Quality Prediction Method by Modeling the Sustained Effects of Irregular Process Disturbances

In manufacturing domain, there are increasing needs for quality control by utilizing various data collected from on-site. Machine failure, equipment component replacement, and other process disturbances are now collected by various sensors. The analysis of these data can help on-site managers to detect product quality drifts and to cope with them quickly and properly. However, the irregular and sparse nature of process disturbances causes prediction accuracy issue and modeling time issue. In this research, we propose a product quality prediction method using a stochastic process to model the irregular disturbances, and make prediction based on dense, regular matrix of sustained effects sampled from the stochastic process for modeling time reduction. As the result of applying the proposed quality prediction method to actual manufacturing data, the MSE (mean squared error) is reduced by 84.6% and the modeling time can be shortened to within 3 hours for daily update. Therefore, it can be estimated that our quality prediction method can help on-site managers to detect quality drifts at early stage and have a better control of product quality.