Condition-Based Monitoring for Failure Detection and Cause Identification in Wafer Transfer Robot Using Machine Learning and Explainable Artificial Intelligence Algorithms

Abstract

This study introduces a method for real-time prediction of failures in Wafer Transfer Robot (WTR), which is critical in semiconductor manufacturing. Using data from acceleration sensors and fast Fourier transformation (FFT) for anomaly detection, the approach involves modeling failure data for high-risk components like bearing motor, ball screw, timing belt, robot hand, and end effector. A deep neural network (DNN) differentiates between normal and failure states, and a random forest (RF) algorithm classifies the failure components. Additionally, the Shapley additive explanations (SHAP) algorithm pinpoints the frequency bands liked to specific failure causes, enhancing the identification of failure causes in WTR.