Positioning-shift failure is a major defect in wafer-handling robot arms that adversely affects the yield of semiconductor fabrication. This failure is often associated with wear on the belt teeth, which may lead to undesired slippage or sudden movements between the belt and gear. Coupled with potential self-correction during subsequent pick-up tasks, its intermittent nature poses challenges for testing and validation by robot manufacturers or in the laboratory. In this study, a real-time prognostic framework was developed for wafer-handling robot arms by integrating CCD-based eccentricity monitoring, temporal feature extraction, stacked LSTM prediction, and edge-enabled online deployment. The framework predicts the maximum eccentric quantity in the subsequent 1-min operating interval and supports online warning generation for operation-oriented failure prevention. The prediction results were in close agreement with the measured values, and the developed model maintained high predictive accuracy over extended operating periods. The platform also successfully updated monitoring information after each prediction cycle and automatically issued warning messages when the predicted eccentricity reached or exceeded the predefined warning threshold.
Chung et al. (Wed,) studied this question.