Automatic recognition method for the three-elementary woven structures and defects of carbon fabric prepregs

Abstract

A vision-based automatic recognition method for the three elementary woven structures and defects of carbon fabric prepregs is proposed for the real-time detection of carbon fabric prepregs in the process of automatic placement that can effectively guarantee the quality and performance of the preformed product. The diffuse scattering of reflected light caused by the surface resin is suppressed by using this new visual illumination method, which is first reported. An Artistic Conception Drawing (ACD) revert algorithm based on the repetitive characteristics of the surface patterns is proposed to automatically identify, match, and classify the three elementary woven structures. The common defect types (such as wrinkles and bubbles) in the prepreg placement process is analyzed and the methods for defect identification, location, and classification based on the model generated by the ACD revert algorithm is discussed. More than 1600 images are captured in the actual production line to validate and compare the performance of the methods. The experimental results illustrate that the ACD revert algorithm has the highest recognition rate (96.53%) for all the samples, and the defect recognition rate of the proposed method exceeds 95%, indicating that it can effectively detect carbon fabric prepregs in the automatic placement process.