摘要
Tool health monitoring and maintenance scheduling are crucial to empower smart manufacturing. Focusing on realistic needs, this study aims to develop a data-driven framework that integrates partial least squares and exponentially weighted moving-average for feature selection and model construction to monitor and predict tool health via analyzing status data collected from the sensors and thus derive the optimal maintenance strategy for smart production. Indeed, the proposed approach can deal with multi-collinearity of equipment and process data efficiently. An empirical study is conducted for validation in a leading thin film transistor liquid crystal displays manufacturing company. The results have shown practical viability of the proposed approach to provide an early detection of abnormal tool status, prolong the maintenance cycles for enhancing capacity utilization and productivity, and thus reduce the cost. Indeed, the developed solution is implemented in real settings as partial effort for enabling Industry 3.5.