Open Access
Add to BibSonomy
Abstract
We propose a THz nondestructive analysis method based on multiple echoes and feature fusion. Conventionally, it is difficult to identify the debonding defects of the glue layer (II) due to the thin adhesive layer. To this end, a THz propagation model is established, and a quantitative method for determining the thickness of debonding defects based on multiple echoes is presented. The measurement error for preset defect thickness of 500 µm was 4%. Further, for determining the area of debonding defects, a feature fusion imaging algorithm is proposed to realize the lateral recognition of defects and quantitative analysis is used to improve the recognition ability of defects.
© 2020 Optical Society of America under the terms of the OSA Open Access Publishing Agreement
1. Introduction
Ceramic matrix composite (CMC) is a kind of composite material, which consists of a ceramic substrate reinforced with various kinds of fiber [1]. Owing to their unique characteristics such as high rigidity, high temperature resistance, low weight, and high corrosion resistance, CMCs are widely used in aerospace field [2,3]. They are often used to form a five-layer bonding structure [4] containing CMC, glue layer (I), insulation felt, glue layer (II), and metal substrate. Consequently, the hidden debonding defects are highly detrimental to its bonding strength, and the detection accuracy and precision of these defects are important criteria for determining the efficacy of CMC materials to protect the system. Presently, nondestructive testing (NDT) methods, such as ultrasonic C-scan [5], digital infrared thermography [6], and computed tomography (CT) [7], are widely used in the detection and evaluation of CMC. However, there are several limitations of these testing technologies: (1) the high porosity of CMC material leads to severe signal attenuation when ultrasonic NDT is used, and therefore the quality of glue layer cannot be evaluated effectively; (2) due to the low thermal conductivity of CMC materials, NDT methods such as infrared thermography cannot be used; (3) typically, the substrate of CMC bonding structure is a metal, which cannot be detected by transmission method, thereby limiting the application of X-ray methods such as CT [5–9]. Therefore, the accurate analysis of the bonding quality and defects of CMC bonding structure has been lacking in the literature for a long time.
THz NDT is a new technology, which has the advantages of strong penetrability, high-performance spectral fingerprinting, low energy consumption, harmlessness, environmental friendliness, and suitability for the evaluation of non-polar materials [10]. It is considered to be the only nondestructive technology that can be used for the evaluation of the bonding quality of CMC bonding structure [11].
Until now, several studies have focused on the NDT of CMC bonding structure using THz-TDS. Becker et al. [12] used fully electronic THz sources to investigate wound highly porous oxide composite (WHIPOX) and compared the measured results with those obtained using established NDT methods like X-ray micro-CT, lock-in thermography, and air coupled ultrasound inspection to validate the potential of their method. Ullmann et al. [13] proved the reliability of THz imaging technology for measurements on WHIPOX material. Owens et al. [14] used THz time-domain reflection imaging for nondestructive characterization of the variations in the properties of ceramic composite materials caused by mechanical and thermal strain. Ren et al. [4] proposed a neural network intelligent recognition algorithm based on THz-TDS to detect the upper and lower debonding defects of CMC. Wu et al. [9] used time delay mapping and maximum correlation amplitude mapping to quantitatively detect the hidden defects, delamination, or moist areas in oxide-oxide CMC materials. Dai et al. [15] proposed a continuous wavelet transform method to process the three-dimensional (3D) data of THz detection and used wavelet coefficients to reconstruct the 3D image of the detection sample. Consequently, they detected the defects in the glue layer with a high resolution.
However, there are very few reports on the NDT of CMC composites with a debonding structure using THz technology. In particular, the quantitative NDT of glue layer (II) of CMC adhesive structure is rarely discussed in the existing literature. In addition, most of THz NDT methods are based on a single feature to identify the defects, which cannot provide the complete information of defects, and the recognition ability is weak. Further, with the development of autoclave curve technology in aviation sector, the thickness of the glue layer of CMC bonding structure continues to decrease, and the defect features cannot be effectively identified by using continuous wavelet transform (CWT) only [15], therefore a more practical and quantitative method for defect detection is urgently needed. Moreover, the glue layer (II) is usually bonded to the spacecraft matrix, which also supports the necessity for the quantitative detection of debonding.
