TY - JOUR
T1 - High-fidelity non-linear analysis of metal sandwich panels
AU - Pinho santos, Luis fernando
AU - Pinho, Luis
PY - 2018/9/6
Y1 - 2018/9/6
N2 - The considerably superior specific strength and stiffness of sandwich panels in relation to conventional structural components makes their employment for two-way spanning structural applications a highly attractive option. An effective high-fidelity numerical modelling strategy for large-scale metal sandwich panels is presented in this paper, which enables the capturing of the various forms of local buckling and its progression over the panel domain, alongside the effects of material non-linearity and the spread of plasticity. The modelling strategy is further enhanced with a novel domain-partitioning methodology, allowing for scalable parallel processing on high-performance computing distributed memory systems. Partitioned modelling achieves a substantial reduction of the wall-clock time and computing memory demand for extensive non-linear static and dynamic analyses, while further overcoming potential memory bottlenecks encountered when conventional modelling and solution procedures are employed. A comparative evaluation of the speed-up achieved using partitioned modelling, in relation to monolithic models, is conducted for different levels of partitioning. Finally, practical guidance is proposed for establishing the optimal number of partitions offering maximum speed-up, beyond which further partitioning leads to excesses both in the non-linear solution procedure and the communication overhead between parallel processors, with a consequent increase in computing time.
AB - The considerably superior specific strength and stiffness of sandwich panels in relation to conventional structural components makes their employment for two-way spanning structural applications a highly attractive option. An effective high-fidelity numerical modelling strategy for large-scale metal sandwich panels is presented in this paper, which enables the capturing of the various forms of local buckling and its progression over the panel domain, alongside the effects of material non-linearity and the spread of plasticity. The modelling strategy is further enhanced with a novel domain-partitioning methodology, allowing for scalable parallel processing on high-performance computing distributed memory systems. Partitioned modelling achieves a substantial reduction of the wall-clock time and computing memory demand for extensive non-linear static and dynamic analyses, while further overcoming potential memory bottlenecks encountered when conventional modelling and solution procedures are employed. A comparative evaluation of the speed-up achieved using partitioned modelling, in relation to monolithic models, is conducted for different levels of partitioning. Finally, practical guidance is proposed for establishing the optimal number of partitions offering maximum speed-up, beyond which further partitioning leads to excesses both in the non-linear solution procedure and the communication overhead between parallel processors, with a consequent increase in computing time.
U2 - 10.1680/jencm.18.00022
DO - 10.1680/jencm.18.00022
M3 - Article
SN - 1755-0785
SP - 79
EP - 96
JO - Proceedings of the Institution of Civil Engineers - Engineering and Computational Mechanics
JF - Proceedings of the Institution of Civil Engineers - Engineering and Computational Mechanics
ER -