Automated feature extraction methods were implemented to replicate the important visual classification characteristics of the defect and noise used by inspectors. Both global and local image analysis methods were introduced for quantitative evaluation. A methodology was also developed for merit assessment of particular classification features in combination with the most promising signal processing algorithms. The protocol also addresses the evaluation of the best combination of features and signal processing algorithms for optimal discrimination. This analysis protocol was applied to a case study data set and the performance of both the feature extraction measures and the signal processing algorithms was studied where several signal processing methods were found to improve the discrimination of defects. Future work is planned to present more quantitative data on all of the results of this study. Several questions remain concerning the evaluation of the algorithms for greatest sensitivity to real defects. Additional work is needed to evaluate the impact of irregular void profiles on signal processing algorithm performance. Prominent edge (ring) features around voids will likely not be present for irregular voids. Second, additional signal processing development may be needed to address new inspection requirements for detection and characterization of delaminations, cracks and crushed foam. Lastly, quantitative evaluation and fusion of methods including THz, backscatter X-ray, millimeter wave, and shearography for optimum sensitivity is of interest.
Protocol and assessment of signal processing and feature extraction methods for terahertz NDE for spray-on foam insulation
Protokoll und Bewertung der Signalverarbeitung und Merkmalsextraktionsverfahren für die Terahertz-ZfP an aufgesprühten Schaumisolierungen
2007
8 Seiten, 4 Bilder, 12 Quellen
Aufsatz (Konferenz)
Datenträger
Englisch
Multiple-Nozzle Spray Head Applies Foam Insulation
NTRS | 1993
|Assessment of Foam Materials for Cryotank Insulation
SAE Technical Papers | 2000
|NTRS | 1989
|