Fundamental Validation of an AI-Based Impact Analysis Framework for Structural Elements in Wooden Structures

This study proposes an AI-based framework for impact analysis of wooden structures, focusing on quantitatively assessing how individual seismic elements and their spatial locations influence structural response. A single-story residential building was used as a case study. Numerical time-history analyses were performed using a detailed three-dimensional nonlinear model, and parametric variations in stiffness and strength were systematically generated using an orthogonal array. Machine learning models were then trained to investigate the relationship between these parameters and seismic responses, and explainable artificial intelligence (XAI) techniques, including SHAP, were applied to evaluate and interpret parameter influences. The results suggest that wall elements oriented parallel to the target inter-story drift direction generally have the greatest effect on seismic response. Quantitative analysis indicates that the relative importance of these elements roughly corresponds to their wall lengths, providing physically interpretable evidence. Model comparisons show that linear regression achieves high accuracy in the elastic range, while Gradient Boosting performs better under strong excitations inducing nonlinear behavior, reflecting the transition from elastic to plastic response. SHAP-based analysis further provides insights into both the magnitude and direction of parameter influence, enabling element- and location-specific interpretation not readily obtained from traditional global sensitivity measures. Overall, the findings indicate that the proposed framework has the potential to support the identification of influential structural elements and the quantitative assessment of their contributions, which could assist in informed engineering decision-making.