International Journal of Scientific Research and Engineering Development

Defect detection and localization remain major challenges in apparel production systems, where small inconsistencies such as yarn tension variation, needle damage, machine vibration anomalies, and fabric density defects can propagate through the supply chain and create substantial quality losses. This study introduces an OTDR-inspired sensor-trace analysis framework that adapts the AI-augmented signal processing approach originally developed for fiber fault localization. The proposed system uses embedded sensors installed on garment production lines to capture real-time vibration, tension, and surface-reflection traces that mimic the waveform-based diagnostic principles used in fiber networks. A convolutional neural network (CNN), trained on over 6,000 simulated and real sensor-trace signatures, automatically identifies defect types such as fabric roll inconsistencies, needle faults, and misalignment anomalies. Experimental results demonstrate a 27–41% improvement in defect localization accuracy over conventional thresholdbased QC methods, significantly reducing waste, rework, and production cycle delays. This work provides a scalable and low-cost architecture suitable for modern apparel industries seeking predictive quality control.