Co-authors​ :

 Youssef MRAIDI (GERMANY), Lin LIU , Fabian DIEMAR , Swen ZAREMBA , Dr.Klaus DRECHSLER (GERMANY) 

The Towpreg manufacturing process involves impregnating continuous dry fibers with resin to create a pre-impregnated material for composite manufacturing. Towpregs are crucial in advanced composites for their controlled resin content, ensuring consistent and precise material properties. Maintaining a constant towpreg width during production is essential for subsequent composite manufacturing processes such as filament winding or automated fiber placement (AFP). This paper presents an inline process monitoring system based on artificial intelligence (AI) for detection of the carbon fiber roving and measurement of the width in an efficient way. A robust optical measurement technique is necessary in order to monitor the tow width during the process providing high-resolution images of the tow and its surrounding. This study utilizes Convolutional Neural Networks (CNNs) for towpreg detection and segmentation, employing two contrasting approaches: a two-stage method with Mask R-CNN and a one-stage method with YOLOv8. Depending on the model, the accuracy of the measurement ranges from 92% to 98%, matched with the true width (measured manually). Moreover, this routine can be expanded for inline defect detection such as yarn splits or fiber breakage during the towpreg manufacturing process.