2019 Pittsburgh AISTech

Sessions Speakers Exhibitors Products Press Releases Show Specials Floor Plan Buzz

A Deep Probabilistic AI Framework for Predictive Quality Control (Room 306)

We will present a Sense-Predict-Recommend framework for increasing product quality and minimizing out-of-tolerance rejection. The deep probabilistic AI (DPAI) framework combines latent feature representation and clustering, Bayesian hierarchical regression and simulation-based recommendation based on historical production data. Coil widths outside tolerances result in lost revenue and lower mill productivity. DPAI contributes to optimal process control input values to obtain finished coil products more precisely within width tolerance for a given grade, thickness and finishing temperature. Predictive simulations enabled a significant reduction in over-width and under-width coils, resulting in a positive impact on mill performance.