Knot Detection with AI Vision in Textile and Apparel Manufacturing
How NorrStudio by NorrSpect identifies yarn knots, slubs, and piecing faults in woven and knit fabrics at production speed preventing surface protrusions and structural weak points from reaching cutting, finishing, and the end consumer.
96%
Reduction in knot-related surface defect escapes to garment assembly
0.6mm
Minimum knot protrusion height detectable inline at production speed
98.7%
Detection accuracy across fine-count cotton, wool, and synthetic yarn fabrics
Overview
Yarn knots are an unavoidable reality of textile manufacturing every yarn break during spinning, warping, or weaving must be joined, and that join, however carefully made, creates a localised mass protrusion above the fabric surface. In premium and technical textile production, however, even a well-tied weaver's knot is unacceptable in the finished fabric face: it creates a visible bump, a potential snagging point, and in fine-count fabrics a structural irregularity that affects drape, handle, and dye uptake in the knotted zone.
NorrStudio, developed by NorrSpect, uses surface topology imaging and point-mass detection models trained specifically on yarn joining defects to identify knots, slubs, and piecing faults across the full fabric width at production speed enabling mills to grade, flag, and address knot density before fabric reaches premium buyers with strict surface quality requirements.
About NorrSpect
NorrSpect is a Swedish AI company headquartered in Umeå, Sweden, specialising in industrial visual inspection for precision manufacturing. Its NorrStudio platform is validated and deployed across automotive and industrial sectors including by manufacturers such as Volvo Cars and is now purpose-built for textile and apparel quality inspection. Knot detection thresholds and fabric-specific models are defined and validated during the pilot phase using real production fabric samples from each client facility.
Industry challenge: why knots are difficult to detect and costly to miss
Yarn knots present a unique detection challenge. Unlike planar defects such as stains or weave faults which create a contrast change on the fabric surface a knot is a three-dimensional mass protrusion, often less than 1mm in height and similar in colour to the surrounding yarn. Under flat frontal lighting, a well-tied knot can be completely invisible to both human inspectors and standard camera systems. It only becomes detectable under raking or oblique illumination that reveals the micro-shadow cast by the protruding mass.
The downstream consequences vary by end use. In luxury shirting and suiting, a single knot visible on the fabric face is grounds for garment rejection. In technical textiles and medical fabrics, an oversized knot at a join can create a stress concentration point that reduces tensile strength below specification. In fine-count knitwear, a piecing knot in the yarn can cause a dropped stitch cascade if the knot catches on a needle during the knitting process.
Weaver's knot
The standard warp or weft join knot used to repair yarn breaks at the loom visible as a small bilateral protrusion on the fabric face when tied with excess tail length
Spinner's knot / piecing fault
A yarn join made during spinning when a roving break is pieced — creates a thickened section in the yarn that produces a localised slub or nep on the fabric surface
Knotter tail protrusion
An automatic knotter join with excess tail left unclipped the protruding yarn end stands above the fabric surface and catches on subsequent handling equipment
Slub knot
An intentional or accidental yarn thick place that creates a raised oval mass on the fabric face distinguishable from decorative slub yarns by position, frequency, and size deviation
Warp knot cluster
Multiple warp joins grouped in the same warp position across a short roll length indicating a high-break-rate section of the warp beam requiring beam quality investigation
Foreign fibre knot
A knot incorporating a contamination fibre — typically a different colour or fibre type creating both a surface protrusion and a colour/texture anomaly at the join point
Solution: NorrStudio AI detection for yarn knots and piecing faults
NorrStudio uses oblique and raking light configurations combined with high-resolution line-scan cameras to cast micro-shadows from yarn knot protrusions making them clearly visible to the AI model even when they produce no contrast signal under flat illumination. Point-mass detection models trained on each client's yarn count and fabric construction classify knots by type, size, and position, distinguishing genuine defect knots from intentional textural elements such as decorative slub yarns or bouclé constructions.
