Self-Supervised Algorithm for Predicting Data Based on Knitting Direction in Capacitive Strain Stitch Sensors

Abstract

In this study, the characteristics of knit-based stitch sensors were compared using capacitance and stress values based on the knit direction. The restored predicted data were analyzed using a masked autoencoder algorithm. First, changes in capacitance with respect to strain were compared, followed by a comparison of stress values with respect to strain. The capacitance \((C)\) values reflected differences in the distance between electrodes \((d)\) and the area of the electrodes \((A)\), with the course direction showing a larger variation. Although the stress values \((\sigma )\) exhibited similar trends in the graph, the comparison using the elastic modulus \((E)\) and Poisson's ratio \((\nu )\) confirmed that the course direction had greater deformation under the same tensile strain. Finally, using the masked autoencoder algorithm, the restoration rate of the original data was measured under 10%, 30%, 50%, and 100% noise levels, showing nearly perfect consistency. These results demonstrate the analysis of knit stitch sensor characteristics and the performance evaluation of the masked autoencoder algorithm.