Janata Sadat has achieved a defect rate of 0.28% across its shuttle-loom weaving operations — a figure that rivals the quality benchmarks of premium European technical textile producers — following the full rollout of a proprietary computer vision quality assurance platform developed in partnership with Dhaka-based AI startup SensorEdge Technologies.
The system, deployed across 140 looms at the Narsingdi weaving shed, uses edge-computing inference units mounted above each loom to analyse warp tension, weft density, and shuttle trajectory at 120 frames per second. When an anomaly is detected — a broken warp thread, a missed pick, or a tension deviation exceeding ±4% — the loom controller receives an automatic correction signal within 80 milliseconds, before the defect propagates into the fabric roll.
From Reactive to Predictive Quality Management
Prior to the AI deployment, quality inspectors checked fabric rolls at end-of-loom. Defects discovered at this stage required full roll unwinding, fault isolation, and in many cases complete re-weaving — a process consuming an average of 4.2 labour-hours per incident. The real-time intervention model has reduced rework incidents by 61% quarter-on-quarter and cut per-metre production cost by an estimated BDT 3.20 — savings that are partially passed to long-term contract buyers through revised pricing schedules.
System Performance — Q4 2023 vs. Q4 2022
- 01Defect rate: 1.8% → 0.28% (84% absolute reduction)
- 02Rework incidents per 10,000 metres: 47 → 18 (61% reduction)
- 03Loom uptime: 81% → 88% through predictive maintenance alerts
- 04Quality inspection labour reallocated to value-adding finishing roles
- 05ISO 9001 non-conformance reports reduced by 73% year-on-year
“The looms haven't changed — they are the same Sulzer and Ruti shuttle looms we have operated for two decades. What has changed is that we now have a continuous digital record of every pick made by every shuttle. That level of process visibility was simply not achievable through human inspection alone.”
The company has filed a provisional patent application covering the tension-compensation algorithm and its integration interface with legacy loom controllers. Licensing discussions with two other Bangladesh Jute Mills Corporation member companies are at an early stage. The full technical white paper, including sensor architecture and model training data methodology, will be presented at the Dhaka Textile Technology Symposium in June 2024.
