclusion, this research demonstrates the potential of computational prediction methods for classifying CFST shear wall failure modes and provides interpretable insights into the structural parameters that govern failure behavior, offering engineers a valuable tool for preliminary design assessment and seismic performance evaluation of concrete-filled steel tube shear wall systems.