Bioinspired Low-Frequency Material Characterisation

Strathprints is designed to allow users to access the research output of the University of Strathclyde. Unless otherwise explicitly stated on the manuscript, Copyright © and Moral Rights for the papers on this site are retained by the individual authors and/or other copyright owners. Please check the manuscript for details of any other licences that may have been applied. You may not engage in further distribution of the material for any profitmaking activities or any commercial gain. You may freely distribute both the url (https://strathprints.strath.ac.uk/) and the content of this paper for research or private study, educational, or not-for-profit purposes without prior permission or charge.

When performing acoustic measurements, choice of ultrasonic frequency is subject to a trade-off between axial resolution and penetration; higher frequencies provide greater resolution but suffer more attenuation.In nature, bats and dolphins achieve greater resolution than would be achievable using engineered signals, yet live in highly attenuating media [22].Consequently, they have been studied in relation to developing ultrasonic engineering applications [22].Both animal groups echolocate using broadband signals, but deliver their energy in different ways because of their operating environments.For instance, dolphins live in water where they are acoustically well matched to their environment, whereas bats (mostly) live in air where the acoustic impedance mismatch and attenuation parameters are higher.In general, dolphins use short-duration, highintensity broadband signals, described as "clicks," which contain between 4 and 8 cycles over 40-70 µs, with a centre frequency of 100 kHz or more [23].They can resolve small changes in wall thickness of cylinders that are far less than one wavelength thick [23,24].Also they can discriminate between steel, aluminium, and coral rock aggregates encased in degassed epoxy resin [25], detect 2.5 and 7.62 cm diameter water-filled spheres at distances of 70-100 m [23,26], and detect buried objects [27].In contrast, most bat species emit longer-duration, high-intensity, broadband, frequencymodulated (FM) or combined FM, and constant frequency signals, in pulses of 0.3 to 300 ms, with frequencies from around 10 to 200 kHz [28].Although bats and dolphins employ different signal strategies, it is generally considered that they perceive the distance to objects from echo delay, and the shape of objects from the interference spectrum generated when multiple reflections from different parts of an object overlap [23].
A key development in acoustic signal processing has been the use of broadband signals such as FM chirp signals with pulse compression techniques to derive time of flight and amplitude information for imaging, as has been well demonstrated in air coupled NDTE [10,11], high-frequency material characterisation [20], and marine sonar [21].Properties of the FM signal, particularly linear FM (LFM), include good signal to noise ratio (SNR), autocorrelation properties, and delivery of high ultrasonic energy levels [10,11,21]; they have more robust ultrasound features than other coded waveforms [20].The use of window functions (smoothing edges) has improved measurements using LFM chirps by reducing ripple distortion and rise and fall overshoots [10,11].Gan et al. [10] used LFM chirp signals modulated by a Hanning window for air-coupled ultrasonic imaging, providing improved SNR and autocorrelation characteristics over narrowband pulses [10].Pallav et al. [11] used an elliptical Tukey window for chirp modulation in air coupled systems, giving improved performance over other windows.Misaridis and Jensen [20] concluded that LFM signals provide up to 10 dB gain in SNR over binary complementary Golay-coded waveforms for imaging in highly attenuating media; benefits are afforded by use of an amplitude tapering function such as a Tukey window.Chirp subbottom profilers are used in subsea geophysical imaging to generate repeatable 1 to 24 kHz source signatures that are better controlled than boomer and pinger signals, resulting in higher resolution and penetration [21].A broad bandwidth chirp with a squared taper function is commonly optimal for pulse compression techniques in subsea imaging [21].Despite these advances, however, obtaining high-resolution measurements of material properties and thickness has remained a significant problem when using acoustic measurements in highly attenuating environments.

Development of New Biologically Inspired Methodologies
To address this problem, we present a method for characterising material properties, including thickness, sound velocity, and attenuation of materials using coded signals with better penetration/low frequencies, but gain high resolution by optimising the energy delivery transducer and signal design system.This is possible through the design and application of novel, biologically inspired signals.Details of signal design and modelling are outlined below, prior to a description of tests that demonstrate their successful application.this is because the ZPH signal contains an integer number of cycles at each frequency, enabling ready analysis of this signal using the DFT.Effectively, we are adapting the signal to the tool to improve amplitude, frequency, and phase estimation.

