SIMULTANEOUS RECONSTRUCTION OF THE ACOUSTIC PROPERTIES OF A LAYERED MEDIUM: THE INVERSE PROBLEM v. K. Kinra, Y. Wang and C. Zhu Center for Mechanics of Composites Department of Aerospace Engineering Texas A&M University College Station, TX 77843 S. P. Rawal Martin Marietta Technologies, Inc. P. O. Box 179 Denver, CO 80201 INTRODUCTION In recent years, multi-layered metal materials have found increasing use in various disciplines, particularly in aerospace, electrical, automotive and pressure vessel industries. These applications utilize certain unique physical properties of the layered metals such as heat conductivity, electric conductivity or corrosion resistance. Thus, a nondestructive evaluation technique is essential for these advanced materials. In this work, we present a frequency domain nondestructive evaluation technique for these layered media and the results obtained when the technique is applied to the three-layer clad metal materials. For the layered media, significant amount of effort has been devoted to the forward problem: Given the true values of the acoustical properties of the individual layers, compare the theory and experiments [1-4]. More recently some effort has been devoted to the solution of the inverse problem by Kinra et al [5]. They developed an inversion scheme to simultaneously extract individual layer properties (thickness, wavespeed) from a single ultrasonic measurement. Successful inversion was obtained for up to four parameters. In their experimental work, however, they found out that, due to the water layer in the simulated layered medium, large discrepancies between the theory and the experiment exist at the resonant frequencies of the water layer. This phenomenon was explained by the possible leaky Lamb waves in the water layer. In this paper, we apply this technique to several commercially available clad metal specimens in which all individual layers are metallic materials. The objective is to investigate the practical applicability of the inversion technique to real-world problems. Review of Progress in QlIQ1IIitative Nondestruclive Evaluation. Vol. 14 Edited by D.O. Thompson and D.E. Chimenti. Plenum Press. New York.199S 1433
THEORY Consider a three-layered specimen occupying the space a < x < hi + ~ + h3 immersed in an elastic fluid (water) and insonified by a normally incident time-harmonic longitudinal plane wave. Here, we use the subscript ( )i to identify the immersion medium ( )0 and the layers ( )1. ( hand ( h The transfer function can then be defined as the ratio of the amplitude of the transmitted wave field to that of the incident wave field. Following the well-known Thomson-Haskell approach [6-7], the transfer function can be obtained as [3] where T=~ and IJ Z;+Zj (2) are the displacement transmission and reflection coefficients, respectively. Z refers to the acoustic impedance pc. As indicated in the parameter list, the complex transfer function depends on the frequency of the wave and the acoustical properties of the layers: density p, wavespeed c and thickness h. EXPERIMENTAL PROCEDURES The experimental setup used to measure the transfer function is shown in Fig. 1. First, we digitize the reference signal (with the specimen removed from the wave path) and the signal through the specimen. Next, the Fourier transforms of both the reference and specimen signals are computed using a Fast Fourier Transform (FFT) algorithm. This procedure decomposes the original time domain signals to their frequency components. According to the definition of the transfer function, for any frequency component, we have u u~' Waler Tan k DtCITA L OSCILOSCOPE P RSO A L COM P U T R Figure 1. Experimental setup. 1434
* FI(ro;) H (0) p)-- i' - - FR(O)) (3) where Fl. FR represent the Fourier transforms of the incident and transmitted signals, respectively. INVERSION PRINCIPLE The essence of our parameter inversion scheme is the widely used principle of least square: the best estimate of the parameters is the one which minimizes the difference between the theory and the experiment in the least square sense, i.e., the one that minimizes (4) Mathematically, the inversion is represented by a nonlinear minimization problem which can be numerically solved by using the Newton-Raphson method [8] or the simplex method [9]. An important issue in the inverse problem is to estimate the error ofthe parameters. We note that the errors of the parameters depend on the properties of the error function E UD. For example, if the error function is not sensitive to certain parameter, then that parameter cannot be expected to be accurately estimated. This provides the motivation for the sensitivity analysis and error surface study. SENSmVITY ANALYSIS AND ERROR SURFACE Let p represent anyone of the acoustical parameters of the specimen. the complex sensitivity of the transfer function to p is defined as Specimen: Ag/CulNi 50 40 30 20-9: :II ell 10 0-10 -20-30 -40 I I I I! ~ / I -50 o 2 4 6 8 1 0 1 2 1 4 1 6 1 8 20 Frequency (MHz) Figure 2a. Sensitivity of magnitude of H* to hi' 1435