In this study, we have proposed an efficient method for analyzing the thickness of debonding defects in glue layer (II) of the CMC bonding structure. The wavelet coefficients are used to reconstruct the THz pulse signal, and the thickness of the defects is calculated by the time-of-flight difference between the multiple and base echoes. Further, a novel feature fusion imaging method is presented for the detection of debonding area. This method is firstly employed for NDT of CMC bonding structure by THz-TDS. Further, it is used to combine multiple THz detection images with different features into a single one to improve the imaging quality and to avoid the shortcomings related to imaging with different features. Overall, the debonding defects in glue layer (II) are quantitatively detected in terms of two parameters: defect thickness and defect area.
2. Experimental methods
2.1 THz-TDS imaging system
We developed a THz-TDS system for defect analysis, which is schematically shown in Fig. 1. The laser pulse generated by a mode-locked Ti sapphire femtosecond laser is divided into pump and probe pulses by a 60:40 beam splitter. The stronger pump pulse (60%) passes through a chopper with a frequency of 1.11 kHz and stimulates a THz emitter composed of a photoconductive antenna (PCA) to generate THz radiation. The THz radiation is collimated and focused onto the sample and is then re-collimated and focused onto a THz receiver by off-axis parabolic mirrors. The time trace of THz pulse signal is detected by the receiver and sampled by adjusting the optical path difference between the pump and probe pulses with a computer-controlled mechanical stage. The width of the detected THz spectrum in the THz-TDS system is 0.2-2.5 THz. Further, the spectral resolution is 3.1 GHz, signal-to-noise ratio (SNR) is greater than 70 dB, fast scanning range is 160 ps, and the time resolution is 0.1 ps. The PCA transmitter and receiver are connected through a collinear adapter and fixed on a two-dimensional (2D) guide rail. Based on the movement of the 2D guide rail, the time-domain spectrum of the sample can be recorded point by point.
2.2 Sample
The dimensions of the sample and defects are shown in Fig. 2. The sample is a five-layer bonding structure, which is consistent with the actual situation. Firstly, room-temperature curing adhesive was evenly applied to insulation felt, and polytetrafluoroethylene (PTFE) film was then used to preset four defects. The four defects were placed at the top, bottom, left, and right sides of the adhesive layer (II), and their thicknesses were 200, 350, and 500 µm, respectively. The insertion depth of the mark was extracted after curing to simulate the air debonding defect [16].
2.3 Simulation analysis
The THz wave is reflected when it passes through different layers of the CMC bonding structure, and the peaks and valleys can be alternately obtained for different flight times of the THz time-domain waveform. To effectively recognize the peak and valley values, the Gauss window deconvolution method was used to filter the time-domain waveform [17]. Figure 3 shows the actual waveforms of glue layer (I) and glue layer (II), which include debonding defects, and the actual waveform without debonding defects. The debonding defect in the glue layer (I) is represented by the valley from the upper interface of the defect and the peak from the lower interface of the defect, and it can be clearly identified. The debonding defect in the glue layer (II) is represented by deepening of the valley. Compared to the defect in glue layer (I), it is more difficult to identify this defect due to the presence of side lobes, scattering, and dispersion.
Fig. 3. Schematic of the THz time-domain waveform when the THz wave propagates through the CMC bonding structure.
To examine the feasibility of calculating the debonding thickness of the glue layer (II) using multiple echoes for selecting the effective imaging method, the defect features and multi-echo characteristics were simulated. The modeling of CMC sample was based on the study by Duvillaret et al. [18], where debonding defects in glue layer (II) were considered with different thicknesses. The propagation of THz waves is schematically shown is shown in Fig. 4. Here, the echo of each interface is recorded as ${E_1}$, ${E_2}$, ${E_3}$, ${E_4}$, ${E_5}$, ${E_6}$, ${E_{fp}}$. The reflected THz wave from the sample with debonding defect can be expressed as follows:
The bold symbols ${{\textbf E}_{\textbf 5}}$, ${{\textbf E}_{\textbf 6}}$, ${{\textbf E}_{{\textbf {fp}}}}$ in Eq. (1) are the characteristics of debonding defects, where ${{\textbf E}_{\textbf 5}}$ is the valley of debonding defect, ${{\textbf E}_{\textbf 6}}$ is the peak from the lower surface of debonding defect to the metal plate, and ${{\textbf E}_{{\textbf {fp}}}}$ is the multiple echo, which is caused by Fabry-Perot oscillations. According to Fresnel's law, the greater the difference in the refractive indices between two media, the higher the reflection energy. As the number of reflections increases, the reflection energy is considerably reduced. Combined with the simulation results, the higher-order echoes (greater than second-order echo) are very weak. Therefore, only the secondary echo generated by debonding defects is considered in this paper.
