Defect Classification
Statistical modelling of fabric inspection data requires adaptive machine learning algorithms that can identify and group unknown types of weaving anomalies without predefined categories. The hierarchical dirichlet process provides a non-parametric Bayesian framework that automatically adjusts the number of defect classes based on the complexity of the scanned textile roll. This method allows automated inspection systems to learn new, rare fabric faults as they appear during high-speed production rather than relying on a static, limited database of known errors.
Algorithmic Mechanism
The algorithm works by sharing statistical data across different levels of the production hierarchy, such as individual loom outputs and overall mill performance. As a high-resolution camera scans the moving web of fabric, the system extracts texture and density metrics. It then uses the hierarchical dirichlet process to group these metrics into distinct anomaly categories, determining which variations are normal structural fluctuations and which are true defects.
System Integration
Integrating this mathematical model into a mill’s quality management system reduces the manual calibration time needed for different fabric constructions. Conventional systems require a technician to set manual thresholds for every new pattern, which slows down the setup process. The adaptive model automates this transition by treating each fabric style as a new level in the statistical hierarchy.
Process Boundary
The model requires high computational power, which can limit its use to offline quality analysis rather than real-time loom monitoring. It is most effective when analysing long-term production histories.
