Alternative Integrators and Parallel Computing for Efficient Nonlinear Response History Analyses

Abstract eng:
This paper is an attempt to address solutions to two common problems associated with nonlinear response history analysis (NRHA). The first is the convergence problems that occur at high levels of nonlinearity, while the second is the computational time which can be prohibitively excessive in certain cases. To overcome the first problem, suitability of the use of alternative integration methods in NRHA instead of the commonly used implicit Newmark integration is investigated. These alternative methods are Explicit Newmark and OperatorSplitting methods, which do not require iterations and convergence checks. NRHA conducted on highway overpass bridges with multiple-column bents showed that the Operator-Splitting method provides very close results to those from the Implicit Newmark method for some of the considered models even for high levels of nonlinearity. The Explicit Newmark method was not successful for the investigated bridge system because of the stability limits which were not feasible to satisfy. However, explicit methods, when applicable, can be useful since they eliminate the convergence problems and reduce the simulation time. Means to improve convergence are also investigated for the implicit Newmark integration method, especially when the use of alternative integration methods is not possible. To address the second problem, parallel and cloud computing resources are utilized together with the alternative time integration methods. Particularly, the parallel version of OpenSees (OpenSees MP) running on NEEShub, currently providing access to up to 4232 processors, is used where significant computing time savings are achieved.

• Export as: BibTeX | MARC | MARCXML | DC | EndNote | NLM | RefWorks
• Add to your basket