Partitioned modelling for nonlinear dynamic analysis of RC buildings under earthquake loading

B.A. Izzuddin, L. Macorini and G. Rinaldin

www.imperial.ac.uk/csm

2CC2013 – Cagliari

Overview

Introduction

Nonlinear dynamic analysis of RC buildings

Partitioned modelling for parallel processing

Application study

Conclusion

3CC2013 – Cagliari

Introduction

Retrofitting and strengthening RC buildings in earthquake regions

Nonlinear dynamic analysis for seismic assessment• Accuracy vs computational demand

Partitioned modelling on distributed memory HPC• Overcoming memory bottleneck for large-scale structures• Reduced simulation time through parallelisation

4CC2013 – Cagliari

Interactions between frame, floor slabs and lateral resistance system• Geometric and material nonlinearity

Modelling of frame members with 1D elements• Fibre elements with nonlinear material models

Modelling of floor slabs and shear walls with 2D elements

Step-by-step time-integration scheme• Accuracy, stability and dissipation of higher modes (e.g. HHT)

Prohibitive memory and computational demands for real RC buildings

Nonlinear dynamic analysis of RC buildings

5CC2013 – Cagliari

Case 2: A parent and a child partition (parent also models a part of structure made with other elements)

Case 3: A parent and a child partition (Same as case 2 but parent and child roles reversed)

Partitioned Modelling for Parallel Processing

Placeholder super-element on parent side

Dual partition super-element on child side

Case 1: A parent and 2 child partitions (parent has only partition super elements)

6CC2013 – Cagliari

Child partitions represented in parent by placeholder super-elements

Parent and child partitions processed in parallel

Child partition wrapped by dual super-element along interface boundary

• Parallelisation through communication between placeholder/dual super-elements

• Effective recovery of super-element resistance/stiffness via frontal solution method

Benefits of partitioned modelling approach using distributed memory HPC

• Overcoming memory bottlenecks and parallel element computations in subdomains

• Additional performance benefits due to parallelisation of frontal solution with reduced front widths

Partitioned Modelling for Parallel Processing

7CC2013 – Cagliari

Application Study

Irregular 4-storey RC building

• Modelling of beams/columns with 1D elements of fibre-type

• Modelling of floor slab diaphragm action using equivalent planar bracing units

• Geometric and material nonlinearity

Seismic excitation in two horizontal X-Y directions

8CC2013 – Cagliari

Application Study

Three computational models with different number of partitions

• Model (A): monolithic (1 process)

• Model (B): 4 child partitions (5 processes)

• Model (C): 14 child partitions (15 processes)

Partition interface boundaries at column locations for models (B) and (C)

• Small number of parent nodes compared to child partitions

• Effective for reducing communication overhead between processes

• Avoids wall-clock time being imposed by solution of equations at parent level

• Ideal speed-up equal to number of child partitions

9CC2013 – Cagliari

Application Study

Identical accuracy for monolithic and partitioned models

Excellent speed-up for partitioned models

Exceptional speed-up for Model (C) exceeding number of child partitions

• Considerable reduction in front width

• Effective implementation of parallel frontal solver on distributed memory HPC systems

Model Child partitions

Maximum front width

Wall-clock time

Speed-up

Monolithic (A) - 318 13hr 37min -

Partitioned (B) 4 510 4hr 45min 2.89

Partitioned (C) 14 198 0hr 30min 27.3

10CC2013 – Cagliari

Conclusion

Nonlinear dynamic analysis of buildings subject to earthquake loading

• Computational demand can be prohibitive for real structures

Partitioned modelling approach for parallel HPC

• Based on parent/child partitions and associated processes

• Identical accuracy to monolithic approach with use of dual super-elements for

recovery of condensed resistance/stiffness at partition interface boundary

• Computational benefits in terms of speed-up and overcoming memory bottleneck

Application to 4-storey RC building subject to earthquake loading

• Exceptional speed-up of 27 with only 14 child partitions

• Additional benefits arising from effective parallelisation of frontal solver leading to

reduced front widths

Practical prospect for nonlinear seismic assessment of real structures

Partitioned modelling for nonlinear dynamic analysis of RC buildings under earthquake loading

B.A. Izzuddin, L. Macorini and G. Rinaldin

www.imperial.ac.uk/csm