Calculating Theoretical Transmission and Reflection Spectra. Use of the coded signals in determining material
properties, including longitudinal sound velocity, the attenuation constant, and thickness measurements, is explored through modelling.In this, and the calculation of theoretical transmission and reflection spectra, it is assumed that a plane acoustic pressure wave is incident in the z direction, normal to the layer interfaces, and that the layers are infinite and homogeneous perpendicular to z.T h ejth layer is characterised by the density ρ j and speed of sound c j .T h e pressure in the jth layer is given by where i = √ −1, ω is the angular frequency, the wavenumber is and α j is the attenuation coefficient in Npm −1 .Following Mikeska and Behrens [17], the attenuation coefficient is assumedtobeatmostaquadraticfunctionoffrequency , f , having the form: The regions 0 and n are assumed to be semi infinite, and the layers 1, ..., n−1 have thicknesses d 1 , ..., d n−1 ,r e s p e c t i v e l y .
The transmission and reflection coefficients through the layered structure are calculated using a transfer matrix approach.The transfer matrix across the jth layer is given by The appropriate boundary conditions at the interfaces between layers are equality of pressure and normal particle velocity [31].Applying these at the boundary between layers j and j + 1 yields, the transfer matrix across the jth interface: The transfer matrix Π for the whole structure is given by the product of the transfer matrices of the layers and their interfaces: This transfer matrix relates the pressure wave in region 0 to that in region n via the expression: Setting the incident condition A 0 = 1, noting that B n = 0, and assuming that z 0 = 0 yields the transmission coefficient: and the reflection coefficient: The transfer matrix approach allows modelling of propagation through an arbitrary numbers of layers.It produces the same results as recursive application of the expressions for single-layer transmission and reflection coefficients [31], but it calculates the overall coefficients directly rather than iteratively.In this paper, the model used to fit the experimental data consists of one layer of test material between semi-infinite regions of water, in which case the single-layer expressions also give the same direct results.To fit the data, it is assumed that the density and thickness of the test material are known, as are the density and speed of sound of water.It is further assumed that the attenuation coefficient in water is negligible.The speed of sound is determined from the spacing of the resonances, ∆ f , in the transmission or reflection spectra of the thickest sample using Subsequently, the attenuation coefficient parameters a 1 and b 1 are found by curve fitting the modelled data to the experimental data for the thickest sample.These parameters are then used to produce modelled data for differing thicknesses of the same material.

Experimental Testing
To evaluate the effectiveness of LFM and ZPH signals in lowfrequency material characterization, tests were performed at the joint University of Leicester/British Geological Survey Ultrasonic Rig Facility located in Keyworth, Nottingham, UK.This comprises a water tank (1.65 m × 1.43 m × 0.90 m) with infrastructure for the precise mounting and positioning of transducers and target materials, housed in a temperature controlled laboratory with water temperature continually logged (Figure 1(a)).Custom built transducers (Alba Ultrasound Ltd., Glasgow, UK) comprise highsensitivity piezoelectric composite main elements having a −6 dB fractional bandwidth approaching 100% around 100 kHz.The transducers operate in the rarely explored frequency range of 40-200 kHz-an order of magnitude lower than commonly used in acoustic spectroscopy, but higher than those generally used in marine sonar.The motivation for using this frequency range is to readily emulate the echolocation systems of bats and dolphins and explore bioinspired applications.Signals are generated and acquired using a highspecification modular system containing a ZT530PXI, 16bit arbitrary waveform generator and a ZT410PXI, 16bit digital storage oscilloscope (ZTEC Instruments Inc.).A C++ programme was developed to automate the signal transmission and reception procedure.An illustration of the setup is shown in Figure 1 a reflection from the target is also collected on the transmitter; a reference signal is collected separately via a pulse/echo from the air-water interface, with the Tx placed on the bottom of the tank facing the surface.The setup differs to that used for measuring insertion loss (IL) and echo reduction (ER) in sonar applications [13], where IL is obtained by measuring the signal before and after placing a target in between the projector and hydrophone, and where ER is obtained by measuring the incident and reflected sound by a hydrophone placed between the projector and target.As measurements of IL and ER are made separately, alignment issues can affect the measurements [13].In our setup, Tx may be used for transmitting and receiving reflections, thus transmission measurements can be made simultaneously; the transducer positions remain constant whilst the sample may be moved in and out of the ultrasonic propagation path, minimising alignment problems.The transducers are further from the target than the hydrophone, thus signal overlap is reduced.Extraneous reflections from tank components are not analysed as they arrive far later than the pulses of interest.
During testing data are acquired with a 10 MHz sampling frequency, for 30000 points, using 32 averages.The transmitted, received, and reflected signal data are saved in binary format data and exported into Matlab for analysis.Spectral analysis is performed on signals using the DFT and evaluated using in-house Matlab code.The length (N) and sampling frequency (F s )o fs i g n a l sa r es e l e c t e d to ensure that the spectral resolution represented by the ratio (F s /N ) should fit within predetermined frequency bins and avoid spectral leakage.The linear magnitude spectra are converted to decibels (dB) by referencing the energy magnitude spectra through a target to the energy magnitude spectra through water.The reference spectra for transmitted signals are acquired for transmission through water, whereas for reflected signals, they are reflected off the air water interface, where the two-way travel time is kept the same.This processing step removes the effect of propagation through water and the frequency response of the transducer.Both ZPH and LFM signals, as described above, are used in the tests.Design recommendations for LFM signals from previous studies have included sweeping a band slightly larger than the passband of the transducer, with a hightime bandwidth product and using a window function [10,11,20,21].Our 40-200 kHz bandwidth LFM signal, which sweeps a band slightly larger than the −6dB range of our transducers, is approximately 70-170 kHz.The length of the signalsisselectedtohaveaslargeatimebandwidthproduct as possible, yet short enough to ensure no extraneous tank reverberation interference.The signals are also designed so that acquired signals can be analysed using the DFT with frequency binning.The signals have similar features to biological signals, but with well-constrained properties so that they can be analysed accurately using available signal processing tools.A 10%-tapered Tukey window is applied to both LFM and ZPH signals.The window improved the signal characteristics and reduces the transducer "turn on" and "turn off " impulse responses.The signals (created with a 10 MHz sampling frequency) are 500 µsl e n g t h( t oa v o i d spurious reflections), and ZPH signals are created with 2 kHz steps.Data are collected using a 500 µs window, starting at the onset of the received or reflected signal, which allows 2 kHz-frequency resolution to be obtained.Transmission and reflection magnitude spectra are calculated as described above.Frequency domain spectra are calculated for the 40-200 kHz range, with 2 kHz resolution, by taking a 5000-point window of the acquired data.
In order to optimally use bioinspired signals with a bandwidth similar to that employed by bats and dolphins, we have selected targets that will allow discrimination of density at the appropriate frequencies.These include relatively homogeneous polypropylene (PP), polyvinylchloride (PVC), and brass panels of variable thickness (0.5-40 mm thick, comparable to those used in previous panel tests-see above), but thin enough to make it impossible to separate front and back face reflections using echo delay techniques at the adopted wavelengths.These are determined instead using pulse overlap techniques and interference spectra from which material property information including longitudinal sound velocity, the attenuation constant, and prediction of thickness (via knowledge of the sound velocity) may be derived.As sound diffraction is a problem in lowfrequency measurement [13][14][15]18], the rectangular (300 × 400 mm) panels have lateral dimensions significantly greater than the wavelengths used (7.5-37.5 mm for 40-200 kHz); consequently the whole beam remains within the panel boundaries.

