| Infrared thermal imaging is an evolving approach useful in nondestructive evaluation of
materials for industrial and research purposes. This study investigates the use of this method in combination with multivariate data analysis as an alternative to chemical etching; a destructive method currently used to recover defaced serial numbers stamped in metal. This process involves several unique aspects, each of which works to overcome some pertinent challenges associated with the recovery of defaced serial numbers. Infrared thermal imaging of metal surfaces provides thermal images sensitive to local differences in thermal conductivity of regions of plastic strain created from stamping pressures in mechanically stamped pieces and a heat affected zone in laser engraved samples, both extending to depths below the visible characters. These are exposed to the surface when the serial numbers are removed. These thermal differences are quite small and thus not readily visible from the raw thermal images of an irregular surface created by removing the stamped numbers. As such, further enhancement is usually needed to identify the subtle variations. The multivariate data analysis method, principal component analysis, is used to enhance these subtle variations and aid the recovery of the serial numbers. Multiple similarity measures are utilized to match recovered numbers to several numerical libraries, followed by application of various fusion rules to achieve consensus identification. Confidence indices are applied to the identification results by conformal prediction as a measure of the accuracy of the identification.
Key Words: Serial number restoration, Lock-in infrared thermography, Principal component analysis, Zernike moments, Pseudo Zernike moments, Similarity measure, Conformal Prediction |