An Image Selection Strategy Based on Swin Transformer and CWT for Robotic Grinding Monitoring

This paper proposes an image selection strategy (ST-CISS) based on Swin Transformer (ST) and continuous wavelet transform (CWT), with the aim of efficiently screening out high-quality image datasets and providing high-quality samples for equipment wear detection during robot grinding. The method first generates multiple datasets through CWT transformation with different parameter combinations and records the corresponding transformation time. Subsequently, an image dataset quality evaluation index based on the correlation coefficient is proposed, and an ST model for robot grinding fault diagnosis is developed. Finally, the optimal image dataset is selected by considering the transformation time, quality evaluation index, and classification accuracy. The experimental results demonstrate that the ST-CISS method effectively selects high-quality datasets and provides efficient data processing and training methods for grinding robot fault diagnosis.