Fabric-inspection system uses neural nets and wavelets
Inspecting textile fabrics is a challenging task because many types of yarns, weave patterns, and quality standards exist. In addition, automated inspection is expensive, costing as much as $1 million per year in some manufacturing plants. The value of fabric affected by defects can also be significant, because recurring problems in high-speed looms can damage thousands of yards of fabric if not quickly found and corrected. Most manufacturing plants employ inspectors to watch for off-quality problems during weaving, but some defects are still not identified until final inspection.
To automate this process, an on-line inspection system that couples vision, neural networks, fuzzy logic, and wavelets to identify defects in fabric has been successfully tested at the textile-manufacturing plant of Johnson Industries (Columbus, GA) in Phoenix City, AL. Developed by researchers at the Georgia Institute of Technology (Atlanta, GA), the equipment could ultimately provide an integrated electronic-feedback system that would monitor and control quality processes throughout the manufacturing cycle. Developed with support from the National Textile Center University Research Consortium (Wilmington, DE), the system automatically identifies defects as the fabric comes off the loom, allowing the manufacturer to immediately correct process problems.
"This system allows operators to quickly correct any problems on the machine and at the same time improve the quality of the manufacturing process," explained Lew Dorrity of Georgia Tech`s School of Textile and Fiber Engineering. The system uses high-speed cameras to scan fabric as it winds onto a take-up roll after weaving. A computer analyzes the information provided by a vision system, identifying abnormal patterns and determining whether they should be considered defects.
Besides detecting off-quality fabric, the inspection system also provides information to pinpoint factors that cause defects. In the design of the system, researchers developed new and innovative software algorithms for the detection and classification of defects, according to George Vachtsevanos of Georgia Tech`s School of Electrical and Computer Engineering. "Signatures from images that are characteristic of the type of defect that might be present are extracted using a wavelet/neural network algorithm combined with fuzzy-logic decision code. "The software integrates learning and optimization tools that avoid false alarms and improve the recognition accuracy," Vachtsevanos says.
Already the technology has been licensed to Appalachian Electronic Instruments (Ronceverte, WV), a manufacturer of textile-related equipment. Appalachian expects to make the prototype into a commercial system that can be retrofitted to existing looms and installed in new machines.
Current R&D activities at Georgia Tech are focusing on improving the software routines to detect a wider class of defects for a variety of fabric styles and the introduction of custom-designed electronics to replace the personal-computer platform. For further information, contact George Vachtsevanos, e-mail: george. [email protected].
