In-situ adaptive reduction of nonlinear multiscale structural dynamics models
- A computational framework of the in-situ, adaptive Projection-based Model Order Reduction (PMOR) is proposed for solving a general nonlinear dynamic multiscale problem. This framework focuses on the reduction at the microscale by using the method of database of local Reduced-Order Bases (ROBs), which is constructed and updated on-the-fly to avoid the weakness of extrapolation in the conventional non-adaptive training. Instead of performing a classical offline-online training approach, an online algorithm is designed for this adaptive process. The proposed framework treats the deformation gradient, defined at each macroscale material point, as a vector parameter domain. This characterization enables it to locate each discrete boundary value problem of the microscale Representative Volume Element (RVE) in the database and assign to it online the most appropriate local ROB, by computing the distance from the query point to the centroid of ROB in the parameter domain. Excessively large local ROB will be split into two new local ROBs to achieve a computational efficiency. The model accuracy is controlled by sampling new information from a High Dimensional Model (HDM) and updating the existing local ROB. Performance results obtained for two computationally intensive nonlinear, dynamic, two-level multiscale applications reveal that the proposed computational framework for the in-situ, adaptive microscale projection-based model order reduction is capable of delivering wall-clock time speedup factors while maintaining a desired level of accuracy
|Type of resource
|electronic resource; remote; computer; online resource
|1 online resource
|He, Wanli, (Mechanical engineer)
|Degree committee member
|Stanford University, Department of Mechanical Engineering.
|Statement of responsibility
|Submitted to the Department of Mechanical Engineering
|Thesis Engineering Stanford University 2020
- © 2020 by Wanli He
Also listed in
Loading usage metrics...