The performance of TrueGrid can be improved by changing the way the model is built. Below are the key issues:
1. Every time a single mesh generation command is issued, the entire part must be rebuilt.
2. The resolution of the geometry can slow the projection algorithm.
3. The command insprt command causes an increased load every time the session file is re-run.
4. Smoothing commands (relax, unifrm, esm, tme) can be very slow, depending on the number of iterations selected.
5. Edges or vertices projected to the intersection of tangent surfaces are problematic. The iterative Newton method cannot work in this case and TrueGrid resorts to a primitive search algorithm to find the intersection.
6. Edges in a part are the most expensive calculation, except for smoothing.
7. If there is not enough RAM, TrueGrid®, like any program, TrueGrid® will have to swap memory which will have a dramatic effect on performance.
Here are some suggestions to improve the speed of TrueGrid®.
1. Build smaller parts and use the BB command liberally to glue parts together.
2. Reduce the mesh size while building the part and then increase the mesh density after everything is completed and run the model in batch mode for the last time. Design the mesh density so that it is a half or a third the required mesh density. Then use meshscal of 2 or 3, respectively, to increase the mesh density in your final rerun.
3. Issue multiple commands before having TrueGrid® draw (and consequently rebuild) the mesh.
4. Never issue one of the smoothing commands while in the initial stages of positioning the vertices, attaching edges, projecting faces, or other routine steps to shape the mesh. If you issue a smoothing command, then every time you enter a new command, the smoothing is recalculated. Save the smoothing for the last step in the development of the part.
5. The insprt command is not re-executed every time you issue a new command, so you only have to pay the price once every time you rerun. But this too can become a problem. You can avoid this by planning your part out in advance, removing the need for many insprt commands. Alternatively, once you have run the part with insprt commands, you can recreate a cleaned up version without the insprt commands by using the tghist file that is automatically generated.
6. The intersection of tangent surfaces must be avoided at all cost. The primary way to avoid this is to make a composite surface (sds option of the sd command) and project both faces that share the edge to the composite surface. If you need the edge to be on the intersection of the two tangent surfaces, create a 3D curve and attach the edge to the curve. This avoids the calculation of the intersection of two tangent surfaces. If it is not possible to combine the two tangent surfaces because they are not trimmed where they meet, use the curf command (instead of the default curs command) to permanently attach the edge to the curve, again avoiding the costly calculation of intersecting tangent surfaces. This discussion also applies to vertices in the part.
7. The positioning of nodes along an edge is the second most expensive calculation in TrueGrid®. Only smoothing is more expensive. Usually you cannot reduce the number of edges in a part. But if you are not aware of this fact, you may inadvertently generate a part with many more edges than are needed, decreasing the efficiency.
8. It is usually an easy matter to increase the RAM. You should get approximately 120 bytes per nodes (32 bit system).
9. If you can, use the getol command of about 30 or 50 to reduce the number of polygons used to approximate the surface geometry from a CAD model. Keep in mind that once you build and use a binary IGES file (saveiges and useiges), changing the getol will not effect the existing IGES binary file. If this low resolution affects the projection quality, increase the accuracy to 3. This keeps the number of polygons that are searched for the initial projection to a minimum and then kicks in the highly accurate Newton method on the algebraic surface. This trade off of number of polygons and algebraic surface evaluations can be significant.
