Performance Prediction and Geometric Sensitivity Analysis of Tapered Hollow Circular Dampers Using Finite Element Method (FEM) and Explainable AI
Metallic yielding dampers dissipate seismic energy through controlled plastic deformation, yet their design still relies on closed form estimates that neglect local effects or on computationally expensive finite element (FE) analyses. This study proposes a damper that has tapered fuses at top and bottom connected with circular column by plates. Analytical formulations for the first yield force, plastic hinge force, and yielding displacement were derived by equating external work to internal strain energy and were compared with monotonic Ansys analyses using beam elements (BEAM188) and solid elements (SOLID186). The comparison showed that plate warping, fuse distortion with ovalization, and stress concentration at the tube-plate junction reduce the stiffness below the beam theory prediction. These effects were investigated through plate-thickness, column/fuse combination, and rigid plate studies. A data set of 1,000 samples was generated using the Halton sampling method, supplemented by extreme cases located at the boundaries of design space. The analyses were automated with PyMAPDL and verified against Ansys workbench. Seven surrogate models ExtraTrees, XGBoost, CatBoost, a neural network, Gaussian process, conditional VAE, and DeepONet were trained to predict the normalized force-displacement curve. DeepONet performed best (, (), and all models reached above 0.96. SAHP-based geometric sensitivity analysis of initial stiffness, maximum force, and the maximum curvature knee point showed that the plate-thickness ratio and the fuse taper ratio dominate the post yield response. The proposed framework offers a fast and interpretable alternatives to repeated nonlinear FE analysis for preliminary design of tapered hollow circular dampers.