Digital Twin-Assisted Optimal Sensor Placement for Real-Time Monitoring of Probe Cards in EWS Applications

This paper presents a digital twin-assisted framework for optimal sensor placement in real-time monitoring of probe cards used in Electrical Wafer Sort (EWS) applications. The framework employs a sensor network to gather operational data from probe cards, supporting future development of machine learning-based strategies for failure detection, maintenance, and monitoring. By coupling a physics-based finite element model with a deep learning architecture that incorporates an attention mechanism, the approach identifies the most informative sensor locations to enhance structural health monitoring efficiency. To address data imbalance, physics-informed data augmentation methods-Gaussian noise addition, frequency jitter, and scaling-are applied, alongside class-weighting during training. Experimental results on simulated frequency response function data from a probe card model show a 92.82% test accuracy, with sensors deemed important by the attention module consistently displaying higher attention scores. These findings demonstrate that the attention mechanism not only achieves high classification performance for failure detection, but also provides interpretable insights for sensor optimization.