Multi-Task Learning for Anomaly Detection and Remaining Useful Life Prediction in Semiconductor Manufacturing Systems

Abstract

In the current research on predictive maintenance for semiconductor manufacturing systems, although multi-task learning has shown promising results in certain applications, effectively integrating multiple tasks, such as anomaly detection and remaining useful life (RUL) prediction, into a unified semiconductor manufacturing framework remains a significant challenge. The interactions between different tasks can lead to increased model complexity and computational overhead. This study addresses these challenges by designing a multi-task learning framework for semiconductor manufacturing systems. The framework utilizes an LSTM-based autoencoder to extract features from time-series sensor data, and experimental results on real-world semiconductor manufacturing data demonstrate its effectiveness in improving early fault detection. It successfully tackles the two major issues of anomaly detection and RUL prediction, balancing the priority of each task to enhance overall performance, and contributes to the advancement of predictive maintenance research in manufacturing.