Jtam.dvi JOURNAL OF THEORETICAL AND APPLIED MECHANICS 46, 4, pp. 917-931, Warsaw 2008 ELASTIC WAVE PHASED ARRAY FOR DAMAGE LOCALISATION Wiesław Ostachowicz The Szewalski Institute of Fluid-Flow Machinery Polish Academy of Sciences, Gdansk, Poland Gdynia Maritime University, Faculty of Navigation, Poland e-mail: wieslaw@imp.gda.pl Tomasz Wandowski Paweł Malinowski The Szewalski Institute of Fluid-Flow Machinery Polish Academy of Sciences, Gdansk, Poland e-mail: tomaszw@imp.gda.pl In this paper, a problem of piezoelectric transducers used for damage localisation in an aluminium plate is investigated.Waves propagate in a specimen and reflect from boundaries and from discontinuities. Because damage reflected signals very often have a lowamplitude, a phased array concept isutilized toamplify the signal.Theamplification isbasedonthe wave interference phenomenon. The star configuration consists of four linearphasedarrays.Asa result, adamagemap is createdcombining four outputs of linear phased arrays. Several configurations with a varying number of transducers are investigated. The resulting damagemaps are compared so as to balance the number of transducers and the possibility of flaw localisation using these maps. The presented investigation has only a numerical character. Key words: phased array, Lambwaves, damage localisation 1. Introduction Wave-based structural healthmonitoring is a non-destructive approach to en- sure safety of structures. Especially skin panels of airplane fuselages are here an object of interest. Permanently attached piezoelectric transducers PZT (le- ad zirconate titanate) serveas a source of elasticwaves andalso as their sensor. Elastic waves in the form of Lamb waves in thin walled structures are very 918 W. Ostachowicz et al. sensitive to various discontinuities such as cracks, regions of corrosion, dela- minations and matrix cracking (the last two for composite materials). Such discontinuities must be detected and localised in an early stage of develop- ment before they can hazard lives of airplane crew and passengers. Therefore, this subject is under investigation by many researchers (Deutsch et al., 1997; Fomme et al., 2004; Greve et al., 2005; Peil and Loppe, 2006; Zhongqing et al., 2006). Various configurations have been studied for damage localisation such as distributed placement of four PZTs for crack identification (Ye et al., 2006) or clock-like manner sensor array (Zak et al., 2006). In this paper a star configuration for damage localisation is investigated. 2. Specimen and elastic waves A damage localisation procedure is conducted on a flat square aluminium panel (1000mm×1000mm×10mm). Lamb wave propagation in this simple structure has been modelled by Spectral Element Method (Zak et al., 2006) (40× 40 elements, 36 nodes per element). The excitation and registration of waves in the specimen is realized in nodes of the Spectral Elements. The applied excitation signal (Fig.1) has been chosen as a wave packet (5-cycle 100kHz sinemodulatedwith aHanningwindow). It allows one to concentrate the excitation signal energy in a narrow frequency band. Fig. 1. Excitation Signal – 5-cycle 100kHz sine Hanning windowmodulated Generally, when piezoelectric transducers are used to generate waves in plates, both symmetric (S0,S1,S2, . . .) and anti-symmetric (A0,A1,A2, . . .) modes exist and the number of these modes depends on the product of the excitation frequency and the plate thickness. There are only two fundamental Elastic wave phased array for damage localisation 919 modes, A0 and S0 (Fig.2), that propagate up to almost 2MHz·mm in alumi- nium. Therefore, the chosen frequency of excitation ensures that none of the higher modes will propagate in the specimen, which surely would make the signal analysismore difficult. The numerical model used allows analysing only propagation of the fundamental anti-symmetric Lamb wave mode (A0). Fig. 2. Fundamental Lambwavemodes: (a) symmetric S0, (b) anti-symmetric A0 As it will be shown in Section 6, the wave group velocity c is essential for the signal processing algorithm presented in this paper. Few numerical expe- riments have been conducted to determine the value of c. The average value of the Lamb wave mode A0 group velocity resulting from these experiments is c=2777m/s. 3. Linear phased array theory Linear phased arrays (Shi-Chang and Yijun, 1999; Purekar and Pines, 2001; Giurgiutiu and Bao, 2004; Sundararaman et al., 2005; Pena et al., 2006) are made of a certain number of piezoelectric transducers placed along a line at the same distance between each other. It is assumed that waves generated by the transducers have an omni-directional character and create a pattern, which is a result of superposition of the waves generated by each transducer individually.A sweep beam,which can be steered, is created if each transducer from the set excites waves with an individually adjusted time delay (Fig.3). Changing the steering angle from 0 to 360◦ and calculating the time delay for 920 W. Ostachowicz et al. each transducer, an area of the structure around the array of transducers can be interrogated. This concept is known from radars, where the dish sweeps the space with a focused