Research on Visual Measurement for Levitation Gap in Maglev System

The measurement for levitation gap is a very important part of levitation control system. The traditional levitation gap sensors have some shortcomings such as small measurement range and specific installation requirements, and usually need nonlinear correction and temperature compensation to meet the control requirements of the magnetic levitation system. In our work, two novel measurement methods for levitation gap based on computer vision are proposed. First, the levitation gap is measured by calculating the image pixel area of the region of interest. The error of the pixel area model can be limited within +/- 0.250 mm, and the mean absolute error (MAE) is 0.095 mm for full scale (FS). Second, the model named SelfConvNet based on convolutional neural network is designed for measuring the levitation gap. The error of SelfConvNet model can be limited within +/- 0.048 mm, and the MAE is 0.013 mm for FS. The measurement results show that the SelfConvNet model is better than SqueezeNet and Visual Geometry Group 16 models, which has high measurement accuracy and strong anti-interference ability. The method based on pixel area has lower measurement accuracy but higher processing speed. Finally, the proposed gap measurement methods have been verified in closed-loop experiment of maglev ball control system.

Automated summary

This paper proposes two novel measurement methods for levitation gap based on computer vision. The first method measures the levitation gap by calculating the image pixel area of the region of interest, with low measurement error and high processing speed. The second method designs a model named SelfConvNet based on convolutional neural network, which has high measurement accuracy and strong anti-interference ability. Experimental results demonstrate the effectiveness and reliability of these two methods.