Specimen: AglCulNi 40r-~--~~------~~------~~--, 20 f1 r--',.---""-' r'\ _2:-rrT~ -40-60 -80 o 2 4 6 8 10 12 14 16 18 20 Frequency OVUHz) Figure 2b. Sensitivity of phase of H* to hi' ah*(m) 'P SH" =P ā -,1' (5) Let (5 p be the absolute error in estimating the parameter p, let the normalized errors be defined as e H = (5H / ii, e$ = &P, and e p = (5 I' / p, where the subscript (") denotes a true value. For j= 1,2,3,... N, let m j be N discrete FFT frequencies, then through a lengthy but straightforward calculation it can be shown that (6) where S H.p and S $,1' are the sensitivity ofthe magnitude and phase of the transferfunction, respectively [5]. S p is the magnitude of the complex sensitivity defined in Eq.(5). Eq.(6) establishes the error propagation rule for single parameter reconstruction. It is clear from Eq.(6) that when the sensitivity is small, a small error in measuring H' (m) will result in a large error in estimating the parameter p, and vice versa. From the plots shown in Fig. 2a and Fig. 2b, the sensitivity obtain maximum values near the resonance peaks of the transfer function. Therefore, it is advisable that the useful frequency range for the inverse problem include several resonance peaks. For multi-parameter inversion, estimating the errors of the parameter is significantly more complicated due to the non-separable effect from various parameters. One way to estimate the errors is to geometrically study the error surface. For example, for a given level of error, say 1 %, Eq.(4) defines a closed surface in the parameter space. The size of the closed surface in each dimension will reflect the error of the associated parameter. The error can be approximately estimated by the linearization of the transfer function near the true values of the parameters. This linearization will 1436
Figure 3. A 3-D error surface plot (Specimen: 304ss/A1J304ss, E=O.OI). yield an ellipsoid for the error surface in the parameter space. For thre<;.. param~ter inversion problem a typical error ellipsoid is shown in Fig. 3 where Xi is defined as hi - hi with hi the true thickness of the i 'h layer. RESULTS AND DISCUSSIONS A. Comparison of the Reconstructed and Measured H* (ro) In any NDE procedures, the problem is usually divided into two stages. In the first stage, the forward problem, the acoustical parameters of the specimens are assumed to be known and the transfer function is computed. The theoretically predicted transfer function is then compared with the that obtained experimentally. In the inverse problem, on the other hand, the acoustical parameters are deduced from comparison between the theoretically and experimentally obtained transfer functions. Although the ultimate goal of any NDE process is to solve the inverse problem, the forward problem is worth studying for the following reasons: 1) The accuracy provided by the forward comparison gives the basis for error estimates in the inverse problem. 2) Forward problem comparison helps to identify the spurious wave phenomena such as the shear modes and leaky Lamb waves. For multi-layered clad metal specimens, only an approximate estimate of the individual thickness can be obtained, for example, using a microscope. Therefore, the forward problem in this study will proceed in the following fashion. The individual layer density and wavespeed are obtained by testing a single layer specimen of the same material. The density of the material is measured using Archimedes principle; the acoustic wavespeed is measured using the resonance technique [10]. The thicknesses of the single layer specimens can be accurately measured with a micrometer to a precision of about ±2.5 /-lm. It appears reasonable to assume that during the cladding process the density and the wavespeed remain unchanged from their bulk properties. To complete the forward problem, the thicknesses of the individual layers must be known. Here we choose to use the reconstructed thicknesses from the transfer function obtained by using a 10 MHz transducer (useful frequency range is 5 to 11 MHz). It 1437
Et:... 0 Q) '"'".E!.~ ~ 1.0' 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0 Specimen: 3 04ssl Al/3 04ss.