Results
Examples of time domain LFM received signals for the transmission and pulse echo are shown in Figures 3(a The transmission and pulse echo spectra differ in shape for different materials.The first maximum in the transmission spectrum (or minimum in the reflection spectrum) occurs at the fundamental frequency, where the thickness equals half the wavelength; subsequent maxima/minima occur at multiples of the fundamental frequency [7,32].The peak spacing is greater for the brass panels than the plastic samples, indicating a higher sound velocity.The spectral shape is indicative of the mechanical Q factor, which is related to attenuation by α = πf/Qυ, Q mech ,whereQ mech = f 0 /( f 1 − f 2 ), where f 0 is the mechanical resonance frequency, f 1 and f 2 are the frequencies where energy is half that at resonance, and υ is the velocity.The peaks are sharper for brass than for the plastics, indicating the higher Q factor of  the material.An underlying slope to the plot can be observed, in addition to the peaks, caused by attenuation-which is a function of frequency [17].
Examples of the received ZPH and LFM signals following transmission through 40 mm PVC and pulse echo are shown in Figures 6(a) and 6(b), respectively.The spectral responses are almost identical, except in the bandwidth limits; this is because of decreased transducer sensitivity.
Test and modelled frequency spectra are shown in Figures 7, 8 and 9, for variable thickness PP, PVC, and brass panels, respectively.As noted, the peak spacing is dependent on the speed of sound in the material and its thickness; when material thickness increases the peak spacing decreases.The resonant features can be observed in panels of 10 mm thickness or greater; for thinner panels, the resonant spacing is larger than the bandwidth.The modelled and test results agree with PP and PVC panels.The test brass spectra agree with the modelled spectra for all thicknesses except 10 mm in transmission and 1 mm in reflection.The model parameters, determined by fitting the spectral response in the case of each 40 mm thick panel are used to predict the spectra for the other targets with knowledge of their          Although modelled and test spectra generally fit well, anomalies are observed in some spectra, that is, notches related to beating frequencies are observed in the spectra at 110 kHz for 5 and 10 mm PP, 130 kHz for 5 and 10 mm PVC, and 110 kHz for 20 mm brass.The thicknesses of these panels are approximately equal to λ/2o rλ/4 for the frequency at the notch.The notches are only observed for the panels, where the fundamental resonance frequency is within the bandwidth.It is not clear why these notches are observed; one explanation is that the resonating target induces resonances in the mounting structure.However, clear differences can be observed in the spectral signatures obtained for panels of different thickness, thus indicating the potential of the technique for thickness determination.
The effect of varying thickness in the model is shown in Figure 10.The modelled and test data are shown for thicknesses in the range 1 to 40 mm, with predictions for ±0.5 mm.It is reasonable to suggest that thickness may be resolved to better than ±0.5 mm using this technique.

