Value chain / Stage 04 · Knitting, weaving and vision inspection
Looms interlace warp and weft, and circular cylinders interlock knit loops into greige cloth. A line-scan camera bar lifts defect detection from 60-70% manual accuracy to above 90%, at 60 metres a minute.
13 / 20 flagged · 7 travel on unseen(R29 range, illustrative roll)19 / 20 flagged · 1 travels on unseen(R29 range, illustrative roll)
01 / Instrument 01 · The roll scanner
A broken end or oil spot missed at Stage 04 is dyed, cut and sewn before anyone rejects the garment. Scan the same roll both ways and press a reading to see it act.
Ringed: fault 06, a warp end break at 34.3 m, flagged by vision and missed by the manual table. Hover or focus any fault to read it.
Flagged so far
13 / 20
14 faults past the bar at 84 m
Defect log, latest first
Supplier market on one scale
2024US$523.4m
2033US$6,522.1m
33.8% a year; Asia-Pacific about 46% in 2024.
This roll in vision mode: 95% (19 of 20), inside the sourced band above 90%. Move across the bar to test a value.
This roll on the manual table: 65% (13 of 20), inside the sourced 60 to 70% band. Arrows show the misses heading to Stage 05.
120 m at 60 m/min = 2 min of web time
GFT arithmetic on the R29 upper speed and the illustrative roll length. Press Run the roll: the bar crosses the roll at the true ratio of 1 m to 1 s.
US$523.4m to US$6,522.1m, 2024 to 2033
The camera bar is a market that buys in, about 12.5 times larger by 2033 on the DataIntelo projection (R28).
02 / Instrument 02 · The construction loupe
Cameroon’s demand needs structured woven cloth for uniforms and occasion wear and breathable knit jersey for basics. Pick a construction: the loupe opens on one fault of the same roll, drawn in that cloth.
Fault 06 · Warp end break · 34.3 m: a missing end leaves an open laneFault 05 · Float skip · 27.2 m: the weft floats over six endsFault 07 · Needle line · 39.9 m: one bent needle draws a line down the waleFault 08 · Weft bar · 45.1 m: one pick of off-shade weft runs across
Hover or tap the diagram to read a crossing.
Each pick passes over and under successive warp ends, so every crossing flips.
Africa textile market
Cameroon clothing and footwear, 2024
US$94.70 and 8.0 pieces a person
Diagrams are schematic, not a mill drawing. Faults are from the illustrative roll RL-0417.
03 / Instrument 03 · The loom film
On an air-jet loom or a multi-feeder knitting cylinder, cloth quality depends on exact tension control in four steps. Watch fault 06 being made in them.
Cycle position
25%
STEP 01 · WARP BEAMING AND SIZING
Before a loom can weave, three thousand to five thousand cones from Stage 03 are wound parallel onto a beam and coated with a starch film on the sizing machine, so each end survives heald and reed friction without fraying.
One end of the sheet is already broken: fault 06 before it is cloth.
STEP 02 · HEALD SHEDDING AND WEFT PICKING
Electronic dobby or cam motions raise selected heald frames and lower others to open a tunnel, the shed, across the warp sheet. An air-jet pulse or rapier gripper then shoots one measured pick of weft through it in milliseconds.
The broken end leaves a gap in the upper sheet of the shed.
STEP 03 · REED BEAT-UP AND SELVEDGE LOCKING
The steel reed rocks forward and pushes the new pick into the cloth fell while the warp ends cross over to lock it. Leno or tuck-in selvedges bind both edges so the roll runs flat into Stage 05 dyeing and Stage 06 cutting.
06 · WARP END BREAK · 34.3 m · flagged by vision, missed by the manual table (illustrative).
STEP 04 · CIRCULAR KNITTING CAM AND LOOP FORMATION
In circular knitting, yarn cones feed latch needles guided by cam tracks. Each needle rises to clear its old loop, catches new yarn from the feeder and knocks over a uniform stitch for breathable jersey and piqué.
07 · NEEDLE LINE · 39.9 m · the bent needle draws it down the wale; the bar boxes it (illustrative).
04 / Instrument 04 · From image to weaveable specification
Era 1 · 2023-24
Design Journal (Taylor and Francis) documents generative AI in apparel in 2023 and 2024 as design-side 2D images: Midjourney and Stable Diffusion exploring style, silhouette, colour and surface pattern. Rapid for ideation, yet a pure image model outputs no warp ends per centimetre, yarn count or seam allowance.
Era 2 · 2025-26
Research cited in the annex (arXiv 2509.04705) shows multi-model pipelines spanning fabric construction, shade, weave and silhouette, coupling neural generation with physical simulation. Designs are checked for drape, stretch and weave feasibility before a sample beam is warped.
Working paper: From mood board to specificationEra 3 · At scale
McKinsey’s State of Fashion 2025 associates AI with inventory costs down 35 to 40%, forecast accuracy up 20 to 30%, forecast errors down 20 to 50%, lost sales down up to 65% and revenue per customer up 15 to 25%; 75% of executives prioritise it.
Associations reported for fashion at scale, not a GFT result. Hover or focus a bar to read it.
05 / Instrument 05 · The roll map
Inspection is wasted if the findings stay on a clipboard. When Stage 04 exports an XY defect file with every roll, Stage 05 and Stage 06 can work around faults automatically.
Red cross: an unseen fault travels on. SKIP: a cell routed around a mapped fault. Dots: the 20 illustrative faults, red where unseen. Point at a cell to read it (10 m per cell).
Mode A · blind release · manual 60-70%
7of 36 cells carry an unseen fault on
With no coordinate log, markers are laid over the misses and sewing finds the defective panels mid-garment.
With manual inspection at 60 to 70 percent accuracy and no digital coordinate log, 20 to 30 percent of slubs, warp breaks and oil spots pass unseen into Stage 05 dyeing and Stage 06 cutting. Markers are laid blindly over faults, and sewing operators find defective panels mid-garment, depressing First-Pass Yield and OEE.
Split: Mode A left of the handle, Mode B right. Drag the handle or pick a mode.
Mode B · mapped · vision above 90%
16cells routed around mapped faults, 1 unseen
Every roll carries a defect map tied to its roll ID, so Stage 06 shifts pieces around logged faults.
Under line-scan computer vision (above 90 percent accuracy across more than 40 defect types at up to 60 metres a minute), every roll carries a digital defect map tied to its roll ID. Spreading and cutting tables in Stage 06 shift pattern pieces around logged faults or splice at exact metre marks, protecting the sewing line.
See the digital strategy behind the camera Cells and faults are illustrative, tuned to the R29 ranges.
06 / Instrument 06 · The doffing gate
A loom weaves yarn into cloth; a line-scan camera weaves that cloth into a trustworthy digital record before a single drop of dye is spent.
RL-0417 · 120 m · greige · illustrative
Mode carried from the scanner: line-scan vision.
Sources: GFT’s Cameroon textile programme annex (prepared 31 January 2024; refresh flagged 2026). Roll RL-0417, its faults and cells are illustrative. Not advice.
Fabric inspection benchmarks (manual accuracy 60 to 70 percent, missing 20 to 30 percent; computer-vision accuracy above 90 percent across more than 40 defect types at up to 60 m/min), AI defect market projections (US$523.4m to US$6,522.1m at 33.8 percent a year), generative design citations and African and Cameroon market figures come from GFT’s Cameroon textile programme annex. The demonstration roll, its 20 faults and every cell count on this page are illustrative, tuned to those ranges. Nothing on this page constitutes investment, legal or tax advice.