ray for target searching.When the waves from the radar find an object, they (the waves) are reflected from it and they return with some time delay. If the velocity of the wave generated by the radar and the angle of turn of the radar dish are known, then the position of the object can be determined by simple calculations. This technique is also used in damage detection but instead of a rotating radar dish, phased arrays are used for the beam-forming procedure. The principle of work of a phased array is explained below. Fig. 3. Simulation of wave front creating for 16 transducers with spacing equal to the half wave length Let us consider a set of n transducers that can generate and receive elastic waves with an omni-directional character. Furthermore, it is assumed that the rays from the transducers are parallel. This assumption is justified only if the transducers are placed close to each other and the discontinuity causing wave reflection is far enough from the array. The greater the distance between the transducers the farther the discontinuity should be tomake the idea of beam- forminguseful.Thebeam-formingprocedure isbasedon the signal interference principle.Theprocedureof producingawave front canbedivided in two cases: for a signal transmitted and for a signal received (Giurgiutiu and Bao, 2004). In order to focus the total signal in the direction indicated by the angle θ (Fig.4), the ith array element should excite a wave with an individual time delay ti relative to the 1st PZT (Fig.4), given by the formula ti = licosθ c (3.1) where li is the distance between the ith and the 1st transducer and c is the wave group velocity. The same procedure must be repeated for the signals Elastic wave phased array for damage localisation 921 arriving to the sensors, which are backscattered from any obstacles situated at an angle θ because the ninth transducer is closer than the first one to discontinuities at the angle θ. Fig. 4.Wave front creation principle The focusingmay be realized in twoways: electronically (an electronic sys- tem can excite waves with individual time delays) or by a computer algorithm (all transducer excite waves individually and signal processing procedures ba- sedontime shiftingareapplied to thegathered signals).Theangle θ is changed from 0 to 360◦ and as a result the wave is amplified at angles corresponding to obstacles and faded for the rest. 4. Phased array algorithm In this paper, results of numerical simulations are only discussed and the focusing of the beams is performed by a computer program as explained in the previous section. In this numerical simulation, waves are generated and registered by each transducer of the phased array. The process of wave generation and sensing is conducted in n steps. At each step, one transducer generates waves which are registered by all trans- ducers. For the nth element phased array this makes a total of n2 signals. Then a signal processing procedure is applied, which relies on time shifting for each collected signal with a time delay related to the spacing of transducers and the angle at which the beam is created. Signals registered in all sensors, when the excitation comes from the first transducer, are shifted in time to this transducer and then summed. In the next step, the same procedure is 922 W. Ostachowicz et al. repeated for the second transducer. This procedure is repeated for each sensor of the array. As a result, n signals are obtained. This is how the beam-forming procedure for generating waves is carried out. Similar operations are performed for the beam-forming procedure for the arriving signals. This can be achieved by shifting all the n signals to the chosen transducer and summing them. This transducer is the point of origin for a virtual sweep beam. Due to the discrete character of all signals, the calculated time delay is divided by a time step resulting from the simulation and rounded (an integer number is needed) to obtain a number of points for the discrete time shift. The time delays are calculated for beam angles from 0 to 360◦ with a chosen step. The signals are then transformed from the time domain into a physical domain of the structure using formulas x=Rcosθ y=Rsinθ (4.1) where R = ct/2, t is the time and (x,y) are coordinates of the investigated surface area. It gives in effect a damage map in which the z coordinate represents the amplitude of the signal. The usedMATLAB R© plot procedures interpolate the signals for the rest of angles. As a result, a three dimensional damage map covering the area of the investigated structure is created (Giurgiutiu andBao, 2004; Yu andGiurgiutiu, 2005). Values on the z axis represent the amplitude of the interrogating signal at the point (x,y) of the structure. In order to facilitate the map analysis, the squared signals are plotted. In order to extract the damage signature, a special signal processing algo- rithm is proposed. It is assumed that the excitation signals and the damage reflected signals have similar features. Therefore, the features of the total signal reflected from thedamage canbecompared to the excitation signal. For this purpose, a signal processing procedure, which uses a virtual rectangular