PillNot

Purpose

PillNot is a GitHub Pages hosted, static web application for early-stage woven fabric development. It helps engineers explore how raw materials, yarn specifications, construction choices, and wet/mechanical processing may influence ISO 12945-2 pilling outcomes.

Model Basis

The model is deterministic and explainable. The same inputs always produce the same predicted pilling class, stage influence, parameter-level feature importance, and trade-off profile.

• Raw materials: polyester-predominant blends with viscose, wool, and elastane.
• Yarn: yarn count, twist level, yarn hairiness, and spinning process.
• Construction: weave pattern plus warp and weft density.
• Processing: heat-treatment temperature and chemical treatment.

Model coefficients are separated in model-config.js, allowing future calibration without changing the user interface or core application wiring.

Path to Class ≥3 Recommendations

The simulator now provides scenario-specific recommendations designed around the challenge target of ISO 12945-2 class ≥3. Depending on the inputs, the app may recommend:

• Increasing yarn twist from Low to Medium or High.
• Comparing Open-End against Ring or Air-Jet spun yarn.
• Increasing construction density toward a stronger woven envelope.
• Testing resin or anti-pilling finishing where cost and hand-feel permit.
• Moving heat treatment closer to the prototype optimum zone around 155°C.

Validation Roadmap

1. Collect representative woven fashion articles across polyester/viscose, polyester/wool, and polyester/elastane blends.
2. Run standardized ISO 12945-2:2000 pilling tests on each article.
3. Record fiber composition, yarn count, twist, yarn hairiness, spinning process, weave, density, heat treatment, and chemical finishing.
4. Compare predicted class against lab-observed class.
5. Calibrate coefficients by product family and fabric type.
6. Publish model version, calibration dataset scope, and expected error bands.
7. Use the app as a pre-lab screening tool, not as a replacement for final physical testing.

Industrial Integration Notes

• Import recipes from PLM, ERP, laboratory, or mill trial systems.
• Store model coefficients as versioned configuration files.
• Export scenario summaries for product-development review meetings.
• Use the optional OpenAI API layer to convert structured model output into engineer-readable explanations, baseline comparisons, lab trial plans, and presentation-ready briefs.
• Use baseline-vs-scenario comparison to document what changed between lab iterations.
• Calibrate separate coefficients for product families, mills, yarn suppliers, and finishing lines.
• Integrate with ISO 12945-2 result history to continuously improve prediction accuracy.

AI Advisory Layer

The OpenAI API integration is intentionally advisory. PillNot first calculates the pilling class, stage influence, feature influence, trade-offs, and Class ≥3 path using transparent JavaScript logic. AI then explains that structured output in textile-engineering language, compares baseline and scenario changes, generates lab-trial plans, and helps users communicate trade-offs.

The demo also includes fallback and manual prompt modes so the presentation can continue without network access or a live API call.

The AI layer should not be used as an unvalidated prediction engine. ISO 12945-2 testing remains the validation basis for production decisions.