The simulated waveforms are shown in Figs. 5(a)–(c), where the thicknesses of debonding defects are 200, 350, and 500 µm, respectively. According to the Eq. (2), as the defect thickness of glue layer (II) decreases, the time-of-flight difference between multiple echoes and base echo becomes smaller.
Fig. 5. Simulated THz time-domain waveforms for debonding defects with thickness of (a) 200 µm, (b) 350 µm, and (c) 500 µm.
At the same time, according to the Fabry-Perot effect [19], the thickness of debonding defects can be calculated by using the difference between the flight times of the secondary echo and the base metal plate echo as follows:
where $\Delta t$ is the time difference between the multi-echo and metal echo, d is the thickness of glue layer (II), and n is the refractive index of the defect (approximately equal to the refractive index of air).Moreover, it is clear from Fig. 5 that the debonding defect is not only characterized by the multi-echo signal but also by the deepening of the valley in front of the metal substrate. Therefore, the optimum imaging effect can be obtained by minimum imaging and amplitude imaging of the valley and characteristics of the multiple echoes.
3. Data analysis
In traditional THz imaging algorithm, the time-domain waveform is first acquired for each space point of the sample, and then the corresponding frequency-domain waveform is obtained by applying Fourier transform. According to the specific information in the time-domain or frequency-domain waveform such as the maximum and minimum peak amplitudes, the image is established pixel by pixel. Generally, the information to produce the best imaging effect varies from sample to sample, and therefore it must be individually determined for each sample [20]. To this end, we propose a feature fusion imaging method, which combines the advantages of various feature imaging methods. This method is schematically shown in Fig. 6. It can be divided into four steps:
- A. As shown in the Fig. 6, for the debonding defect of glue layer (II), a, b, and c can be used for imaging, where a and b are the amplitudes of two characteristic positions after deconvolution, and c is the amplitude of the secondary echo after deconvolution.
- B. The image contrast is calculated by Weber contrast [21] and root mean square (RMS) contrast [22], which are expressed as follows:
Here, $Cw$ and $C\sigma$ are the Weber contrast and RMS contrast, respectively. ${L_t}$ represents the average brightness of the target image (prefabricated defect area), and ${L_b}$ represents the average brightness of the image background (no prefabricated defect area); ${I_{w \times h}}$ denotes an image whose width is $w$ and height is $h$, $I(x,y)$ is the value of a pixel at a specified position in the image, ${\mu _{{I_{w \times h}}}}$ is the brightness of the image (average value of all the pixels).
- C. The Weber contrast and RMS contrast of different imaging methods are normalized, and the average value of normalized coefficient is then calculated to obtain the final weight of each image. where $C{w_i}$ and $C{\sigma _i}$ are the Weber contrast and RMS contrast of each image, respectively. ${\overline {Cw} _i}$ and ${\overline {C\sigma } _i}$ are the normalized Weber contrast and RMS contrast of each image, respectively, and ${k_i}$ is the weight factor used in image fusion.
- D. The appropriate weight coefficient is applied in each image and a linear superposition of these images is used to realize fusion imaging, i.e., where $I{M_i}(m,n)$ is the normalized image obtained from the different imaging methods, ${k_i}$ is the coefficient of different imaging methods, and $\sum\limits_{i = 1}^n {{k_i}} = 1$. $I{M_{best}}(m,n)$ is the image obtained by feature fusion imaging.
4. Results and discussion
4.1 Analysis of defect thickness
Here, we examine the proposed quantitative analysis based on multiple echoes. The THz-TDS system is used to collect the THz signal of the CMC bonding structure, as shown by black curves in Fig. 7.
Fig. 7. Time-domain waveform reconstructed by CWT for defect thickness of (a) 200 µm, (b) 350 µm, and (c) 500 µm.
For the CMC bonding structure in which the thickness of debonding defects is 500 µm, the debonding thickness can be directly obtained from the position of multiple echoes. However, when the thickness of the debonding layer is small, e.g., 200 µm and 350 µm, the multi-echo signal is usually mixed with the substrate echo signal, as shown in Figs. 7(a)–7(b).
Therefore, the CWT method is utilized for separating the THz multi-echo signal and substrate echo signal to obtain the accurate position of the corresponding peak. The wavelet transform is expressed as follows [23]:
It is clear in Fig. 7 that after CWT, the reflection peaks become narrower and the tiny interface peaks are enhanced. When the thickness of the debonding defect is 200 or 350 µm, a good peak separation is realized by CWT, while for 500µm thickness of the debonding defect, this separation effect is enhanced by using CWT as compared to the directly obtained result based on multiple echoes.
