Automated Gate Rejection Algorithms for Real Time Recycled Fiber Blend Variations
Real-time NIR spectroscopy coupled with automated divert gate algorithms stabilizes furnish quality, rejecting off-spec recycled stock before pulper entry.
Bayesian Classification
Probabilistic Categorization
Statistical sorting methods determine the likelihood of a material sample belonging to a specific grade or quality group based on prior observed data. Automated bayesian classification helps mill operators identify the origin of contaminants or the likely cause of surface defects by comparing sensor data to known patterns of historical failure. It functions by updating the probability of a hypothesis as more evidence becomes available.
Defect Identification
Vision systems on a coating line use these calculations to differentiate between a simple scratch and a serious slime hole that might cause a press break. The algorithm assigns a category based on features like edge sharpness, length, opacity and orientation. Because bayesian classification handles uncertainty well, it reduces the number of false positives that would otherwise stop a high-speed machine.
Real-time processing allows the system to adjust its internal model with every new scan of the moving web.
Training Input
The reliability of the output depends on the quality of the initial data sets used to define the categories. A system relying on bayesian classification requires a diverse history of board samples to distinguish between natural wood grain variations and printing errors. It provides a statistical confidence level for every decision.
Once the confidence level falls below a set floor, the system triggers a manual inspection by a technician.