Discussion
The fact that different materials of different thicknesses have diagnostic characteristics in relation to modulation in the time domain as well as associated spectral notches in the frequency domain demonstrates the ability of our method to discriminate both between materials and material thickness.The thickness resolution obtained using frequency domain spectra, is of the order of ±0.5 mm.Considering the wavelengths of our signals: 7.5 mm at 200 kHz, 15 mm at 100 kHz, and 37.5 mm at 40 kHz, this gives resolutions of λ/15, λ/30, and λ/75, respectively.It has been shown that dolphins can achieve resolutions of λ/50 using similar frequency ranges [24], and, therefore, the resolution demonstrated in the present study is similar to that accredited in nature to dolphins.
Previous studies on underwater sonar have evaluated ER and IL for different materials in a similar frequency range [16][17][18][19].Mikeska and Behrens [17] used frequencies of 50-500 kHz using steps of 5 kHz.They determined that sound velocity estimations were accurate to 3% by varying sound speed in the model and comparing modelled results to test data.Barnard et al. [16] used discrete frequencies in the 100-500 kHz range, with approximately 100 kHz resolution, to compare modelled and test data for transmission coefficients with varying grazing angles.The fits of modelled to test data achieved were poorer than those demonstrated in the current study; however, Barnard et al. used multiple transducers to cover the frequency range, and measurements were made at discrete frequencies.The broadband nature and sensitivity of the devices in our work are thought to be responsible for the improved results.Thibieroz and Giangreco [18] used a 0-100 kHz broadband pulse to interrogate a 30 mm thick steel panel.Perturbing phenomena affected the fit between modelled and test data, and this was due to diffraction effects and extraneous reflections.A signal processing technique was proposed to overcome these phenomena though some discrepancies were still observed-which were

Figure 1 :Figure 2 :
Figure 1: Experimental setup.(a) laboratory picture for the experimental set up.(b) diagram of experimental setup showing transmitting transducer (Tx), receiving transducer (Rx).Reflected and outgoing signals collected on channel 1 (Ch 1) of the oscilloscope, received signals collected on channel 2 (Ch 2).Computer (PC) used for controlling signal delivery and acquisition.
) and 3(b).Examples of the normalised spectral response obtained by transmitting LFM and ZPH signals through water are shown in Figure 3(c).It can be seen that the received LFM and ZPH signals produce spectral responses that are almost identical, except at the extremes of the frequency range where slight differences are caused by low transducer sensitivity.Examples of the time domain pulse-echo received signals for different thicknesses of PVC and brass panels are shown in Figures 4(a) and 4(b).The spectral responses obtained for 40 mm thick panels in transmission and pulse echo (when LFM signals are applied) are shown in Figures 5(a) and 5(b), respectively.The spectral responses show periodic features the dimensions of which vary depending on material type.

Figure 3 :
Figure 3: Time domain signals acquired for 40 mm thick polypropylene (PP), polyvinylchloride (PVC), and brass targets; (a) transmitted signals with water signal shown for reference, (b) reflected signals, and (c) Frequency spectra for the different signals received through water: normalised spectra for the linear frequency-modulated (LFM) chirp and zero phase (ZPH) ladder chirp.

Figure 4 :
Figure 4: Time domain-reflected signals acquired for 40 mm thick targets.

Figure 7 :
Figure 7: Experimental (black diamonds) and modelled (grey lines) spectra for polypropylene (PP) targets of various thicknesses, obtained using the linear frequency modulated (LFM) chirp; traces are offset by the values shown in brackets.

Figure 8 :
Figure 8: Experimental (black diamonds) and modelled (grey lines) spectra for polyvinylchloride (PVC) targets of various thicknesses, obtained using the linear frequency modulated (LFM) chirp; traces are offset by the values shown in brackets.

Figure 9 :
Figure 9: Experimental (black diamonds) and modelled (grey lines) spectra for brass targets of various thicknesses, obtained using the linear frequency modulated (LFM) chirp; traces are offset by the values shown in brackets.