The defect thickness of glue layer (II) can be determined according to Eq. (4). CMC bonding structures with different thicknesses of debonding defects in the glue layer (II) were sampled; 100 sets of time-domain waveforms were randomly selected around the defect region. The mean values and errors of the calculated defect thicknesses are shown in Table 1.
Table 1. Calculated vs. Actual Defect Thickness
Table 1 shows that the calculated thickness is 513 and 378 µm for the preset defect thickness of 500 and 350 µm, respectively, which indicates a good agreement between the calculated and actual values. However, when the thickness of debonding defects is 200 µm, the detection accuracy is low because the multi-echo features are annihilated by the base echo. Therefore, when the defect thickness is higher than 350 µm, the thickness of the defects can be accurately obtained using the time delay between the multiple echoes and the base echo, and the quality of the adhesive film can be precisely assessed.
4.2 Analysis of defect area
An image of the extracted PTFE film is shown in Figs. 8(a), 9(a), and 10(a). Here, the tape is used to mark the insertion depth for calculating the specific area of the debonding defect. The sample images including debonding defects with thicknesses of 200, 350, and 500 µm, which are obtained using different imaging methods, are shown in Fig. 8, Fig. 9 and Fig. 10, respectively.
Fig. 8. Different images of debonding defects with a thickness of 200 µm: (a) physical image of sample defects, (b) minimum imaging, (c) amplitude imaging, (d) multi-echo amplitude imaging, (e) minimum imaging after CWT, (f) amplitude imaging after CWT, (g) multi-echo amplitude imaging after CWT, and (h) feature fusion imaging.
Fig. 9. Different images of debonding defects with a thickness of 350 µm: (a) physical image of sample defects, (b) minimum imaging, (c) amplitude imaging, (d) multi-echo amplitude imaging, (e) minimum imaging after CWT, (f) amplitude imaging after CWT, (g) multi-echo amplitude imaging after CWT, and (h) feature fusion imaging.
Fig. 10. Different images of debonding defects with a thickness of 500 µm: (a) physical image of sample defects, (b) minimum imaging, (c) amplitude imaging, (d) multi-echo amplitude imaging, (e) minimum imaging after CWT, (f) amplitude imaging after CWT, (g) multi-echo amplitude imaging after CWT, and (h) feature fusion imaging.
In the above figures, the red and yellow regions represent the debonding defect regions, and the blue regions indicate the normal (non-defective) regions. In fact, the three imaging methods selected in this paper are used to select a specific amplitude for imaging. The selection of different characteristic peaks leads to different imaging effects. It can be seen in Figs. 8(b) and (e), Figs. 9(b) and (e), and Figs. 10(b) and (e) that minimum imaging [24] can efficiently recognize the edge of debonding defects, but it has a high rate of misjudgment in the normal areas. As shown in Figs. 8(c) and (f), Figs. 9(c) and (f), and Figs. 10(c) and (f), although amplitude imaging [25] exhibits better recognition ability, it causes more noise. After CWT, the overall image smoothness is improved, but the edge recognition is still poor. Further, multi-echo amplitude imaging is considerably affected by the thickness of the defect. For the debonding defect with a thickness of 500 µm, good imaging results are achieved using this method, as shown in Figs. 10(d) and (g). Compared with minimum imaging, the misjudgment rate for non-defective areas is greatly reduced in feature fusion imaging, as shown in Fig. 8(h), Fig. 9(h), and Fig. 10(h). Further, compared with amplitude imaging, the recognition of debonding edges and the contrast of non-defective areas are improved. Furthermore, feature fusion imaging exhibits good imaging effect for defects with different thicknesses.
The actual area of defects in the upper, lower, left, and right sides was recalculated through the extracted PTFE film. Subsequently, binarization was used to calculate the area of defects in each slice of the image to validate the higher accuracy of the feature fusion imaging over physical imaging. The results are shown in Table 2.
According to Table 2, the total error is 139.0 mm2 and the percentage error is 10.9% when the thickness of debonding defect is 200 µm, and the percentage error is 11.4% when the thickness of debonding defect is 350 µm. When the thickness of debonding defect is 500 µm, the total error is 120.6 mm2 and the percentage error is 6.78%. These results prove the high stability and accuracy of feature fusion imaging for the characterization of debonding defects and for calculating the area of debonding defects. According to Table 3, the areas of debonding defects calculated by different imaging methods are quite different. Fusion imaging is superior to the other imaging methods for debonding defects of different thickness with good recognition ability of defect area.