Detects yarn knots and piecing faults as small as 0.6mm in protrusion height across the full fabric width at production speed
Classifies knot type weaver's knot, spinner's piecing, knotter tail, foreign fibre enabling targeted process intervention at the source
Distinguishes defect knots from intentional decorative slub and textured yarn elements using fabric-specification-aware model training
Identifies warp knot clusters high-frequency join zones indicating warp beam quality issues or high-break-rate loom positions
Detects foreign fibre knots via combined surface topology and colour anomaly analysis at the join point
Generates roll-level knot density maps total knots per linear metre enabling roll grading against buyer-specified knot tolerance thresholds
Integrates knot density data with warp beam traceability systems for upstream yarn supplier quality feedback
Solution
NorrStudio AI Inspection Knot Detection Module
Inspection scope
Fine-count woven fabrics, suiting and shirting cloth, knit fabric rolls, technical textiles
Hardware
Line-scan cameras, oblique and raking lighting rigs, motion-sync encoder
Output
Real-time knot alerts, roll knot density maps, warp beam health signals, PDF QA reports
Integration
ERP / WMS, warp beam traceability systems, yarn supplier feedback loops, cutting room CAD
Deployment time
Pilot phase calibrated to client yarn count, fabric construction, and buyer knot tolerance before full deployment
Use case: suiting fabric mill knot density control for luxury menswear buyers
The problem: A fine worsted suiting fabric mill producing 100s and 120s count wool fabrics for European luxury menswear brands was consistently failing buyer knot density audits the brand specification permitted a maximum of two knots per linear metre on the fabric face, but the mill's manual inspection process had no reliable method of counting or locating knots at batching speed. Rolls were either passed on inspector judgement with frequent escapes or subjected to full slow-speed re-inspection that added 25–35 minutes per roll to the outgoing QA process.
The NorrStudio solution: NorrStudio was deployed at the loom exit batching frame across the mill's worsted rapier looms. Raking light rigs were configured for the fine-count wool construction. Models were trained to detect weaver's knots and knotter tail protrusions on the 100s and 120s count fabric face, with the buyer's two-knots-per-metre threshold encoded as the roll grading criterion. Warp knot cluster alerts were configured to flag specific warp beam sections with elevated break rates for beam preparation quality review.
Results:
Metric | Before NorrStudio | After NorrStudio |
|---|---|---|
Knot density audit failures at buyer incoming inspection | 11–15% of rolls per shipment | <1% of rolls per shipment |
Outgoing knot inspection time per roll | 25–35 min (slow-speed re-inspection) | <3 min (automated at batching speed) |
Roll grading against buyer knot threshold | Inspector judgement only | Automated pass / conditional / reject grading per roll |
Warp beam quality issues identified | Not traceable to beam source | 3 high-break-rate beam sections identified in first month |
Foreign fibre knot escapes | Occasional — detected at buyer | Zero escapes — flagged inline at loom exit |
Traceable knot density documentation per roll | None | Full knot density map and count report per roll, archived |
How does NorrStudio detect knots that are the same colour as the surrounding yarn?
NorrStudio uses oblique and raking illumination rather than frontal lighting to detect knots. Because the light source strikes the fabric at a low angle, a knot protrusion even one that is identical in colour to the surrounding yarn casts a micro-shadow that the AI model detects as a surface height anomaly. This makes knot detection independent of colour contrast and effective across both light and dark fabric colours.
Can NorrStudio distinguish defect knots from intentional decorative slub yarns?
Yes. NorrStudio models are trained on each client's fabric specification, including the approved slub profile if the fabric uses a decorative slub yarn. The model learns to classify slub instances that fall within the approved size and frequency range as conforming, while flagging instances that exceed the approved slub specification or occur outside the expected slub yarn positions as defects.
Can NorrStudio count knots per linear metre and grade rolls against a buyer-specified threshold?
Yes. NorrStudio generates a roll-level knot density map expressing knot count per linear metre across the full roll length. Buyer-specified knot tolerance thresholds are encoded into the grading logic, and each roll is automatically classified as pass, conditional, or reject based on whether its knot density falls within, near, or above the agreed threshold.
How does NorrStudio identify foreign fibre knots versus same-fibre knots?
NorrStudio combines surface topology detection which identifies the protrusion with colour anomaly analysis at the join point. A foreign fibre knot incorporating a different-colour or different-lustre yarn produces both a height signal and a localised colour deviation at the knot position. The combination of both signals allows the system to classify it as a foreign fibre knot rather than a standard weaver's knot.
Can knot cluster data be used to improve warp beam preparation quality upstream?
Yes. NorrStudio's warp knot cluster detection identifies zones of elevated join frequency at specific warp positions a reliable indicator of high-break-rate sections on the warp beam or yarn package. This data is surfaced as a beam-level quality signal that can be fed back to the warping and sizing department, enabling proactive beam quality intervention before the warp enters the loom.
Similar Topic
