From Process Variables to Machine Learning

Presented by Malcolm Keif, Rochester Institute of Technology

From Process Variables to Machine Learning examines how structured measurement data can make process control more practical and effective in printing environments. Drawing on example datasets from inkjet, gravure, and screen printing, this session shows how different ways of structuring and labeling printing data align with different machine learning tasks and model types. By linking key process and quality variables to regression, classification/detection, and segmentation workflows, this presentation demonstrates how data modeling can support more direct and useful process-control decisions. Broad outline: - Printing variables - Measurement - Specifications and tolerances - Machine learning tasks vs. model choice - Inkjet dataset: tabular data → regression - Gravure dataset: labeled defect data → classification/detection - Screen-print dataset: image-mask data → segmentation - Comparison of data structure, task, and model fit