Quantifying the Relative Influence of Parameters on Pilling Results in Woven Fabrics. A data-driven, explainable simulator for textile pilling prediction across all four processing stages.
Live Simulator: aionsystem.github.io/TEXTILE-PILLING
- Overview
- Project History
- The Problem
- FORGE v1.0 Audit
- Parameter Framework
- Scoring Engine
- Simulator Interface
- Installation
- Usage
- Validation
- Calibration Status — Honest Ceiling
- Repository Structure
- Methodology Documentation
- License
- Acknowledgments
- Contact
This simulator predicts pilling behavior in woven fabrics based on parameters across all four stages of textile processing — fibre, yarn, construction, and finishing. It is an integrated tool relating fibre composition, yarn construction, fabric structure, and finishing treatments into a single, explainable, confidence-scored prediction.
| Stage | Parameters Covered |
|---|---|
| Fibre | Type, blend ratio, fibre length, denier, crimp |
| Yarn | Twist, hairiness, spinning method, count |
| Construction | Weave type, warp/weft density, fabric mass |
| Finishing | Singeing, mercerization, anti-pilling treatment, resin |
What this solution delivers:
- Relative influence scoring — which parameters matter most and by how much
- Scenario simulation — test any parameter change at any processing stage
- Pilling class prediction (1–5) targeting ≥3 per ISO 12945-2:2000
- Explainability — plain-language reasons for every prediction change
- Trade-off analysis — durability vs. hand-feel vs. cost impact
This project did not originate as a single submission — it has gone through several stages, and this README reflects the current state, not the earliest one.
| Stage | Context |
|---|---|
| v0.1.0 | Originally built for the InoCrowd Pilling Challenge |
| FORGE v1.0 Audit | Adversarial five-lens review applied — EV 0.73, certified |
| v1.1 | Current version — audited, methodology fully documented, parameter framework finalized |
| NICE Challenge 2026 | Submitted as one of three proposals to NICE's Yangtze Delta Innovation Grant. Selected as the strongest of the three for industrial readiness — falsifiable against an objective third-party standard (ISO 12945-2:2000) rather than a self-defined scoring framework |
| Current | Released open source. Not pursuing the NICE roadshow/relocation track; the tool and methodology remain freely available for any manufacturer, researcher, or NICE industrial partner who wants to use or build on it |
Problem: Textile manufacturers need to reduce pilling without compromising comfort or appearance. No single integrated tool relates parameters across spinning, weaving, dyeing, and finishing into one actionable, explainable prediction.
Target: Achieve pilling class ≥3 under ISO 12945-2:2000 (Martindale method).
Material scope: Woven fabrics from polyester, viscose, wool, and elastane blends (polyester dominant); spun yarns from staple fibres.
What this solution explicitly avoids:
| Exclusion | Rationale |
|---|---|
| Addressing only one processing phase | Pilling is a multi-stage phenomenon — single-phase tools miss systemic interactions |
| Targeting only one fabric type | Industrial reality demands adaptability across blend compositions |
| Lacking explainability | Manufacturers need to know why a change improves class, not just that it does |
This project was run through FORGE v1.0 — a five-lens adversarial review combining structured certainty scoring, a red-team pass, and a multi-role evaluation council — before being finalized as v1.1.
| Metric | Result |
|---|---|
| FORGE EV (Expected Validity) | 0.73 |
| Status | Certified |
The scoring engine and all parameter weightings in this repository are presented with their honest epistemic status declared — see Calibration Status below. FORGE certification reflects rigor of the audit process applied to the specification and reasoning; it does not assert that the model's literature-derived weights have been validated against real production data. That distinction is treated as load-bearing, not a footnote.
