Modelling & Simulation

One of the key research areas in the Digital Bauhaus Lab is the development and improvement of simulation methods in order to predict physical fields in structures and components. Joint research project of engineers and computer scientists will focus on new methods and algorithms along the process chain of »modelling – simulation – visualization – validation«. Within this research area, entitled as »Digital Engineering«, the integration of highly sophisticated material models on different scales into engineering models taking into account physical nonlinear material behavior, such as damage and fracture, and stochastic variations of the different input variables and an automated and adaptive model construction are resulting in large and highly nonlinear equations systems, which have to be solved in parallel on appropriate cluster hardware.

Efficient hybrid parallel solver techniques, combining computing and graphic processors can be applied to solve large nonlinear equation systems with up to 70 mio. d.o.f. on multi-core workstations. The compute cluster installed in Bauhaus-Digital Lab will enable the participating researchers to improve spatial and temporal resolution of their models and therewith improve the prediction quality of their models further.

Prof. Dr.-Ing. habil. Carsten Könke, Prof. Dr. rer. nat. Tom Lahmer, Prof. Dr.-Ing. Timon Rabczuk (Institute of Structural Mechanics)

Selected Projects

  • Multiscale models to predict damage and fracture behavior in technical and biological materials

    Starting with high resolution image data to describe the heterogeneous material structure of either composite materials, such as concrete, or biological materials, such as human bone material, concurrent multiscale models using Finite Element Methods are generated and solved in order to predict the damage behavior of these materials.

  • Figure 01: Voxel type FEM model of a trabecular bone material
    Figure 02: displacement field as contour plot for model shown in Fig. 01
  • Parallel hybrid solver strategies with reduced main memory requirements

    Large linearized equation systems with several million degrees of freedom can be efficiently solved with parallel iterative solver techniques on compute clusters. In order to reduce the necessary main memory requirements of the algorithms a new method is developed which allows a matrix free solution technique. Therewith only local stiffness information is compiled and can be easily distributed among the CPU/GPU multi-core hardware. Additional efficiency is obtained by applying hybrid discretization techniques using regular grid type meshes in regions of linear-elastic behavior and aligned meshes in zones with localized damage behavior. 

  • Figure 03: hybrid discretization techniques for heterogeneous composite materials