Real-time hybrid simulation is an innovative method of testing structural systems subjected to earthquake loading. In a hybrid simulation, the complete structural system is included by modeling some of the components of the structure in the laboratory (the experimental substructure) while the remaining components of the structure are modeled analytically (the analytical substructure). A reasonable concern pertaining to real-time hybrid simulation is ensuring that all of the necessary steps (calculations of responses, application of displacements, measurements of forces) within a time step can be completed so that synchronization is maintained between the substructures. If this does not occur, the accuracy of the results of the simulation is compromised. This concern is further heightened as models for analytical substructures become more intricate to accurately represent real-world structures, thereby increasing the number of necessary computations. Therefore, it is desirable to find a way to decrease computational time to enable real-time hybrid simulations of larger structural systems, where the analytical substructures have a large number of elements and degrees of freedom. A possible solution is to use grid-based processing in which the simulation is conducted using multiple processors instead of a single processor to perform the state determination of the elements in the analytical substructure to determine the associated restoring forces.
To test this, comparisons are made between the usages of 2 xPC processors versus 1 xPC processor to execute simulations of numerical models. A model to define the analytical substructure for the structural system of interest is constructed using Lehigh University's HybridFEM program in MATLAB. The numerical simulation of the structure takes place next, followed by a hybrid simulation where the experimental substructure is comprised of two magnetorheological (MR) dampers. The structures used to conduct this study are based on the 3-story, 9-story, and 20-story ASCE benchmark structures.
Results show that as the computational demand on a particular xPC is decreased via increase of the computational demand on a second xPC, the computational load and time for the original xPC is reduced, and the computational load and time for the second xPC is increased.
Stephanie Tong University of Illinois at Urbana-Champaign
REU Site: Lehigh University Advisor: Dr. James M. Ricles Mentors: Yunbyeong Chae, Thomas Marullo
Cite this work
Researchers should cite this work as follows: