·Table of Contents
·Aeronautics and Aerospace
Automatic estimation of ultrasonic attenuation for porosity evaluation in composite materials
L.-K. Shark and C. Yu
Department of Engineering and Product Design
University of Central Lancashire
Preston, PR1 2HE, UK
J. P. Smith
Warton, Preston, PR4 1AX, UK
To speed up the evaluation of the porosity content in composite materials, which is currently based on time- consuming manual measurement of the amplitude values of ultrasonic echoes, a novel method for automatic estimation of ultrasonic attenuation is proposed in the paper. The proposed method consists of two stages, namely, signal restoration followed by attenuation estimation. In the signal restoration stage, the regions containing ultrasonic echoes are first identified automatically, and the wavelet based denoising method is then applied with filter length optimised and threshold fuzzified to minimise the changes in the echo regions and to zero the fluctuations outside the echo regions. In the attenuation estimation stage, the porosity content is evaluated based on the relative amplitude and energy values of the two denoised echoes reflected respectively from the top surface and the bottom surface of the component under test. The effectiveness of the proposed method is demonstrated using practical ultrasonic signals acquired from a number of purpose-made laminated composite material samples with known porosity levels.
Porosity is a defect that appears as small interlaminar void in composite structures. The presence of excessive porosity can severely affect the mechanical strength of composite structures under loading due to stress concentrations. Consequently, in the aerospace industry, ultrasonic measurement is employed as a routine non- destructive testing (NDT) technique to evaluate the porosity level in an aircraft component manufactured using composite materials. The process involves transmission of high-intensity ultrasonic waves through the component under test and time consuming manual assessment of the amplitude values of returned echoes by experienced inspectors [1,2]. To reduce the inspection time, to increase the consistency in porosity estimation, and to de-skill the task of porosity evaluation by allowing the use of non-specialist operators, a novel method is proposed in this paper for automatic estimation of ultrasonic attenuation.
Ultrasonic signals are noisy in nature due to the back scattering phenomenon produced by the inherent microstructure of the material. From the previously published literature, the advantages of using the wavelet transform for signal restoration have been well-established . The approach is based on thresholding the wavelet coefficients produced by the forward discrete wavelet transform (DWT) of the ultrasonic signal, thereby
allowing only significant wavelet coefficients (with their magnitude values greater than the threshold) to be used in the inverse DWT (IDWT) to suppress noise. However, the quality of the ultrasonic echoes restored depends heavily on the choice of wavelet, its order and threshold value. Presented in Section 2 is a novel signal restoration method, whereby an automatic signal segmentation process is included before denoising to facilitate the optimisation of the wavelet order, and the dilemma in setting an appropriate threshold is solved by using a
special fuzzy membership function . Normally, the porosity level is evaluated based on the amplitude ratio of the second echo reflected from the bottom surface of the component with respect to the first echo reflected from the top surface of the component. Section 3 presents the validity of using the energy ratio as an additional measure to improve the reliability of porosity estimation. To demonstrate the effectiveness of the proposed method, some experimental results based on practical ultrasonic signals acquired from purpose-made samples are presented in Section 4 with some general conclusions given in Section 5.
2. Signal Restoration
There are two steps used in signal restoration which can be described as
- Automatic signal segmentation to determine the regions containing the ultrasonic echoes based on the statistical properties.
- Wavelet based denoising with filter length optimised and threshold fuzzified to minimise the difference between the denoised echoes and the original echoes in the identified echo regions and to minimise the amplitude values of the ultrasonic signal outside the identified echo regions.
2.1 Automatic Signal Segmentation
Figure 1 shows a typical ultrasonic record acquired from a particular point of a laminated composite material part. It is seen to consist of a transmitted signal followed by two echoes reflected from the top and bottom surfaces respectively. While the transmitted signal is seen to possess the strongest amplitude, the second echo is seen to be weaker than the first echo due to material attenuation of the ultrasonic energy. Furthermore, between these signals is noise due to back scattering.
Fig 1: A typical ultrasonic record |
The aim of automatic signal segmentation is to separate the transmitted signal and echoes from noise, and is achieved by exploiting the differences between noise and the transmitted signal/echoes in term of their statistical properties. Since echoes and the transmitted signal are likely to be strongly correlated due echoes being produced by the transmitted signal, and since noise caused by back scattering is likely to be random due to the irregularity of the material microstructure, the following steps are devised for the automatic signal segmentation:
- Estimating the standard deviation of noise, s, based on the last 50 samples at the end of the ultrasonic record. If the measurement system is correctly set-up with the component under test being placed at a sufficient high level above the bottom of the water tank, these 50 samples will happen well after the second echo.
- Locating the transmitted signal by searching the sample position, P, corresponding to the
maximum amplitude value in the whole ultrasonic record.
