UBC Theses and Dissertations
Performance-based seismic design using designed experiments and neural networks Zhang, Jiansen
There are many uncertainties involved in seismic design process. Such factors as earthquake ground motions, variability of structural geometries and material properties, and approximation in analytical model contribute to the non-performance of the structure. Therefore, reliability methods are applied in structural engineering to assess the structural performance. However, seismic reliability assessment may necessitate a large number of performance function evaluations, each requiring a nonlinear dynamic structural analysis, which is a formidable, if not impossible task. In performance-based seismic design, a set of design parameters must be found to meet the associated target reliability levels for different performance objectives. This is conventionally achieved by trial-and-error using repeated forward reliability analysis, which is inefficient. Hence, it is desirable to develop an efficient and effective procedure that can reduce the colossal computational efforts, making seismic reliability assessment and performance-based design tractable. This study has explored for the first time applications of Design of Computer Experiments and Artificial Neural Networks for seismic reliability analysis, as well as performance-based seismic design, taking into account structural nonlinear dynamic behavior and all the major uncertainties involved. Experimental design is utilized to construct response databases for Neural Networks learning. Neural Networks act as a surrogate of the computer program, improving computational efficiency by approximating structural responses. Case studies have been carried out to demonstrate the applicability and efficiency of the proposed methods in seismic reliability assessment and performance-based seismic design.
Item Citations and Data