Table 2. Defect Area Calculated Using Feature Fusion Imaging.
Table 3. Defect Area Calculated Using Different Imaging Methods.
5. Conclusions
A pulsed THz-TDS system was employed to detect the debonding defects in the glue layer (II) of a five-layer CMC bonding structure. Firstly, the feasibility of multi-echo analysis for determining the thickness of defects was validated by simulations, and the effective feature imaging method was selected. Secondly, another echo analysis method based on CWT was used to measure the thickness of debonding defects in the glue layer (II), which circumvented the issues due to dispersion, scattering, and aliasing. The maximum percentage errors for the debonding defects with thicknesses of 200, 350, and 500 µm were obtained as 40.5%, 8%, and 2.6% respectively. This indicates that the method exhibited a good thickness recognition ability for debonding defects with thicknesses of 350 and 500 µm. For lateral detection of defects, fusion feature imaging was conducted with a variety of imaging methods, which improved the imaging contrast and the accuracy of defect recognition. Finally, the area of debonding defects with thicknesses of 200, 350, and 500 µm was calculated. The maximum percentage errors for the calculated area with respect to the actual area were 10.9%, 11.4%, and 6.78%, respectively.
Overall, we proposed an effective quantitative method for thickness measurement based on multiple echoes to address the outstanding issue of weak defect identification due to thin bonding layer and debonding defects. Further, the proposed feature fusion imaging algorithm realized lateral recognition of defects, and quantitative analysis was used to further improve the recognition ability of defects. In future, we hope to utilize additional signal processing methods to further analyze the defect signals, and image processing algorithm will be modified for further improving the recognition accuracy of defective and non-defective areas.
Funding
National Demonstration Center for Experimental Opto-Electronic Engineering Education; Higher Education Discipline Innovation Project (D17017).
Disclosures
The authors declare no conflicts of interest.
References
1. S. Schmidt, S. Beyer, H. Knabe, H. Immich, R. Meistring, and A. Gessler, “Advanced ceramic matrix composite materials for current and future propulsion technology applications,” Acta Astronaut. 55(3-9), 409–420 (2004). [CrossRef]
2. R. Naslain, “Design, preparation and properties of non-oxide CMCs for application in engines and nuclear reactors: an overview,” Compos. Sci. Technol. 64(2), 155–170 (2004). [CrossRef]
3. J. Lincoln, B. Jackson, A. Barnes, A. Beaber, and L. Visser, “Oxide-oxide ceramic matrix composites-enabling widespread industry adoption,” in Advances in High Temperature Ceramic Matrix Compo sites and Materials for Sustainable Development, Ceramic Transactions, Volume 263, M. Singh, T. Ohji, S. Dong, D. Koch, K. Shimamura, B. Clauss, B. Heidenreich, and J. Akedo, eds. (Wiley, 2017), pp. 401–412.
4. J. Ren, L. Li, D. Zhang, X. Qiao, Q. Lv, and G. Cao, “Study on intelligent recognition detection technology of debond defects for ceramic matrix composites based on terahertz time domain spectroscopy,” Appl. Opt. 55(26), 7204–7211 (2016). [CrossRef]
5. J. Dong, B. Kim, A. Locquet, P. McKeon, N. Declercq, and D. S. Citrin, “Nondestructive evaluation of forced delamination in glass fiber-reinforced composites by terahertz and ultrasonic waves,” Composites, Part B 79, 667–675 (2015). [CrossRef]
6. U. B. Halabe, A. Vasudevan, P. Klinkhachorn, and H. V. GangaRao, “Detection of subsurface defects in fiber reinforced polymer composite bridge decks using digital infrared thermography,” Nondestr. Test. Eval. 22(2-3), 155–175 (2007). [CrossRef]
7. H. Ullah, A. R. Harland, and V. V. Silberschmidt, “Experimental and numerical analysis of damage in woven GFRP composites under large-deflection bending,” Appl. Compos. Mater. 19(5), 769–783 (2012). [CrossRef]
8. C. Stoik, M. Bohn, and J. Blackshire, “Nondestructive evaluation of aircraft composites using reflective terahertz time domain spectroscopy,” NDT&E Int. 43(2), 106–115 (2010). [CrossRef]
9. D. Wu, C. Haude, R. Burger, and O. Peters, “Application of terahertz time domain spectroscopy for NDT of oxide-oxide ceramic matrix composites,” Infrared Phys. Technol. 102, 102995 (2019). [CrossRef]
10. K. Kawase, T. Shibuya, S. I. Hayashi, and K. Suizu, “THz imaging techniques for nondestructive inspections,” C. R. Phys. 11(7-8), 510–518 (2010). [CrossRef]
11. S. Costil, S. Lukat, P. Bertrand, C. Langlade, and C. Coddet, “Surface treatment effects on ceramic matrix composites: Case of a thermal sprayed alumina coating on SiC composites,” Surf. Coat. Technol. 205(4), 1047–1054 (2010). [CrossRef]
12. S. Becker, T. Ullmann, and G. Busse, “3D Terahertz imaging of hidden defects in oxide fibre reinforced ceramic composites,” in 4th International Symposium on NDT in Aerospace, (2012)
13. T. Ullmann, Y. Shi, N. Rahner, M. Schmücker, P Fey, G. Busse, and S. Becker, “Quality assurance for the manufacturing of oxide fiber reinforced ceramic composites for aerospace applications,” in International Symposium on NDT in Aerospace, (2012).
