# Aetec Q64 TTL — Original CAD & automation demonstration

Updated 11 September 2026.

Option 1 retains the supplied machine assembly. Option 2 modifies selected original conveyor members, removes the two groovers and relocates guards. Both are shown simultaneously at the same installation position in two views of the drawing-based factory. The separate factory viewer is unchanged.

## Geometry fidelity

- Source: `Q64 TTL Pipe Robotic Welding Line.STEP`, 420,388,120 bytes, AP214 from SolidWorks.
- 9,033 part instances, 757 unique leaf definitions, 9,944 assembly nodes. All 757 definitions have rendered geometry.
- CAD faces are tessellated at 0.10 mm chordal tolerance and 0.15 radians angular tolerance. The web model is a surface representation of the STEP solids, not the native boundary-representation model.
- 12,488,111 unique triangles; 65,041,207 triangles across instances. No mesh decimation or vertex quantization was applied.
- Meshoptimizer compression is lossless relative to the tessellated buffers. All 2,631 buffer views were decoded and verified byte-for-byte against their input. Repeated parts share geometry without removing detail.
- Source assembly names, placements and colours are preserved. Visible movement uses the actual CAD links and tools. Labels start off. Guards can be hidden for inspection and restored.
- **Compare CAD assembly poses** restores the original assembly, including its overlapping alternative socket sizes, fences, panels, floor logos and human context. These source features have not been deleted.

## Source STEP defects

The STEP contains 104 unresolved references to `#18446744073709551615` in edge loops. OpenCASCADE could not tessellate 290 of the 67,464 unique faces (about 0.43%). No complete part definition is omitted, but these affected parts have incomplete surface patches. The app is therefore not a claim of perfect face-for-face reproduction of damaged source solids.

Affected definitions include the ISE20 pressure switch (198 faces), PTF-600F Y1 (52), K3561-1 robot cable part (25), SC-32 cylinder (6), two motor definitions (3 each), J2 robot part (1), Tip Sample C (1) and Fanuc teach pendant (1). A clean re-export of these source parts is needed for complete surface recovery. The detailed face records are in `models/q64/conversion-report.json`.

## Corrected process and operator inputs

All socket centre distances use the same First End cut-face datum (local negative Z). They are absolute distances, not cumulative socket-to-socket spacing. Socket #1 sets the clock orientation. Later UI rows accept offsets relative to #1; blank means the same orientation. Looking from First End toward the far end, positive clock angles turn clockwise. Internally, each socket stores its resolved absolute angle. Socket axes remain radial, perpendicular to the pipe axis.

For cutting, the chuck rotates the selected opening to the underside. The original plasma torch aims upward, and the cut disc drops below the outside surface. R1 withdraws before the pipe rotates 180 degrees to present the opening upward for socket placement and welding. The gripper approaches open, closes after arriving at pickup, holds during the two tack welds, opens before withdrawing and stays parked during full welding. R1 withdraws before R2 releases. Both robots remain parked during chuck translation/rotation operations.

The process centre is normalized `(0.65033, 1.4715, 0.60)` m. This moves the demonstration forward along the pipe relative to the earlier location; robot bases and the original machine geometry are unchanged. CAD-derived tool geometry is kept. Upward cutting uses an authored aiming tilt and roll selected for fixture clearance, rather than assuming a vertically mounted replacement torch. Full weld and tack postures can differ. These angles are geometric demonstration choices, not a plasma cut-quality or controller calibration result.

One production recipe accepts 1–200 identical pipes, with 1–12 individually dimensioned sockets. Six examples include the requested First End distances of 500 and 4,000 mm with a 45-degree relative angle. The large-header example starts at 500 mm to leave more chuck clearance. Additional recipes can be queued separately. The optional combined demonstration contains 43 pipes. Repeat whole batch retains the completed count and visible outfeed buffer. At 1x, authored joint speeds are limited to 60 degrees/second; 8x and 16x are visualization speeds. Cycle times are editable assumptions.