- Identifying the end of the transmitted signal by comparing s with s1 (k) which is the standard deviation of the transmitted signal computed at position k based on 10 successive samples. With the comparison done on a point-by-point basis starting from position P, the point corresponding to s 1 (N0)£h 0s is recognised as the end of the transmitted signal, where h0 is a weighting factor.
- Locating the first echo by searching the sample position, I0 , corresponding to the maximum amplitude value, A0 , after the transmitted signal.
- Identifying the start of the first echo by comparing s with s2(k) which is the standard deviation of the first echo computed at position k based on 10 successive samples. The comparison is again done on a point-by-point basis starting from position I0 , but moving towards the start of the ultrasonic record. The point corresponding to s2(N1)£ max(h1s, r0A0 ) is recognised as the start of the first echo, where h1 and r0 are weighting factors. The reason of including r0A0 is to avoid the situation of failing to find the start of the first echo when s is too small.
- Identifying the end of the first echo based on the same strategy by searching from I0 towards the end of the ultrasonic record with the point corresponding to s2(N2)£ max(h2 s, r1A0) being recognised as the end of the first echo.
- Locating the second echo by searching the sample position corresponding to the maximum
amplitude value after the first echo and repeating (v) and (vi) with modified weighting factors to identify the start and the end of the second echo, N 3 and N 4 .
Figure 2 shows a result of the start and the end of the first and second echoes estimated by of the proposed automatic signal segmentation method. Although the echo intervals estimated are not precise, they do contain the main peaks of each echo. Furthermore, it should be apparent that the accuracy and the execution time of the proposed method can be improved by incorporating additional information such as component thickness, approximate ultrasonic velocity in the component, and the sampling rate of the ultrasonic measurement system.
Fig 2: Echo intervals identified |
2.2 Wavelet based denoising
With the positions of the two echoes being identified by the automatic signal segmentation method, a region based error criterion can be established for the following denoising operation. If the denoised signal is required to be as similar to the original signal as possible in the two identified echo intervals and as near to zero as possible outside the two identified echo intervals, then the error criterion can be written as
where s(i) is the original signal, is the denoised signal, and W denotes the echo regions.
The minimisation of the error criterion by the wavelet based denoising method depends on the choice of wavelet, its order and threshold value. After experimenting with a number of different wavelets on ultrasonic signals, the symlet wavelet was found to outperform other wavelets based on visual comparison due to its linear phase property. To select the optimum order of the wavelet, a search operation is carried out to perform denoising based on the wavelet filter bank implemented using the 2nd symlet wavelet up to the maximum 8th order symlet wavelet. The optimum filter length corresponds to the minimum e value produced. Although the search of optimum wavelet order may be viewed as a time-consuming process, it may be only necessary to do it once at the start of a ultrasonic scan or the search range can be limited to the adjacent orders of the optimal order previously estimated.
A number of different methods for threshold value selection are available. One is based on the Stein's unbiased risk estimate (SURE) , and the other is a fixed form threshold with some asymptotic near-minimax properties . While the latter is known as a conservative one as it tends to `overkill' the wavelet coefficients, the former tends to `underkill' the wavelet coefficients. Consequently, there is a problem in selecting between the SURE threshold and the fixed from threshold. By treating the SURE threshold and the fixed form threshold as the lower bound and the upper bound respectively, a possible compromise is use a fuzzy membership function to allow the wavelet coefficients lying between lower bound and the upper bound to contribute partially in the signal reconstruction according to their magnitude values. Let w m,n denote the nth wavelet coefficient at the mth decomposition stage with tmsure and tmfixed denoting the SURE threshold and the fixed form threshold respectively, an appropriate fuzzy threshold function proposed previously  is given by
and where D is a small constant to represent the membership value of almost zero.
For the original ultrasonic signal shown in dash line in Figures 3(a), the signal restored based on the proposed wavelet denoising method is shown in solid line and was obtained using the 8th order symlet wavelet (which was found to be the optimum order) with 5 decomposition stages. Compared with Figure 3(b) which shows the signal restored using the 3rd order symlet wavelet with the same number of decomposition stages and the fixed form threshold, the proposed denoising method is seen to offer better performance with the restored echoes fitting better to the original echoes and containing significantly less amplitude distortions.
|Fig 3: (a) Proposed denoising, (b) Standard denoising|
3. Attenuation Estimation
With noise significantly reduced in the echo regions, ultrasonic attenuation can be evaluated to estimate the porosity level. The evaluation is normally based on the peak amplitude ratio of the second echo reflected from the bottom surface of the component with respect to the first echo reflected from the top surface of the component. In order for the amplitude ratio to be a valid measure of the ultrasonic attenuation, the implicit assumption is that the second echo should have the same waveform shape as the first echo. This assumption is not always true in practice, because the material acts as a lowpass filter resulting in higher attenuation of the high frequency components contained in ultrasonic signals, especially in the case of employing high frequency ultrasonic transducers. Consequently, an energy ratio should provide a more appropriate measure of ultrasonic attenuation and it can be determined using
To investigate the accuracy of each measure under noisy environment, computer simulations were carried out to denoise a known ultrasonic signal embedded in the additive white Gaussian noise and to evaluate the amplitude ratio and the energy ration. In the computer simulations, the known ultrasonic signal was simulated using two Gaussian envelope sinusoidal signals to represent two echoes with the amplitude of the second echo being precisely half of that of the first echo. By averaging over 1000 simulation results, the amplitude ratio and the energy ratio as functions of signal-to-noise ratio (SNR) is shown in Figure 4(a). From Figure 4(a), the estimation error of both measures is seen to decrease as SNR increases. To yield an estimation error less than 5%, the denoising operation needs to be able to produce a SNR value greater than 7 dB. Figure 4(b) shows the variance of the estimated amplitude and energy ratios as functions of SNR. Compared with the energy ratio, although the amplitude ratio is seen to be closer to the true ratio from Figure 4(a), the variance of the amplitude ratio is seen to be much worse from Figure 4(b). Consequently, both measures are used in the evaluation of ultrasonic attenuation to increase the reliability of porosity estimation.
|Fig 4: (a) Attenuation estimated, (b) Variance of attenuation estimated|
4. Experimental Results
The approaches described in Sections 2 and 3 were applied to the practical ultrasonic signals acquired from purpose-made samples. The samples are laminated composite material plates with 3 mm thickness. The porosity levels along the edges of the samples have been identified by using the electron microscope to vary from grade A with less than 0.2% porosity to grade F with greater than 5% porosity (see Table 1). The ultrasonic signals were acquired by scanning over an area of 4 mm x 2 mm of each sample at 0.1 mm resolution using a focused transducer operating at 20 MHz.
|Table 1 :|
Using the middle values of each porosity range as the nominal porosity level for each grade, except grade A and grade F which are denoted by 0.1% and 7% respectively, Figure 5 shows the relationship between the porosity level and the ultrasonic attenuation estimated by averaging the amplitude ratio and the energy ratio obtained over the scan area. From Figure 5, the ultrasonic attenuation is seen to increase as the porosity level increases. Furthermore, the strong agreement between the amplitude ratio and the energy ratio provides the confidence in the results obtained.
Fig 5: Attenuation versus porosity |
Figure 6 shows two images corresponding to the amplitude and energy attenuation distributions of a sample. The agreements between the two images are again apparent. Based on the histogram of the attenuation distribution showing the number of pixels per each attenuation interval defined according to the specified porosity ranges, an accurate and consistent porosity grade can be automatically derived for components manufactured using composite materials.
|Fig 6: (a) Amplitude attenuation, (b) Energy attenuation|
The paper presented a novel method for porosity evaluation based on automatic estimation of ultrasonic attenuation. In order to provide a high accuracy and reliability, the positions of ultrasonic echoes are first identified based on their statistical properties, thereby facilitating the following denoising operation to be carried out with optimum wavelet order and a fuzzy threshold. Furthermore, the determination of the ultrasonic attenuation is based on not only the amplitude ratio but also the energy ratio between the two echoes. The proposed method was tested using practical ultrasonic signals acquired from purpose-made composite materials
with different porosity levels, and is seen to form a good basis for the practical implementation of a reliable and automatic porosity evaluation system in the future.
The work is funded by the European Community under the Industrial and Material Technologies Programme (Brite-EuRam III, Contract No:BRPR-CT98-805).
- McRae, K.I. and Zala, C.A.:"Attenuation and porosity estimation using the frequency-independent parameter Q", Review of Progress in Quantitative Nondestructive Evaluation, Vol.14, pp703-710, 1995.
- Daniel, I.M., Wooh, S.C. and Komsky, I.:"Quantitative porosity characterization of composite materials by means of ultrasonic attenuation measurements ", Journal of Nondestructive Evaluation, No.11, pp1-8, 1992.
- Strange,G., and Nguyen, T. Wavelets and Filter Banks, Wellesley-Cambridge Press, 1996.
- Shark, L.-K. and Yu, C.:" Denoising by optimal fuzzy thresholding in wavelet domain ", IEE Electron. Lett., Vol.36, pp581-582, 2000.
- Donoho, D.L. and Johnstone, I.M.:"Adapting to unknown smoothness by wavelet shrinkage ", J. Am. Statist. Ass., Vol.90, No.432, pp1200-1224, 1995.
- Donoho,D.L.:" De-noising by soft-thresholding ", IEEE Trans. Info. Theory, Vol.41, No.3, p613-627, 1995.