On-line model updating in earthquake hybrid simulation

Hybrid simulation has emerged as a relatively accurate and efficient tool for the evaluation of structural response under earthquake loading. In the conventional hybrid simulation, the responses of few critical components are obtained by testing while the numerical module is assumed to follow an analytical idealization. Where there is a much larger number of analytical components compared to the experimental parts, the overall response may be dominated by the idealized parts hence the value of hybrid simulation is diminished. It is proposed to update the behavior of the material constitutive relationship of the numerical model during the test, based on the data obtained from the physically tested component. Identifying the parameters that govern the constitutive relationship behavior from the experimental module is a challenging task. Hence, an approach based on optimization tools is developed to determine the model parameters that minimize the error between the numerical and experimental modules. Interior point methods and genetic algorithms are adopted as gradient and non-gradient optimization tools, respectively. Each of which provides different features that are suitable for various types of applications in earthquake response assessment. On the other hand, neural network is utilized as an alternative identification approach. Neural network is advantageous in case the analytical constitutive relationships are not suitable to represent the actual model behavior, as it can be trained independent from analytical guidance to find the mathematical formulas that correlate the input strain to the output stresses. UI-SIMCOR the platform utilized to conduct the hybrid simulation analyses. It can communicate with several finite element programs. Amongst others, ZeusNL is used to analyze the numerical modules due to its efficiency in representing cases of extreme loading and non-linear problems. For model updating purposes, the source codes and the communication protocols between UI-SIMCOR and ZeusNL are modified to be able to exchange the stress-strain information during the hybrid simulation test. Several steel and concrete constitutive models included in ZeusNL library are implemented in the proposed approach. In addition, the components required for the neural network procedure are introduced to the program. The scope of the work also includes verifying the model updating concept through analyzing several numerical problems. These problems include the assessment of regular and irregular structural systems. Moreover, it is shown that through updating the parameters of a simple constitutive model, it can capture the behavior of a more advanced one. Additionally, a number of previously conducted experiments are investigated. A procedure is presented to determine the constitutive model information from the tested component. This procedure is implemented to identify the constitutive model parameters representing a steel beam-column connection and a multi-bay concrete bridge subjected to combined loading. The identified parameters are then utilized to update the analytical model incrementally. The results of both examples show the effectiveness of model updating in minimizing the errors, compared to the pure analytical solution. The proposed approach is expected to enhance the capability of conventional hybrid simulation test to assess structures with several critical components such as high-rise buildings and multi-bay bridges.

Name thesis:degree_name

Ph.D.

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Civil Engineering

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2015

Author dc:creator

• Elanwar Amin, Hazem

Contributors dc:contributor

• Elnashai, Amr S.
• Spencer, Billie F., Jr.
• Lange, David A.
• Kuchma, Daniel A.
• Fahnestock, Larry A.
• Cha, Eun J.

• Model Updating
• Hybrid Simulation
• Earthquake Engineering
• Optimization
• Neural Networks

Statement dc:rights

• Copyright 2014 Hazem Elanwar Amin

Language dc:language

en

Handle dc:identifier

http://hdl.handle.net/2142/73040

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/73040