| Parameter | Impact on Pilling | Optimal Direction |
|---|---|---|
| Fibre length | Short fibres pill more | Longer → better |
| Fibre denier | Fine denier pills more | Coarser → better |
| Fibre tenacity | High tenacity = pills persist | Moderate → better |
| Fibre crimp | High crimp = less pilling | Higher → better |
| Polyester % | Anchor fibre — holds pills on surface | Higher = more retention |
| Viscose % | Weak when wet — increases pill formation | Lower → better |
| Parameter | Impact on Pilling | Optimal Target |
|---|---|---|
| Yarn twist | Higher twist = less pilling | 700–900 tpm |
| Yarn hairiness | Lower hairiness = better | H < 4.0 |
| Spinning method | Ring-spun produces more hairiness than open-end | Open-end preferred |
| Yarn count | Finer count = more pilling | Coarser → better |
| Parameter | Impact on Pilling | Optimal Target |
|---|---|---|
| Weave type | Plain < Twill < Satin (tightness) | Plain = tightest → best |
| Warp density | Higher density = less pilling | ≥ 45 picks/cm |
| Weft density | Higher density = less pilling | ≥ 40 picks/cm |
| Fabric mass | Heavier = better pill resistance | ≥ 150 gsm |
| Treatment | Effect on Pilling |
|---|---|
| Singeing | ✅ Reduces pilling — burns off surface fibres |
| Mercerization | ✅ Reduces pilling — increases fibre coherence |
| Anti-pilling chemical | ✅ Reduces pilling |
| Resin finishing | ✅ Most effective — binds surface fibres |
| Shearing | ✅ Reduces pilling |
| Softening agent | |
| Heat-setting | ➖ No significant effect on pilling class |
PSS_base = 0.50 (neutral starting point)
PSS_final = 0.1 + (fibre_score × yarn_score × construction_score × finishing_modifier) × 0.4
Each parameter group contributes a weighted percentage to the final prediction. Weights are derived from parameter count and published correlation strength, and are adjustable within validated ranges:
| Group | Default Weight | Adjustable Range |
|---|---|---|
| Fibre composition | 35% | 25–45% |
| Yarn construction | 25% | 20–30% |
| Woven construction | 25% | 20–30% |
| Finishing treatments | 15% | 10–20% |
Pilling_Class = 1 + (1 − PSS_final) × 4
Classification:
Class 5 — PSS_final ≤ 0.20 No pilling
Class 4 — PSS_final 0.21–0.40 Slight pilling
Class 3 — PSS_final 0.41–0.60 Moderate pilling ← TARGET (ISO 12945-2:2000)
Class 2 — PSS_final 0.61–0.80 Severe pilling
Class 1 — PSS_final ≥ 0.81 Very severe pilling
Pass/Fail: Class ≥ 3 = PASS per ISO 12945-2:2000.
Confidence = 1 − Uncertainty_Mass (UM)
Uncertainty penalties applied when inputs deviate from validated ranges:
Fibre blend outside typical industrial ratios (polyester < 40% or > 90%) +0.10 UM
Twist outside 700–900 tpm +0.10 UM
Density outside 35–55 picks/cm +0.05 UM
Untested finishing treatment combination (>3 treatments) +0.05 UM
Blend includes >10% elastane +0.05 UM
Confidence is capped at 0.95 — no claim of 100% certainty in textile prediction. A confidence score below 70% triggers a flagged output: the prediction is reported as uncertain and the deviation is named explicitly.
