Implementing low budget machine vision to improve fiber alignment in wet fiber placement

Abstract

Machine vision is revolutionizing modern manufacturing, with new applications emerging regularly. The composites industry, relying on precision in aligning fibers, stands to benefit significantly from machine vision. Ensuring the exact fiber orientation is critical, as deviations can compromise product mechanical properties and lead to failure. Machine vision, particularly in wet fiber placement (WFP), offers a solution for monitoring and enhancing quality control in composite manufacturing. WFP involves pulling fiber bundles, impregnating them with resin, and precisely transporting them to mold tooling for layer-by-layer fabrication. The challenge lies in handling tacky, wet fiber bundles, making tactile sensors impractical. This makes WFP an ideal candidate for contactless process monitoring. The objective of this study is to employ a low budget machine vision in WFP, utilizing a webcam connected to a single-board computer. Artificial intelligence is trained using images of fiber bundles just before placement on the tooling mold. The module detects and measures the position and orientation of a roving in the starting position, enabling the initiation of the WFP process. The methods employed are thoroughly evaluated for reliability and feasibility. After completing only 50 training epochs, a roving detection accuracy of 91.3% could be achieved with almost no critical errors. With additional iterations per placement process, the system functions almost flawlessly at its current state.