Production sockets are hollow consumable geometry using entered OD, wall and height. Pipes have wall thickness, visible openings, end grooves and weld beads. These process-generated items supplement the original machine; they do not replace its CAD geometry. Dimension ranges are input limits, not the machine's approved production capacity. A finite search may fail to find a checked route for otherwise valid dimensions; such a new recipe is blocked and the previous batch is retained.

## Outgoing conveyor

The pipe transfers onto the cross conveyor at normalized X = -0.74 m, Z = -3.80 m. It advances sideways by the outgoing pitch and then stops. All waiting finished pipes move by the same indexed step, not continuously down the pipe axis. Pitch uses the preceding pipe's right-side envelope plus the next pipe's left-side envelope and the chosen gap, including socket protrusions and a small bead allowance. All distances are calculated from the actual clock orientations.

Finished pipes remain visible until they reach the far collection edge. That edge assumes automatic collection; it does not simulate an actual pickup robot or operator. A bounded buffer is reconstructed from the batch history, so seeking and repeating a batch preserve the same spacing without an unbounded pile of rendered stock. Roller-to-socket contact, downward socket support and collection mechanics still require an engineering study.

## Collision checking

The checker now uses enclosing box hierarchies built from clusters of triangles in the actual source CAD. It includes 253 moving robot/tool surfaces, 888 nearby fixed CAD surfaces, and 110 moving chuck/jaw/support surfaces. Gripper finger opening follows the current socket diameter. Original source socket alternatives are hidden in simulation and restored in CAD assembly poses.

Clearance is checked against the complete authored joint interpolation using recursive swept bounds. Each joint's displacement contribution is bounded using 4 m reach; the largest bound computed from the original link chain and CAD extents is 2.6934 m. Robot bounds include 10 mm margin per surface and fixed/fixture bounds 2 mm. Chuck translation, jaw rotation/opening and support lift receive additional movement bounds during preflight. These are conservative enclosing volumes, not exact triangle/solid intersection calculations.

Both robot-to-robot pairs and selected self-collision pairs are tested. Same-robot link pairs separated by two or fewer joints are excluded because of articulation/mounting interfaces. The floor, nearby guards, fixed structure and original moving chuck/jaw/support geometry are included. Planning checks whole cutting/welding contours, not just endpoints, and verifies transfers between operations. Every displayed pose is checked again; a detected interference holds the last displayed safe pose.

The scope does not include exact pipe/socket-to-tool contact, carried consumable collisions, all remote conveyors, workpiece-to-factory swept solids, detailed cable deformation, robot controller limits, dynamics or PLC safety logic. Deliberate assembly contacts and overlapping source tool variants within an end-effector are not a tool self-collision assessment. Therefore this is an authored visual/kinematic model, not a certified full-cell collision assessment or a production robot program.

## Aetec presentation

The original round company logo is reused on larger ground decals and mounted on original robot and enclosure surfaces. The machine geometry and source colours remain intact. Updated floor, lighting and camera framing improve the presentation. CAD assembly poses hides the added presentation decals and restores the original source floor logos. Station labels start off and remain switchable.

## Coordinates and factory placement

Standalone web coordinates in metres are `x = STEP X / 1000 - 14`, `y = STEP Z / 1000 - 0.06`, `z = -STEP Y / 1000 - 28`. The full source assembly bounds include its context: 8.148 m wide, 19.078 m deep, and 6.162 m high.

This comparison uses wall A in the rear ground-floor crane bay. The normalized machine is rotated 180° around Y and placed at factory `(4.661, 0, 17.650)` m. The local process centre therefore sits at factory `(4.01067, 1.4715, 17.050)` m, within the drawing's approximately Z = 6.5–24.6 m crane travel zone. The two panes are alternative installations in the same location, not two lines occupying the bay together.

