Defect Data Modeling and Analysis for Improving Product Quality and Productivity in Steel Industry

Abstract

Abstract The steel industry is committed to improving product quality and productivity with low production cost. To achieve these objectives, reducing product defects is of utmost importance. There is an increasing interest in developing a model to predict the occurrence of defects online. However, traditional statistical models such as multiple linear regression and Poisson model are not adequate enough to describe the observed defect count data due to the unique characteristics of non-negative integers, overdispersion, high skewed distribution, and excess zeros in the data. This study develops an online quality monitoring system based on the zero-inflated regression modeling. Zero-inflated models are two-component mixture models that combine a count component and a point mass at zero. Intuitively, a mass of zeros observed in defect data can be attributed to two states: a safe (perfect) state, where no defect occurs, and a nonsafe (imperfect) state, in which defects are possible but not inevitable. Zero-inflated models are suitable for modeling discrete and non-negative integers data and can handle both over-dispersion and excess zeros. The effectiveness of the zero-inflated models was verified through their application to the real defect data of a steelmaking plant. The application results demonstrated that the prediction accuracy of the zero-inflated models is superior to the PLS, Poisson, and negative binomial models.