┌─────────────────────────────────────────────────────────────────┐
│ PILLING-SIM v1.1 — TEXTILE PILLING SIMULATOR │
├─────────────────────────────────────────────────────────────────┤
│ PHASE 1: FIBRE │
│ Polyester %: ████████████░░░░░░ 60% │
│ Viscose %: ██████░░░░░░░░░░░░ 30% │
│ Wool %: ██░░░░░░░░░░░░░░░░ 10% │
│ Elastane %: ░░░░░░░░░░░░░░░░░░ 0% │
│ Fibre length: ████████░░░░░░░░░░ Medium (>30mm) │
├─────────────────────────────────────────────────────────────────┤
│ PHASE 2: YARN │
│ Twist (tpm): ████████░░░░░░░░░░ 750 │
│ Hairiness (H): ██████░░░░░░░░░░░░ 3.2 │
│ Spinning: ○ Ring-spun ● Open-end │
├─────────────────────────────────────────────────────────────────┤
│ PHASE 3: CONSTRUCTION │
│ Weave type: ● Plain ○ Twill ○ Satin │
│ Warp density: ██████████░░░░░░░░ 48 picks/cm │
│ Weft density: ████████░░░░░░░░░░ 42 picks/cm │
├─────────────────────────────────────────────────────────────────┤
│ PHASE 4: FINISHING │
│ Singeing: ● Yes ○ No │
│ Mercerization: ● Yes ○ No │
│ Anti-pilling: ● Yes ○ No │
│ Resin finish: ○ Yes ● No │
├─────────────────────────────────────────────────────────────────┤
│ RESULTS │
│ ┌─────────────────────────────────────────────────────────┐ │
│ │ PREDICTED PILLING CLASS: 3.4 ✅ (≥3 = PASS) │ │
│ │ CONFIDENCE: 82% (UM: 0.18) │ │
│ │ │ │
│ │ RELATIVE INFLUENCE: │ │
│ │ Fibre composition: 32% │ │
│ │ Yarn construction: 28% │ │
│ │ Woven construction: 25% │ │
│ │ Finishing treatments: 15% │ │
│ │ │ │
│ │ EXPLANATION: │ │
│ │ → Polyester 60% anchors pills — moderate retention │ │
│ │ → Twist 750 tpm is optimal for this blend │ │
│ │ → Plain weave + anti-pilling pushes class to 3.4 │ │
│ │ → Adding resin finish would reach class 4.0 │ │
│ └─────────────────────────────────────────────────────────┘ │
├─────────────────────────────────────────────────────────────────┤
│ TRADE-OFFS │
│ Durability: ████████████░░░░░░ Good │
│ Hand-feel: ██████████░░░░░░░░ Acceptable │
│ Cost estimate: +8% vs. baseline │
└─────────────────────────────────────────────────────────────────┘
# Clone the repository
git clone https://github.com/AionSystem/TEXTILE-PILLING.git
cd TEXTILE-PILLINGNo dependencies — pure HTML, CSS, and JavaScript. Open index.html in any modern browser.
GitHub Pages deployment:
- Push to
mainbranch - Enable GitHub Pages in repository Settings → Pages
- Live at:
https://aionsystem.github.io/TEXTILE-PILLING/
| Action | Method |
|---|---|
| Adjust fibre blend | Move sliders — total must equal 100% |
| Modify yarn twist | Drag twist slider (optimal: 700–900 tpm) |
| Change weave type | Click radio button |
| Toggle finishing treatments | Activate/deactivate switches |
| Read real-time prediction | Updates instantly on any input change |
| Export results | Click Generate Report button |
Interpreting the output:
- Class ≥ 3 — Acceptable for market under ISO 12945-2:2000
- Confidence > 70% — Prediction is reliable; inputs are within validated ranges
- Influence % — The parameter group where improvement has the highest leverage
- Uncertainty Mass > 0.30 — At least one input is outside validated industrial ranges; result is flagged
The scoring engine is calibrated against published textile engineering research. No proprietary data was used.
| Source | Finding | Implementation |
|---|---|---|
| ISO 12945-2:2000 | Martindale method — pilling classification standard | Classification scale 1–5; target ≥3 enforced |
| AATCC Test Method 61-2013 | Accelerated pilling assessment standards | Weighting calibration for finishing treatments |
| Wang Lu et al. (1994) | Pilling mechanism — 4-stage formation, critical abrasion cycle ranges | Core mechanism model |
| Yang, Zhang & Shen (2017) | Fibre length, denier, crimp, and scale correlations | Phase 1 parameter weightings |
| Ghosh, Das & Saha (1987) | Hairiness vs. linear density and twist multiplier | Phase 2 yarn parameter weightings |
| Pilling of Fabrics (1956) | Weave tightness, density, and twist effects | Phase 2–3 parameter weightings |
Edge cases tested:
| Scenario | Expected Result | Simulator Output |
|---|---|---|
| 100% polyester, no finishing | Pill retention high — moderate class | Class 2.8–3.1 |
| 100% viscose, no finishing | Weak when wet — lower class | Class 1.9–2.3 |
| Low twist + high hairiness | Class failure | Class < 3.0 flagged |
| All anti-pilling treatments active | Class target exceeded | Class 4.0+ achieved |
This section exists because overclaiming model accuracy in an engineering tool is worse than disclosing its limits.
