abstract
- © 2018The new big data driven manufacturing quality philosophy, Process Monitoring for Quality (PMQ), proposes Big Data ¿ Big Models, a new modeling paradigm that includes a big data-driven learning process that requires many models to be created to find the final one. Since many candidates are created, one of the main challenges is to select the model that efficiently solves the tradeoff between complexity and prediction. Most mature manufacturing organizations generate only a few Defects Per Million of Opportunities (DPMO); therefore, manufacturing-derived data sets for classification of quality tend to be highly unbalanced. The Penalized Maximum Probability of Correct Decision (PMPCD) is developed to solve the posed tradeoff. According to simulation and experimental results, the model selection criterion induces parsimony by selecting the model with the minimum number of features needed for an effective/efficient defect detection.