AN AUTOMATED, FIELD PORTABLE, NON-CONTACT NDE SYSTEM BASED ON OPTICAL FOURIER/ELECTRONIC NEURAL NET IMAGE CLASSIFIER

PHYSICAL OPTICS CORPORATION (POC) PROPSES A NOVEL, HIGH SPEED, LARGE AREA, NON-CONTACT NONDESTRUCTIAVE/EVALUATION (NDE) SYSTEM BASE ON THREE MATURE TECHNOLOGIES: 1) HIGHLY SENSITIVE OPTICAL (HOLOGRAPHIC) INTERFEROMETRY; 2) HIGHLY PARALLEL, OPTICAL FOURIER TRANSFORM FEATURE EXTRACTION, AND 3) AN ELECTRONIC ARTIFICIAL NEURAL NETWORK CLASSIFIER. THE NOVELTY OF OUR APPROACH LIES IN INTERGRATING THE THREE COMPONENT TECHNOLOGIES SO THAT EACH ON COMPLIMENTS THE OTHERS. A DOUBLE-EXPOSURE HOLOGRAPHIC INTERFEROMETRY TECHNIQUE WILL BE USED, WHERE THE MATERIAL TO BE INSPECTED IS NON-CONTACT THERMALLY OR ACOUSTICALLY STRESSED. HOLOGRAMS ARE THEN RECORDED AT TWO DIFFERENT TIMES AS THE STRESS DECREASES. THIS CAN BE DONE IN REAL-TIME BY USING A PULSE LASER AND REAL-TIME HOLOGRAPHIC MATERIALS AND THUS BE IMMUNE TO VIBRATION. THE RECONSTRUCTED IMAGE, SHOWING A GROSSLY AMPLIFIED STRAIN PATTERN, IS FOURIER TRANSFORMED, THUS SEPERATING THE HIGH-FREQUENCY DEFECT PATTERN FROM SLOWLY VARYING LOW-FREQUENCY BACKGROUND PATTERNS. THIS MAKES IT POSSIBLE TO USE A LOW-DIMENSIONAL FEATURE VECTOR FOR SUBSQUENT CLASSIFICATION BY USING A COMMERCIAL, LOW COST, ELECTRONIC, ARTIFICAL NEURAL-NETWORK SYSTEM. THIS NEURAL-NET BASED SYSTEM PROVIDES THE POTENTIAL FOR REAL TIME, FULL-VIEW, FLAW RECOGNITION AND LOCATION WELL BEYOND THE CAPABILITY OF CURRENT INSPECTION TECHNIQUES.