What's validated: The mechanism model (4-stage pilling formation), the directional relationships between each parameter and pilling outcome, and the ISO 12945-2:2000 classification mapping are all grounded in published, cited textile research.
What's not yet validated: The specific numeric weights (the 35/25/25/15 group split, the exact coefficients within each phase, the optimal twist range of 700–900 tpm) are derived from literature synthesis, not fitted against real Martindale-tested production samples. They are explicitly tagged as assumptions in the full methodology document.
Path to full calibration: Real-sample testing across the parameter space — realistically several dozen Martindale-tested fabric samples spanning different fibre blends, twist levels, weave types, and finishing combinations — would allow the literature-derived weights to be replaced with regression-fitted ones. This is the single highest-leverage next step for anyone adopting this tool industrially.
| Assumption | Label |
|---|---|
| Polyester acts as anchor fibre similar to nylon | [ASSUMPTION] |
| Viscose reduces pilling resistance when wet | [ASSUMPTION] |
| Elastane heat sensitivity affects finishing | [ASSUMPTION] |
| Optimal twist range 700–900 tpm | [ASSUMPTION] |
| Relative influence weights (35/25/25/15) | [ASSUMPTION] |
| Higher fabric mass reduces pilling | [ASSUMPTION] |
TEXTILE-PILLING/
│
├── index.html ← Main simulator interface
├── style.css ← Styling (AION system design language)
├── script.js ← Scoring engine + UI logic
│
├── docs/
│ ├── methodology.md ← Full parameter framework + scoring formulas
│ ├── validation.md ← Test cases + ISO compliance notes
│ └── tradeoffs.md ← Durability / hand-feel / cost analysis
│
├── assets/
│ └── images/ ← Screenshots, interface previews
│
├── README.md ← This file
└── LICENSE ← GPL-3.0
Full methodology is documented in docs/methodology.md. Key sections:
- Data collection — Published textile research with full citations
- Parameter correlation — Pairwise interaction matrix across all four phases
- Scoring architecture — PSS formula derivation, weight calibration, uncertainty quantification
- ISO 12945-2:2000 compliance — Martindale method mapping and class boundary rationale
The simulator is a parametric engineering model built entirely from published research. All formulas and weights are traceable to cited sources, with assumptions explicitly labeled.
GNU General Public License v3.0
See LICENSE for full terms. Open for research, academic, and non-commercial use.
Commercial licensing: If you need to integrate this into a closed or proprietary product — for example, an industrial QC pipeline that can't be open-sourced under GPL copyleft — a separate commercial licensing path is available. Contact aionsystem@outlook.com.
| Source | Role |
|---|---|
| InoCrowd Pilling Challenge | Original problem statement — quantifying pilling parameter influence |
| ISO 12945-2:2000 | Martindale method standard — pilling classification reference |
| Wang Lu et al. (1994) | Pilling mechanism research — 4-stage formation model |
| Yang, Zhang & Shen (2017) | Fibre parameter correlations and weighting calibration |
| Ghosh, Das & Saha (1987) | Yarn hairiness research |
| AATCC Test Method 61-2013 | Pilling assessment standards |
| AION Constitutional Stack | Epistemic scoring, FORGE audit methodology, and uncertainty quantification architecture |
Sheldon K. Salmon — AI Reliability Architect · Red-Team Framework Designer · AionSystem
| Channel | Link |
|---|---|
| GitHub | github.com/AionSystem |
| Portfolio | aionsystem.github.io |
| linkedin.com/in/sheldon-k-salmon-b0901b378 | |
| aionsystem@outlook.com | |
| ORCID | 0009-0005-8057-5115 |
PILLING-SIM v1.1 · FORGE v1.0 Audited · EV 0.73
Four-stage parameter framework · ISO 12945-2:2000 · Confidence-scored predictions · Open Source