Nonlinear response of steel structures under excessive thermal loads

This thesis develops a practical SAP2000--MATLAB workflow for nonlinear fire analysis of steel frame structures, with OpenSees used as an independent benchmark. The work addresses the gap between simplified fire checks and advanced thermo-mechanical finite element analysis by using structural software that includes temperature-dependent steel properties, thermal expansion, geometric and material nonlinearity. The procedure is based on EN 1993-1-2 elevated-temperature steel properties. MATLAB generates stress--strain curves, creates temperature-specific material and section definitions, assigns thermal frame loads, runs nonlinear load cases and extracts response histories. Material degradation and thermal expansion are treated separately: temperature-specific sections control stiffness and strength reduction, while thermal loads represent free thermal elongation and restraint-induced effects. The methodology is checked using an IPE80 benchmark and then applied to a three-dimensional steel-frame case study, first under uniform elevated temperature and later under non-uniform real-fire temperature histories. The fire data are processed into nodal and member temperature histories, and the final real-fire analysis uses 175 temperature stages. SAP2000 results are compared with OpenSees using absolute vertical displacement histories and relative midspan-deflection histories. The results show that the workflow can reproduce the main displacement trends obtained from OpenSees. The agreement is strongest under the reference mechanical load level and becomes more sensitive when the permanent load is amplified, because stiffness degradation, nonlinear redistribution and local beam deformation become more influential. Overall, the workflow provides a controlled procedure for practical nonlinear fire assessment of steel frames, while its main limitation is the lack of full thermo-mechanical state transfer between externally modified SAP2000 temperature-stage models.