Semisupervised Classification With Sequence Gaussian Mixture Variational Autoencoder

Evenness of filament yarn is a crucial indicator that significantly impacts the quality of downstream textile products. Therefore, accurate real-time prediction and classification of the coefficient of variation (CV) value, which serves as an indicator of evenness, are of utmost importance. However, current detection methods predominantly rely on offline evenness testing devices, compromising the real-time capability and accuracy of evenness detection. To address this challenge, a semisupervised sequence Gaussian mixture variational autoencoder (VAE) model is developed for predicting and classifying the CV value. This model combines a mix VAE and a sequence-to-sequence structure, integrating a classifier to achieve semisupervised classification of time-series data. To validate the effectiveness of the proposed method, both software and hardware enhancements were implemented on the existing capacitance-based yarn evenness testing device, enabling uninterrupted measurement of yarn evenness and length. The collected data were then used to train the model. Experimental results demonstrate that the proposed model achieves an accuracy rate of 85% in classifying the CV value of the filament yarn.