deploy.

2026-09-15 01:12:50 +08:00
parent 2db889ad89
commit 1a75503105
99 changed files with 4168 additions and 726 deletions
@@ -106,6 +106,10 @@ they are the fastest way to find out what a run actually did.</p>
 <td>physics is on and steps touched, but no touched step carries a physics brief</td>
 </tr>
 <tr>
+<td><code>Play-RapidCut--Detected</code></td>
+<td>a rapid traverse (G00) removed stock — one warning per NC line, anchored on the line's first cutting step, with the line's removed volume and largest cutting depth; the line total is compared against the project's <code>RapidCutVolumeThreshold_mm3</code> (<code>MachiningActRunnerConfig</code>, default 0: any contact); with collision detection on, contact with the spindle stopped is reported there instead</td>
+</tr>
+<tr>
 <td><code>Tool-FluteCount--Zero</code></td>
 <td>physics is on and a milling cutter resolves to zero flutes</td>
 </tr>
@@ -228,6 +228,20 @@ seed pollution.</li>
 <li><a class="xref" href="../../api/Hi.Numerical.Acts.ActMcXyzLinearContour.html">ActMcXyzLinearContour</a> steps by <strong>euclidean</strong> tip travel instead of the
 largest per-axis component, so <code>LinearResolution_mm</code> now caps actual tool-tip travel per step and a
 diagonal move produces up to √3× more steps at the same setting.</li>
+<li><strong>Cutter–workpiece engagement no longer depends on how the contact surface was tiled.</strong> The scan
+that turns a step's contact contours into the engaged angle ranges of
+<a class="xref" href="../../api/Hi.Milling.Engagements.LayerMillingEngagement.html">LayerMillingEngagement</a> switched to a bounding-box approximation on
+dense steps, could pull a merged range back when a shorter slice nested inside it, started bottom
+rings the floor never reached, and read the winding of each triangle — so the same cut produced a
+different engagement, and therefore different cutting forces and temperatures, depending on how
+many polygons the geometry kernel happened to cut the contact surface into. Every triangle is now
+scanned exactly and orientation-free; a cylinder wall and a flat floor sector with known engagement
+arcs reproduce them to 1e-9 rad at four tessellations, and the managed and native paths stay
+bit-identical. Re-running an existing program moves the forces on most engaged steps (58–95% of
+them on the bench fixtures); the largest moves are floors that used to register as empty — the
+whole bottom disc dropped out — and side ranges that were fragmented into pieces separated by
+phantom 1° gaps. Coefficient sets trained before this change embed the old engagement of their
+training program; retrain on a replay if that fit matters.</li>
 <li>The SoftNc pipeline is the default engine.</li>
 <li>Executed <code>SyntaxPiece</code>s freeze to UTF-8 (below), so their <code>JsonObject</code> is a fresh read-only snapshot
 per call rather than a retained live graph.</li>