Benchmarking Seismic Evaluation Methodologies for Existing Reinforced Concrete Buildings

This dissertation consists of three essays on topics related to seismic assessment of reinforced concrete buildings. The first and second essays present studies on the use of machine learning techniques to improve nonlinear models of RC frames. The title of the first essay is "machine learning tools to improve nonlinear modeling parameters of RC columns." The second essay is titled "parametric variation study of RC column nonlinear modeling parameters using deep neural network model." The two essays are aimed at improving the fidelity of nonlinear models by taking advantage of a new generation of data analytics methods. A methodology to develop simple equations for nonlinear deformation modeling parameters is proposed. In this methodology machine learning tools are used to determine the most suitable mathematical expressions to describe the shape of the relationship between input variables and modeling parameters from complex data sets of component experiments. The proposed methodology was evaluated using an experimental set of rectangular and circular reinforced concrete columns using a robust database (ACI 369 column database).The third essay presents a benchmarking study for the nonlinear analysis procedure in the ASCE 41 Standard encompassing three building case studies. Standardizing seismic evaluation of RC buildings is controversial and challenging because the process inherently relies on engineering judgment. The main motivation to standardize the seismic evaluation process is that engineers may arrive at different outcomes depending on the assumptions they adopt to create building models, conveying different measures of expected performance. The variability of possible outcomes is problematic for building owners, building officials, and local authorities, given the financial implications of retrofit or replacement. All these constituencies benefit from an objective process based on a uniform measure of risk. Standards such as ASCE 41 seek to provide a common set of rules for creating building models and assess the expected level of performance for standard seismic hazards, in a process that is uniform and consistent. The ASCE 41 Standard provides a consistent framework for seismic evaluation, based on several methodologies of analysis. Of all the methodologies in the standard, the Nonlinear Dynamic Procedure (NDP) is the most complex and most accurate. The NDP requires the creation of nonlinear numerical models based on modeling parameters for building components specified in the standard. Modeling parameters for some types of components (columns, beams and flexure controlled structural walls) have been carefully calibrated using test data, while others were defined based on engineering judgement. Thorough validation of system performance calculated using modeling parameters specified in the ASCE 41 standard is important given the reliance of the methodology on simulated component behavior. The 2009 NIST Report 09-917-2, titled "Research Required to Support Full Implementation of Performance‐Based Seismic Design", identified benchmarking of current performance-based design methodologies contained in ASCE/SEI 41, Seismic Rehabilitation of Existing Buildings, (ASCE, 2006), as a high priority research need. The third essay evaluated the accuracy of building numerical models created with component modeling parameters and other rules in the ASCE 41 Standard, by comparing calculated metrics of building performance with observed and recorded values.