Classification of Composite Defects Using the Signature ClassificationDevelopment System

摘要

The Johns Hopkins University Applied Physics Laboratory and the CarderockDivision of the Naval Surface Warfare Center are developing a Signature Classification Development System (SCDS) to transfer classification technology to nondestructive evaluation (NDE) field equipment. SCDS is a personal computer-based software tool-kit for developing classification algorithms. It includes support for digital signal processing, gating of the signatures, generation of feature vectors, and classification of vectors using artificial neural networks. SCDS successfully classifies ultrasonic signatures from defects in thick section, graphite/epoxy composite test panels. Seven test panels were fabricated with programmed defects embedded one-eighth or halfway into the panel. Six of the panels contain defects representing delaminations, porosity, and contaminations, and one panel serves as a reference standard. Ultrasonic signatures were recorded from the test panels using an ultrasonic C-scan system. SCDS was used to process the signatures and generate feature vectors for input to the artificial neural networks. The classifier achieved a 94% accuracy for one defect, and perfect accuracy for two other defects.