Repository navigation
perf(statics): stop the axial refinement when it leaves the compression where it was - #131
Conversation
|
@codex review |
Codex Review Summary: ✅ CompletedThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
|
Codex Review: Didn't find any major issues. Chef's kiss. Reviewed commit: ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
If Codex has suggestions, it will comment; otherwise it will react with 👍. Codex can also answer questions or update the PR. Try commenting "@codex address that feedback". |
…on where it was With tensile_safety True the AutoMesh static solve splits the elements around an element-mean axial compression until the compression falls below its tolerance, the minimum axial strain stops improving for four passes, or the mesh reaches its element budget. A compression of the equilibrium itself (a line too long for its span, held by its bending stiffness) does not move under refinement, but tiny round-off changes of its value counted as improvement, so the loop doubled the mesh to the budget (32768 elements) on each of the static routes it tried: a power cable deck with its sections given twice failed after 76 s. The loop now also stops when two successive passes leave the violation within 0.1 % of the previous pass's, naming the compression as the equilibrium's. A compression that is a mesh effect falls as its elements are split, by far more than 0.1 % a pass, so a resolvable state still refines as before. Measured: the doubled power cable fails in 6.0 s (75.7 s before) with the same compressive resultant (-514.88 N); cable_statics_matrix writes all 521 output files byte for byte as before (99.6 s, 101.6 s before); every shipped example (TMax cut to four steps) writes identical output files. Tests: the parity tests run the doubled power cable through the driver and require the new stop (fails before: the budget stop after 79 s). Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
The [Unreleased] entries of the open changes are added together in one change, so that they do not conflict with each other. Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
|
@codex review |
d087ad3 to
c06ea98
Compare
|
Codex Review: Didn't find any major issues. You're on a roll. Reviewed commit: ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
If Codex has suggestions, it will comment; otherwise it will react with 👍. Codex can also answer questions or update the PR. Try commenting "@codex address that feedback". |
A compression that belongs to the equilibrium itself (rather than to the mesh) does not fall as elements are split. Its violation stayed flat from pass to pass (57.65124, 57.65104, 57.65109, …), yet changes of about 1e-6 still counted as progress. The axial refinement therefore kept doubling the mesh to its 32768-element budget on each of three static routes before it failed.
It now stops once two successive passes leave the violation within 0.1 % of the previous one, and the message says the compression is the equilibrium's, not the mesh's.
cable_statics_matrixEvery example (TMax cut to 4 steps) gives identical output files.
Test. A driver test that requires the early stop; it failed before, taking 79 s. CTest passes on both BLAS kernels; pytest passes.