L. Song

C. Bach, L. Song, T. Erhart et al.

Projection-based nonlinear model order reduction methods can be used to reduce simulation times for the solution of many PDE-constrained problems. It has been observed in literature that such nonlinear reduced-order models (ROMs) based on Galerkin projection sometimes exhibit much larger stable time step sizes than their unreduced counterparts. This work provides a detailed theoretical analysis of this phenomenon for structural mechanics. We first show that many desirable system matrix properties are preserved by the Galerkin projection. Next, we prove that the eigenvalues of the linearized Galerkin reduced-order system separate the eigenvalues of the linearized original system. Assuming non-negative Rayleigh damping and a time integration using the popular central difference method, we further prove that the theoretical linear stability time step of the ROM is in fact always larger than or equal to the critical time step of its corresponding full-order model. We also give mathematical expressions for computing the stable time step size. Finally, we show that under certain conditions this increase in the stability time step even extends to some hyper-reduction methods. The findings can be used to compute numerical stability time step sizes for the integration of nonlinear ROMs in structural mechanics, and to speed up simulations by permitting the use of larger time steps.

Z. Ma, L. Song, X. Feng et al.

Intracranial aneurysm (IA) is a life-threatening blood spot in human's brain if it ruptures and causes cerebral hemorrhage. It is challenging to detect whether an IA has ruptured from medical images. In this paper, we propose a novel graph based neural network named GraphNet to detect IA rupture from 3D surface data. GraphNet is based on graph convolution network (GCN) and is designed for graph-level classification and node-level segmentation. The network uses GCN blocks to extract surface local features and pools to global features. 1250 patient data including 385 ruptured and 865 unruptured IAs were collected from clinic for experiments. The performance on randomly selected 234 test patient data was reported. The experiment with the proposed GraphNet achieved accuracy of 0.82, area-under-curve (AUC) of receiver operating characteristic (ROC) curve 0.82 in the classification task, significantly outperforming the baseline approach without using graph based networks. The segmentation output of the model achieved mean graph-node-based dice coefficient (DSC) score 0.88.