-----THB--OR-y-------.I'I!5 MHz TRANSDUCER 10 MHz TRANSDUCER 20~zTRANSDUCER 2 4 6 8 1 0 1 2 1 4 1 6 1 8 20 Frequency (MHz) Figure 4. Comparison of the reconstructed and measured transfer function. is noted that even though we used thicknesses reconstructed from the 10 MHz transducer data, the comparison between the reconstructed and measured transfer functions still reflects the difference (residual) due to experimental errors. A typical comparison for the magnitude of the transfer function is shown in Fig. 4. Compared to the previous results where one of the layers was formed by water [5], the comparison is significantly improved near the resonance peaks. This suggests the elimination of leaky Lamb waves in solid layered medium. B. Inversion Results The first step in the design of the inverse problem is to select a frequency range based on the sensitivity analysis. In this work, we use three different pairs of transducers for our measurements, with a nominal frequency of 5 MHz, 10 MHz and 20 MHz, respectively; their useful bandwidth is 2-6 MHz, 5-11 MHz, and 11-20 MHz, respectively. In this study, we confine our attention to the reconstruction of individual layer thicknesses. In the inverse scheme, the acoustic parameters are found out by iteratively adjusting them such that the experimentally obtained transfer function is best matched by the theoretical prediction over the frequency range. Mathematically, the closeness ofthe match is quantified by the error function defined in Eq.(5). For the minimization problem, we choose to use the simplex approach. We started with an initial guess of the acoustical parameters. The algorithm then continually modifies the parameters and the error function is becomes progressively smaller until a preset convergence criterion is satisfied. Typical convergence paths for various initial guesses are shown in Fig. 5. The inversion results for the specimens tested are shown in Tables 1-3. CONCLUSIONS A frequency domain technique for multi-layered clad metals is described. Using this technique, the individual layer thicknesses of three layered clad metals can be simultaneously measured from a single ultrasonic experiment. Based on the sensitivity analysis and 3D plot of the error surface, the errors for the measured parameters can be estimated when the error of the transfer function is known. When the technique was applied to several real clad metal layered samples, with 1438
<> Initial Guesses o Converged Values Figure 5. Typical convergence paths for different initial guesses. Table 1. Inversion Results for Specimen AgiCu/Ni: Mean value±precision (htotal=1.768 ± 5 Ilm) Nominal Frequency hl h2 ha htotal (MHz) (mm ± Ilm) (mm ±Ilm) (mm ± Ilm) (mm) 5 0.311 ± 1 1.162± 1 0.286 ± 1 1.759 10 0.314 ± 9 1.144 ± 9 0.306 ±3 1.764 20 0.313±14 1.157 ±24 0.292 ±8 1.762 Table 2. Inversion Results for Specimen 304ss/AU304ss: Mean value±precision (htotal=2.277 ± 5 Ilm) Nominal Frequency hl h2 ha htotal (MHz) (mm ± Ilm) (mm ± Ilm) (mm ± Ilm) (mm) 5 0.375 + 1 1.524 ± 1 0.375 ± 1 2.274 10 0.387 ± 1 1.523 ± 1 0.363 ± 1 2.273 20 0.378± 3 1.521 ± 5 0.362 ±28 2.261 1439
Table 3. Inversion Results for Specimen 436ss/A1J304ss: Mean value±precision(htotal=2.657 ± 5 11m) Nominal Frequency h1 h2 h3 htotal (MHz) (mm ± 11m) (mm ± 11m) (mm ± 11m) (mm) 5 0.439 ± 2 1.815 ± 1 0.398 + 1 2.652 10 0.434 ± 1 1.828±2 0.384 ± 1 2.646 20 0.432 ± 4 1.823 ± 5 0.389 ± 8 2.649 three transducers of different center frequencies, repeatable (with a typical precision of a few 11m) and consistent results are obtained. ACKNOWLEDGMENTS This paper is based on the work supported by the Texas Advanced Research Program (Advanced Technology Program) under Grant No. 9999-03-015 to Texas A&M University, College Station, TX. Thanks are also due to Poly metallurgical Corp., 262 Broad Street, MA 02760 and Clad Metals, Inc., R. D. #2, Canonsburg, PA 15317 for providing the clad metal test specimens. REFERENCES 1. Folds, D. L. and Loggins, C. D. "Transmission and Reflection of Ultrasonic Waves in Layered Media," J. Acoust. Soc. Am., 62, pp. 1102-1109 (1977). 2. Scott, W. R. and Gordon, P. F., Ultrasonic Spectrum Analysis for Nondestructive Testing of Layered Composite Materials," J. Acoust. Soc. Am., 62, pp. 108-116 (1977). 3. Hanneman, S. E. and Kinra V. K., "A New Technique for Ultrasonic Nondestructive Evaluation of Adhesive Joints: Part I. Theory," J. Exp. Mech., 32, pp. 323-331 (1992). 4. Hanneman, S. E. and Kinra V. K. and Zhu, C., "A New Technique for Ultrasonic Nondestructive Evaluation of Adhesive Joints: Part II. Experiment," J. Exp. Mech., 32, pp. 332-339 (1992). 5. Kinra, V. K., Jaminet, P. T., Zhu, c., Iyer, V. R., "Simultaneous Measurement of the Acoustical Properties of a Thin-Layered Medium: The Inverse Problem," J. Acoust. Soc. Am., 95, pp. 3059-3074 (1994). 6. Thomson, W. T., "Transmission of Elastic Waves Through a Stratified Solid Medium," J. of Applied Phys., 21, pp. 89-93 (1950). 7. Haskell, N. A, "The Dispersion of Surface Waves on Multilayered Media," Bulletin of the Seismological Soc. Am., 43, pp. 17-34 (1953). 8. Iyer, V. R. and Kinra, V. K., "Characterization of a Sub-Half-Wavelength Adhesive Layer Using Ultrasonic Spectroscopy," Enhancing Analysis Techniques for Composites Materials, Eds.: L. Schwer, J. N. Reddy and A K. Mal (ASME NDE-Vol. 10,1991) pp. 35-42. 9. Neider, J. A and Mead, R., "A Simplex for Function Minimization," Computer Journal, 7, pp. 308-313 (1965). 10. Kinra, V. K. and Dayal, V., "A New Technique for Ultrasonic NDE of Thin Specimens," SEM J. of Experimental Mechanics, 28, pp. 288-297 (1988). 1440