14. L. Owens, M. Bischoff, A. Cooney, D. T. Petkie, and J. A. Deibel, “Characterization of ceramic composite materials using terahertz reflection imaging technique,” in 2011 International Conference on Infrared, Millimeter, and Terahertz Waves, (IEEE, 2011), pp. 1–2.
15. B. Dai, P. Wang, T. Y. Wang, C. W. You, Z. G. Yang, K. J. Wang, and J. S. Liu, “Improved terahertz nondestructive detection of debonds locating in layered structures based on wavelet transform,” Compos. Struct. 168, 562–568 (2017). [CrossRef]
16. I. Frydrych, F. Xu, Q. D. Duan Mu, L. J. Li, D. Yang, and B. Xia, “Nondestructive evaluation of rubber composites using terahertz time domain spectroscopy,” Fibres Text. East. Eur. 26(1(127)), 67–72 (2018). [CrossRef]
17. J. Takayanagi, H. Jinno, S. Ichino, K. Suizu, M. Yamashita, T. Ouchi, S. Kasai, H. Ohtake, H. Uchida, N. Nishizawa, and K. Kawase, “High-resolution time-of-flight terahertz tomography using a femtosecond fiber laser,” Opt. Express 17(9), 7533–7539 (2009). [CrossRef]
18. L. Duvillaret, F. Garet, and J. L. Coutaz, “A reliable method for extraction of material parameters in terahertz time-domain spectroscopy,” IEEE J. Sel. Top. Quantum Electron. 2(3), 739–746 (1996). [CrossRef]
19. M. Scheller and M. Koch, “Fast and accurate thickness determination of unknown materials using terahertz time-domain spectroscopy,” J. Infrared, Millimeter, Terahertz Waves 30(7), 762–769 (2009). [CrossRef]
20. J. Zhang, J. Chen, J. Wang, J. Lang, J. Zhang, Y. Shen, H. L. Cui, and C. Shi, “Nondestructive evaluation of glass fiber honeycomb sandwich panels using reflective terahertz imaging,” J. Sandwich Struct. Mater. 21(4), 1211–1223 (2019). [CrossRef]
21. W. F. Schreiber, Fundamentals of Electronic Imaging Systems: Some Aspects of Image Processing (Springer, 1993), pp. 60–70.
22. R. A. Frazor and W. S. Geisler, “Local luminance and contrast in natural images,” Vision Res. 46(10), 1585–1598 (2006). [CrossRef]
23. M. Rucka and K. Wilde, “Application of continuous wavelet transform in vibration based damage detection method for beams and plates,” J. Sound Vib. 297(3-5), 536–550 (2006). [CrossRef]
24. J. Ren, D. Zhao, and L. Li, “The coating curing properties study using terahertz time domain spectroscopy,” Proc. SPIE 9677, 967726 (2015). [CrossRef]
25. R. F. Anastasi, E. I. Madaras, J. P. Seebo, S. W. Smith, J. K. Lomness, P. E. Hintze, C. C. Kammerer, W. P. Winfree, and R. W. Russell, “Terahertz NDE application for corrosion detection and evaluation under shuttle tiles,” Proc. SPIE6531, 65310W (2007). [CrossRef]
