Advanced Machine Learning Techniques for Predicting Z-Axis Belt Wear in Wafer Transfer Robots

Wafer-transfer robots (WTRs) are critical to semiconductor manufacturing, where precision and efficiency are essential to ensure high production yields and minimize defects. A key challenge is the wear and degradation of the Z-axis belt, which can cause misalignments, operational disruptions, and costly downtimes. This study presents a data-driven approach leveraging machine learning to predict Z-axis belt wear, enhancing WTR reliability and performance. High-frequency acceleration sensor data from the Upper Blade axis of WTRs were utilized to develop and test various machine learning classification models, including K-Nearest Neighbors (KNN), Logistic Regression, Naive Bayes, Decision Tree, and Random Forest. The Random Forest model achieved the highest predictive accuracy at 99.8%, significantly surpassing traditional maintenance methods. These findings highlight the potential of machine learning to transform predictive maintenance in semiconductor manufacturing. By anticipating and preventing faults with exceptional precision, machine learning-based predictive maintenance enables more reliable and efficient operations, reducing costs and enhancing system longevity. This study not only demonstrates the effectiveness of machine learning in predictive maintenance but also sets the foundation for future applications aimed at optimizing the performance and lifespan of critical manufacturing equipment.