Research on quality detection method of high-performance fiber thin web based on machine vision

With the growing demand for nonwoven materials in high-end applications, traditional offline basis-weight measurements interrupt continuous production, and in-line weight-detection systems on carding machines only provide coarse, localized weight profiles (typically at centimeter-scale resolution), and cannot resolve fiber- or pore-level distribution uniformity. This paper proposes an online quality-inspection method for nonwoven fiber webs using machine vision and deep learning. High-resolution images captured on a bespoke inspection platform are preprocessed via wavelet transforms and image-enhancement techniques. A convolutional neural network regression model then estimates web basis weight in real time. To quantify fiber-distribution uniformity, this paper introduces a porosity-based metric and demonstrates that its coefficient of variation (CV) correlates strongly with the CV of measured web basis weight. The proposed approach delivers data-driven support for real-time adjustment of carding-machine process parameters, thereby overcoming the technical challenge of online uniformity monitoring in high-end nonwoven fabric production.