Recurrence Quantification Compared to Fourier Analysis for Ultrasonic Non-Destructive Testing of Carbon Fibre Reinforced Polymers
Veröffentlichungsdatum
2020-06-12
Autoren
Betreuer
Gutachter
Zusammenfassung
Recurrence quantification analysis (RQA) is investigated on an application of ultrasonic testing on carbon fibre reinforced polymer (CFRP) and compared to methods in time domain and Fourier analysis. Porosity shall be detected with ultrasonic pulse-echo testing.
The work aims at creating an equivalent to the established criterion, the height of the back-wall echo (BWE), which is not available in the inspection of certain CFRP structures. A back-wall echo equivalent (BWE-equivalent) shall be generated out of the time series that consist of intermediate echoes from the inside of CFRP (carbon fibre reinforced polymer) specimens. The results depend largely on whether a resonance effect with ply thickness approximately half the main wavelength occurs, leading to rather regular, sinusoidal intermediate echoes.
A classification into porous or non-porous based on the BWE-equivalent created in this thesis is possible: for a unidirectional CFRP with resonance effect, linear regression on the amplitude spectrum provided excellent cross-validation results with balanced accuracies - mean of true positive (detection) rate and true negative rate (1 minus false alarm rate) - greater than 96%. Applying a Hann window on the intermediate echo time series prior to Fourier transform is essential, and linear regression relies heavily on frequencies above the main frequencies, once leakage is reduced through the Hann window. For a unidirectional material without resonance effect worse but still promising balanced accuracies from approximately 81% to 91% were achieved with RQA, Euclidean distance, feature RATIO, as well as with linear regression on Amplitude spectrum with Hann window.
An ultrasonic inspection with a frequency leading to a resonance effect plus linear regression on the amplitude spectrum after applying a Hann window is recommended for industrial application to classify into porous and non-porous without access to a back-wall echo.
The work aims at creating an equivalent to the established criterion, the height of the back-wall echo (BWE), which is not available in the inspection of certain CFRP structures. A back-wall echo equivalent (BWE-equivalent) shall be generated out of the time series that consist of intermediate echoes from the inside of CFRP (carbon fibre reinforced polymer) specimens. The results depend largely on whether a resonance effect with ply thickness approximately half the main wavelength occurs, leading to rather regular, sinusoidal intermediate echoes.
A classification into porous or non-porous based on the BWE-equivalent created in this thesis is possible: for a unidirectional CFRP with resonance effect, linear regression on the amplitude spectrum provided excellent cross-validation results with balanced accuracies - mean of true positive (detection) rate and true negative rate (1 minus false alarm rate) - greater than 96%. Applying a Hann window on the intermediate echo time series prior to Fourier transform is essential, and linear regression relies heavily on frequencies above the main frequencies, once leakage is reduced through the Hann window. For a unidirectional material without resonance effect worse but still promising balanced accuracies from approximately 81% to 91% were achieved with RQA, Euclidean distance, feature RATIO, as well as with linear regression on Amplitude spectrum with Hann window.
An ultrasonic inspection with a frequency leading to a resonance effect plus linear regression on the amplitude spectrum after applying a Hann window is recommended for industrial application to classify into porous and non-porous without access to a back-wall echo.
Schlagwörter
recurrence quantification analysis
;
ultrasonic testing
;
non-destructive testing
;
porosity
;
carbon fibre reinforced polymer
;
Fourier analysis
;
linear regression
;
state space
;
ultrasound
Institution
Fachbereich
Dokumenttyp
Dissertation
Zweitveröffentlichung
Nein
Sprache
Englisch
Dateien![Vorschaubild]()
Lade...
Name
dissertationCarstenBrandt.pdf
Size
15.49 MB
Format
Adobe PDF
Checksum
(MD5):03a5ed00ab7b568ce5ce9a35e725eb12