Design of a machine learning-based process capability management system to control dispersion and bias of process data

Abstract

Manufacturers commonly utilize process capability indices (PCIs) as part of their statistical process control (SPC) to ensure consistent product quality. However, PCIs only provide information about the presence of a process issue, without explicitly identifying if the issue is caused by dispersion or bias. In this study, we propose a process capability management system (PCMS) that provides visualized results to understand the capability level of all process items at a glance by utilizing heat treatment data. The system visualizes the process capability level of each item by plotting them as dots on a scatter plot, with dot color differentiating process capability levels. This innovative visualization facilitates effective management of numerous process items. In addition, the PCMS incorporates a machine learning classification model to predict and present warning types for items with insufficient process capability, enabling engineers to quickly identify the type of process capability problem and take appropriate measures. Furthermore, by utilizing an XGBoost classifier with the highest machine learning performance and optimizing its hyperparameters, a high prediction performance of 99% is achieved. Finally, a dashboard that provides key information on process capabilities and predicted warning types is introduced to improve the efficiency of engineers in performing process management tasks.