Automated Model Simplification for Physics-Based Simulations

Main Participants: Satyandra K. Gupta, Brian Henry Russ, Madan Dabbeeru, and Atul Thakur

Sponsors: This project is sponsored by NSF and Naval Air System Command.

Keywords: Model simplification, defeaturing, finite element model preparation, context dependent simplification, and physics based simulation


Physics-based simulations play an important role during the product realization process. Let us consider few representative examples. Multi-body dynamics simulations are used to determine the sizes of actuators during the design of robots. Finite element simulations are used in structural and thermal analysis of components in the automotive and aerospace industries. Computational fluid dynamics simulation is used in automotive engine cooling system design. These simulations help in reducing the need for expensive physical prototyping and hence shorten the product development time and reduce the product development cost.  Physics-based simulations are primarily driven by 3D CAD data. The computational performance of simulations depends on the number and complexity of the geometric features present in the CAD model. Features are an integral part of modern CAD model and they are used in virtually all the domains of product life cycle, namely design, manufacturing, analysis and maintenance. Even the presence of a single, relatively small geometric feature can increase the size of the underlying discrete physical simulation problem by as much as 10-fold. If we run a finite element analysis on a part with hundreds of small features the computational time will be very large. Extremely large computational times limit the usefulness of simulations during the design cycles. Complex models may often lead to ill-conditioned matrices and hence working with non-simplified complex models may produce inaccurate results. Hence, simply utilizing more powerful computers will not solve the problem associated with highly complex models. In order to get accurate results in a timely manner, one must utilize simplified models that retain the important details and eliminate the irrelevant ones. 


The objectives of this project are:
  1. Development of model simplification algorithms for context dependent contact preserving off-line model simplification for interactive rigid body dynamics simulations.
  2. Development of model simplification algorithms for fluid - rigid body interaction problems such as for simulation of unmanned sea surface vehicles (USSV) at real time refresh rates.
  3. Development of model simplification algorithms for finite elemet analysis model preparation.

Technical Approach

Context dependent contact preserving off-line model simplification for interactive rigid body dynamics simulations: To simplify a geometric model, many techniques involving vertex, edge and facet decimation have been reported. Decimation based techniques have proved to be very useful for the applications like graphics rendering, finite element analysis model preparation, fast transmission of models over network, etc. One of the main limitations of these techniques from the point of view of rigid body dynamics simulation is that the contact points obtained using the simplified models is drastically different than that from the original models. This is undesirable in case of rigid body dynamics simulation as its fidelity depends upon the accuracy of the contact points returned by the collision detection engine. The potential contact points depend upon the collision context (i.e., which parts are colliding). In many problems the collision context is known in advance or can be easily determined as the parts are known beforehand. This opens up a possibility of storing and retrieving multiple representations of parts based on the collision contexts, where each representation can be simplified for the given collision context. This scheme is promising as the memory is relatively inexpensive compared to the real time computation of contact points for fully featured part models. We utilize the collision context to generate physics preserving simplified models. We simplify models with respect to each other in an off-line manner, i.e. before the simulation is performed, in such a way that possible contact points are preserved using part accessibility considerations.

Model simplification algorithms for simulation of USSV-ocean interaction at real time refresh rate:  We utilize potential flow theory based fluid flow model and assume the USSV as a rigid body, incorporating environmental effect such as wave-boat interaction force, damping and effects such as restoring forces and added mass effect to simulate the USSV. We developed a new simplification technique based on (1) clustering of boat geometry facets such that the surface integral difference over simplified and unsimplified boat surface is minimized, (2) caching the force values and exploiting temporal coherence based on difference in ocean waves around the boat, and (3) parallel computing to achieve significant speed up in the simulation without introducing significant errors.

CAD Model Simplification for FEA: This task  involves suppressing the non-critical model features, such as holes and fasteners, based on the functional criteria. Suppression of these features reduces the computationally expensive simulation times, without affecting the accuracy of the results. We are developing an automated procedure for detecting the non-critical features and supressing them.

Related Publications

The following papers provide more details on our approach.


For additional information and to obtain copies of the above papers please contact:

Dr. Satyandra K. Gupta
Department of Mechanical Engineering and Institute for Systems Research
University of Maryland
College Park, MD-20742
Phone: 301-405-5306
FAX: 301-314-9477