Semantic Segmentation-based Algorithm for Visual Measurement of Rotor Shaft Vibration Features

When detecting rotating objects, the rotating objects are usually in complex industrial environments with complex backgrounds and many interferences. Traditional visual measurement methods are difficult to effectively distinguish between the rotating object and the background, resulting in measurement errors. In this paper, a spindle is used as the image acquisition medium, and a semantic segmentation method based on deep learning is introduced to solve the difficulty of visual vibration measurement of rotating shafts. The semantic segmentation network outperforms the traditional visual measurement methods in extracting the features of rotating objects, and the network model, based on the improvement of U-Net, incorporates the improved CSimAM attention mechanism in each of the encoding and decoding stages of the bottleneck layer, which makes the network more adept at segmenting rotary axes in backgrounds with different lighting, blurring and similarity. The study in this paper not only verifies the effectiveness and superiority of the improved U-KAN network in the visual measurement of rotor shaft vibration displacement, but also provides an important theoretical and practical basis for subsequent research.