The textile industry’s quality control has traditionally relied on manual visual inspection, a labor-intensive process prone to human error. This process constitutes a typical “vigilance task,” where sustained attention over long periods often leads to a decline in detection performance, known as the vigilance decrement, due to high cognitive load. To address these human-centric challenges, this paper proposes a Human-in-the-Loop (HITL) automated fabric inspection system designed not to replace, but to augment the capabilities of skilled inspectors. The system was implemented as a “brownfield” retrofit, upgrading an existing inspection machine with a low-cost, non-invasive hardware and software package. At its core, a Real-Time Models for object Detection (RTMDet) model is utilized to perform the primary, high-load vigilance task of defect scanning. This allows the human operator to focus on the higher-value task of verifying and classifying potential defects identified by the AI. A case study conducted in a real-world jeans manufacturing factory demonstrated that this HITL approach enhanced inspection task efficiency by approximately 2.5 times compared to traditional manual inspection, significantly reducing operator cognitive load and enabling parallel tasking. This study provides a practical blueprint for SMEs in the textile industry to implement effective, human-centric AI solutions within existing operational constraints.

Understanding fiber-fiber interactions in materials using cellulose nanofibril (CNF) is essential for advancing bio-based material design. This study investigates the relationship between fiber network density and crosslinking points to estimate inter-fiber interactions in CNF sheets. The sheets were prepared from CNFs obtained from bacterial cellulose pellicle and bamboo pulp dispersions, subjected to various drying and solvent exchange conditions. Atomic force microscopy (AFM) images were processed and analyzed using SOAX software to quantify crosslinking points. Tensile tests were conducted to evaluate mechanical properties, and interaction energy per crosslinking point was calculated. The results showed that solvent exchange with ethanol increased the apparent number of crosslinking points but reduced the calculated interaction energy, suggesting weaker inter-fiber bonding. The estimated interaction energies ranged from values comparable to moderate hydrogen bonds to those dipole–dipole interactions, depending on the preparation methods. This approach provides a simple and effective method to estimate fiber–fiber interaction energy from network density and mechanical data, offering valuable insights for CNF-based material development.

Abstract: The generation of fiber fragments (FF) from textiles through friction under daily-life has recently attracted attention due to their contribution to microplastic pollution. In this study, a long-term worn lab coat was examined using SEM to characterize location-specific frictional degradation. Distinct features were observed: fiber fracture and subdivision at corners and folds, surface degradation under surface-to-surface loading, and fiber twisting under multidirectional stresses. To reproduce these features, artificial rubbing tests were conducted using the Gakushin-type rubbing tester and Martindale tester. The Gakushin-type rubbing tester demonstrated the potential to reproduce characteristic degradation features in SEM images by adjusting the counterface and plastic bars. By contrast, the Martindale tester showed relatively low selectivity and versatility in degradation conditions, limiting the range of reproducible targets. These results suggest that frictional reproduction using the Gakushin-type rubbing tester provides a practical approach for evaluating FF generation and may contribute to the development of textile products with reduced environmental impact.