Measured clearances are 660 mm from the outermost equipment to wall A, and 1,460 mm from the mini warehouse to the wall. Other original panels project farther than the warehouse. The original front guard is about 182 mm beyond the rear stair enclosure. Positioning was changed to avoid the stair enclosure, the interior column and the crane runway girder. Factory profiles and some column sizes remain drawing-based estimates, and the source PDF floor datum discrepancy remains unresolved.

Both complete source-pose assemblies are checked using the machine's retained mesh triangles against nearby factory boxes. The robot preflight additionally includes these factory obstacles. Upper floors and the roof are cut away for visibility; hiding an object does not remove it from the robot checks. Crane bridge, trolley and hook are shown at the drawing-based position and do not travel in this demonstration.

## Option 2 modification

- Remove source groover assemblies 2296 and 2364 and the associated distal grooving drag chain 2213.
- Shorten longitudinal transfer modules 80 and 998 by 1,000 mm using a splice in normalized Z = 6.9–7.9 m. Intermediate hardware is removed; distal rollers, drive and support assemblies are relocated intact. Profiles retain their cross-sections. Other CAD geometry is shared without decimation.
- Remove the last incoming chain module 4816 and its lift-up module 6171, leaving five chain support stations. The grooving sequences and end-groove graphics are absent from Option 2, including completed pipes.
- Relocate front guards inward. The overall equipment envelope changes from 19.078 × 8.148 m to 17.078 × 8.148 m, a 16.30 m² reduction. This includes guards and source context; it is not a service-access envelope.
- Preserve robot bases, tools, chuck, mini warehouse and outgoing conveyor geometry. The shared finished-pipe receiving position is moved 250 mm along the existing outgoing deck to local Z = -3.80 m, keeping 6 m output away from the factory column.
- Compact stock centres at Z = 4.6 m, or 5.2 m for pipes shorter than 1.5 m. The checked 6 m and 1.2 m examples retain at least two support stations. These checks do not solve pipe sag or support loading.

Both options use the same recipe, elapsed time and playback speed. Retained process stages reuse exactly the same checked robot paths and durations. Option 2 removes two positioning movements plus the grooving dwell: with the default assumptions, 22 seconds per pipe. Once Option 2 finishes, it holds while Option 1 completes. Repeating resets both at the end of the longer batch.

The download in the app is a component change list and recipe in JSON. This version does not generate a modified native STEP solid. The original supplied STEP tessellation remains the geometry source. The splice edits are a layout proposal; final chain routing/tensioning, end connections, guarding, support strength and manufacturing drawings require CAD engineering.

## Validation and practical limits

CPU checks load the retained glTF with the same loader/decoder as the web app. They verify 9,033 CAD instances, source-pose restoration, actual rendered robot transforms against the collision transforms, and matching chuck/jaw/support transforms. The check also rejects a deliberately overlapping robot pose.

Full sample batches are checked for completed pipes and sockets, First End dimensions, final clock angles, downward-facing cutting openings, upward plasma direction, external disc drop, parked robots during indexing, joint-rate limits and batch seams. Outfeed checks cover indexed motion, waiting stock, socket envelope spacing and continuity across batch repeats. A 200-pipe batch is included. Earlier full-sample checks are in validation-report.json. The current comparison, factory placement and default/short batch checks are in comparison-validation.json. An additional 3,020 sampled pipe/output envelopes across all six recipes and both options were checked against nearby factory solids; this is a sampled envelope check, not a continuous carried-workpiece sweep.

The STEP is a static assembly, without calibrated TCPs, approved robot joint limits, PLC logic, travel schedules or real cycle-time measurements. The source video establishes the broad sequence; the latest user instructions override its cutting/outgoing interpretation. Welding heat, penetration, deformation, grip force, conveyor friction and load dynamics are not solved.

No browser interaction or GPU visual/performance QA was performed for this update. Original supplied preview images show the source assembly, not a screenshot of the revised app. A desktop WebGL browser with adequate memory is the intended viewing environment. The complete CAD detail is retained even when it increases loading time. The supplied process video remains available inside the app.

© 2026 Aetec Pte Ltd