Improving model predictions through partitioned analysis: A combined experimental and numerical analysis

Numerical modeling of civil infrastructure often involves assumptions about boundary conditions between adjacent structural components. For steel frame structures, such connections tend to be semi-rigid and often exhibit nonlinear mechanical behavior, making it difficult to determine the rotational stiffness when the connection is evaluated in an isolated manner. This article adapts a partitioned analysis approach using the Block Gauss-Seidel coupling technique to account for the semi-rigid, nonlinear nature of connections in steel frame structures. A case study of a steel frame with bolted connections is presented in which a constituentmodel is developed for the steel frame with rotational springs to represent the connections. Next, high fidelity, three-dimensional constituent models are developed for three different connection types present in the frame. At the cut-off points of the connection models the internal forces and displacements, calculated by the frame model under a predefined external load, are transferred to the constituent connection models as boundary conditions. Subsequently with the connection models, the rotational stiffness is calculated and transferred back to the frame model through the Block Gauss-Seidel iterations. Coupling iterations are repeated until the connection stiffness calculated by two successive iterations converges, at which point load applied to the frame is increased to develop a full moment-rotation curve. Comparison of the coupled model predictions against experiments show that coupled treatment of the constituent frame and connection models significantly increase the predictive capability of the overall model.