Author: Yuan Feng

Publisher:

Published: 2018

Total Pages:

ISBN-13: 9780438930803

Presented is the development of high-fidelity elastoplastic finite element simulation for earthquake soil-structure interaction problems. Part I presents the modeling and simulation systems for earthquake soil-structure interaction problems, Part II presents the high-fidelity modeling aspects of the systems, and Part III presents the high-performance aspects of the systems. Part I presents the modeling and simulation systems for earthquake soil-structure interaction (ESSI ) problems. The simulation system was named ESSI, which shares the same name with the target problem. The ESSI simulation system is based on the finite element analysis (FEA). Three features are presented as follows: (i) A domain-specific language (DSL) was extended in ESSI to facilitate the modeling of FEA. (ii) To reduce the simulation domain in an earthquake, a new interface of seismic input using wave deconvolution and domain reduction method is developed. (iii) To improve the availability of ESSI system, a remote desktop containing ESSI systems, preprocess and postprocess utilities is deployed on the cloud computing platform. The performance and cost of ESSI cloud computing services are discussed.Part II focuses on high-fidelity elastoplastic material modeling in ESSI system. To improve the earthquake-resistant design, the realistic modeling of soil stress-strain behavior plays a crucial role in the elastoplastic finite element analysis. Three features about the elastoplastic material modeling are presented as follows: (i) To simulate the realistic shear behavior unload cyclic loading, multi-surface plastic model is implemented to match the shear modulus reduction and damping properties. (ii) To represent the micro-structure of soil materials, the three-dimensional finite element formula of Cosserat elastoplasticity are developed and implemented. With the additional couple stress, researchers are able to control the free rotations on particles. The Cosserat elastoplastic models achieve mesh-independent plastic zones in the localization problems. (iii) To build trust in the developed elastoplastic materials, the constitutive algorithms are verified using prescribed solution forcing, Richardson extrapolation, and grid convergence index.Part III focuses on high-performance aspects of ESSI system, including coarse-grain parallelism and fine-grain parallelism. In coarse-grain parallelism, hardware-aware plastic domain decomposition (HAPDD) algorithm is developed for load balancing. Elastoplastic finite element simulation has two computation phases: solving elastoplastic material at Gauss points and solving systems of equations of all elements. HAPDD aims to balance the computation load of solving elastoplastic material. Metis/ParMetis and Scotch/PtScotch are used to adaptively repartition the computation graph. A parametric study is conducted to tune the partitioning kernel for the specific application code in earthquake soil-structure interaction. In fine-grain parallelism, a small tensor library is developed to accelerate the elastoplastic algorithms. While the conventional linear algebra optimizes the large-scale matrix operations, however, in elastoplastic algorithms, millions of small tensor algebra dominates the computation time. Explicit SIMD intrinsic functions are used to improve the tensor algebra performance.Based on the three parts above, the ESSI system was significantly improved in thisresearch in terms of simulation features and performance.