Characterization of twist of fancy yarns using wavelet analysis of sensor signal

Abstract

Yarn twist variations may cause stripes in the direction of weft yarns or local defects on a fabric surface. Since fast Fourier transform and time analysis cannot directly detect local frequency variations of yarn signal and defect sensors are designed to detect the diameter decreases without considering frequency analysis, no data associated with twist-related frequency changes can be obtained when inspecting Chenille yarn (Cy) defects. This study proposes the prediction of twist level (T) and twist variations (Delta T) of Cy whose twist changes in accordance with the spatial period of pile density by using wavelet analysis, allowing localized frequency variations to be obtained. Complex-valued Paul wavelet was used to determine the Delta T of signals with small frequency fluctuations, while Morlet wavelet was addressed for signals with high frequency change. The relation of the signal frequency to the pile yarn density and, correspondingly, twist was modeled by equations. To prevent discontinuities in wavelet cross-spectrum (WCS), the twist simulation signal was generated by equalizing twist oscillation amplitudes without changing their phase. To compare ideal twist to the local twist, another simulation signal demonstrating the ideal twist at sample-specific frequency was generated. The WCS of the simulation signals allowing the segmentation of variation intervals was used for determining Delta T and yarn portions, where the twist is compatible with ideal twist, by establishing correlation between scales and twists. For yarn samples with various T and Delta T types, the T and Delta T results obtained by the proposed wavelet-based algorithm showed the mean absolute relative percentage errors of 1.617% and 37.062%, respectively.