This page uses CSS to present the content in the best possible manner.
If you can see this message, then CSS is not enabled in your browser, and the page will not appear as the designer intended.

Detail Removal

Introduction

In finite element analysis, the geoemtry of CAD models is often too complex to analyse in a reasonable timescale, meaning an idealised model must be constructed, often from scratch. The research presented here invistiages the use of the medial axis to automatically suppress details on 2D models, and using topological operations to manually merge, collapse, and kill to suppress details on 3D models. Eventually the medial surface may be to automate these topological operations on 3D models.

Loop and Edge Aspect Ratios on 2D objects

Holes and inner loops of material in a 2D region can be identified from their aspect ratio using the medial axis. In these figures the outline of a 2D region is shown in blue while the medial axis is shown in red. The loop aspect ratios shown here are the length of the medial axis around the loop divided by the average radius of the inscribed disks centred on the medial axis. The smallest aspect ratios indicate those holes which could be removed with least effect on the stress distribution.


The figure below shows the edge aspect ratio of all the edges in the model. Here aspect ratio is defined as the length of the object edge divided by the average radius of the inscribed disks touching it. Note how the small fillet radii are identified.


These edge ratios can be used to select holes and edges to suppress. As shown below, the fillet radius has been removed, the hole reduced to a node, and the slot dimensionally reduced to a crack.



Moreover, as shown in the figure below, other simple features such as notches and protrusions could be identified, and then classified using measures similar to those outlined above. A notch aspect ratio would be the length of the medial axis touching the notch boundary, divided by the average radius of the inscribed disc centred on that portion of medial axis. Protrusions could be classified by the amount of taper in the associated medial axis. Eventually these details could be suppressed as illustrated in the diagram below.



After meshing using TR6 elements, although the simplified model required significantly fewer elements, the deflection results are within 7% of the fully detailed analysis, as shown below:

Topological Merging and Collapsing Operations

To suppress details in 3D solid models, several euler operators have been implemented using the ACIS solid modeller. These permit a user to interactively perform operations such as merge_faces, merge_edges, kill_void, kill_hole, collapse_face_to_edge, collapse_face_to_vertex. The resulting model can then be cut into convex primitives and hex meshed using mid-point sub-division. The pictures below illustrate how a support bracket, and a housing were simplified and meshed, using an interactive version of the chopper. These models took approximately twenty minutes to simplify and mesh, and required considerably less elements than the original fully detailed models.



Error Estimation

After details have been removed, it is important to know the accuracy of the results from the FE analysis. One method being investigated for this is using submodelling. In submodelling deflection results from a coarse global model are applied to the boundary of a submodel of the detail. If the stiffness of the detail is similar to the stiffness of the equilivant region in the simple coarse, then stress values around the boundary of the submodel should be similar to those in corresponding positions of the simple coarse model. Summing up these stress jumps between coarse model and submodel around the boundary of the submodel, should provide an estimate of the accuracy of the simplification. If these errors are hiigh the detail should be reinstated and a futher analysis performed. Below is shown a plate with a small semi-circular notch. The coarse model ignores the notch completely and uses only six elements. Four submodels of radii 0.5, 1.0, 2.0 and 3.0 were then analysised as submodels of this coarse model. The stress jump energy for the 0.5 notch is small relative to the total strain energy in the global model, whilst the 3.0 notch is over a tenth of the total strain energy.

Further work

Research is currently being focused on using the medial surface to automatically select faces and edges to suppress, and on performing estimates of the errors resulting from these simplifications