time windowwith the same lengthas the excitation signal is used.The timewindow isused to extract a portion of the total received signal and then, features from this portion of the signal are compared to the excitation signal (Fig.5). A similarity measure of the received signal portion and the excitation signal for a given point was built as a product of the absolute values of the excitation and the windowed signal portion and then summed for all points of the window. Based on this measure, a damagemap can be plotted by applying thementioned procedure for all points of the structural area under investigation. Damaged areas of the structure will have a high value of the similarity measure. Elastic wave phased array for damage localisation 923 Fig. 5. Illustration of building a similarity measure for signal processing 5. Multi-phased array transducer configuration In this paper, a combination of linear phased arrays is used (multi-phased array) forming a star configuration. The form of this configuration is an effect of authors, own research for the improvement of the ordinary linear phased array based damage localisation. The star configuration for 9 transducers in each linear array is presented in Fig.6. Fig. 6. Star transducers configuration Thedistances between sensors/actuators are not equal, because this is how the nodes in the mesh of spectral elements are placed. The transducers have not beenmodelled; only the nodes are used as sources and receivers of waves. Four cases are considered and compared, for 3, 5, 7 and 9 transducers in a linear array. They correspond to 9, 17, 25 and 33 transducers in the whole configuration. 924 W. Ostachowicz et al. The beam-forming algorithm is independently applied to each array. As a consequence, fourmaps of the scanned area are obtained (Fig.7). Thesemaps are made of signals which represent the difference between the damaged and intact plate signals. Fig. 7. Four componentmaps – the stripe in the middle represents an array for which the map is generated, black box indicates the simulated damage (90% stiffness reduction in one spectral element) As can be noticed, the line on which the transducers from one array lie is a symmetry line of themap (Fig.7). Therefore, a singlemapmay indicate two damage areas (ghost damage) even if there is only one damage (Fig.7a). This symmetry is even better seen on polar plots (Fig.8) illustrating the amplitude (normalised to its maximum value) dependence on the angle. Figures 8a-d correspond to Figures 7a-d. Inorder to remove theambiguity of localization, analgorithmwasproposed joining the four component maps into one (Fig.9a). This final map is simply a product of the four component maps. It sharpens the image of damage more than the sum of the maps and allows a more precise localization of a defect than in the case of only one component map. A great advantage of this approach is also seen in Fig.9b. The direction of the damage is indicated Elastic wave phased array for damage localisation 925 Fig. 8. Fig. 8 Amplitude angle dependence for four arrays (damage simulated as 90% stiffness reduction in one spectral element), (a) horizontally placed array, (b) vertically placed array (c) an array rotated by 45◦, (d) an array rotated by −45◦ Fig. 9. Four maps joined for damage as a 90% stiffness reduction, (a) resulting map, (b) amplitude angle dependence 926 W. Ostachowicz et al. very precisely there. Comparing Fig.9b with Fig.8, one may observe that the proposed algorithm gives better results than any component map. Theabove results (Fig.7,Fig.9a) showthe localization ofdamagemodelled as a 90% stiffness reduction in one spectral element. The scale of colours is linear from white (indicating the minimum level of the signal amplitude) to black (the maximum level of the signal amplitude). Signals used for the damage map algorithm were corrupted with random noise (up to 10% of the signal amplitude) to make the investigation more related to real measurements. 6. Effect of transducers number Investigation of the effect of the number of transducers on damage localiza- tion was begunwith a simple damage scenario, namely one 25mm long crack modelled as separation of the nodes between appropriate Spectral Elements. It is placed approximately at the angle 45◦ with respect to the plate centre, which is also the centre of PZT configuration used to localise the damage. Fig. 10. Damagemap for 25mm long crack localisation for the case of: (a) 9, (b) 17, (c) 25, (d) 33 transducers in the star configuration Elastic wave phased array for damage localisation 927 Figure 10 depicts the damage maps for this case. They correspond to 9, 17, 25,33 PZT transducers in the star configuration, respectively. One may notice that an increase of the number of transducers results in a decrease of the damage signature (the area where the signal is amplified by the al- gorithm). To study this result more precisely, polar plots are presented in Fig.11, which correspond to the damagemaps in Fig.10. Polar plots show the amplitude dependence on the angle. The values of amplitude on these plots have been normalised in relation to the amplitude of the damagemap for the 33-PZT configuration (Fig.10d). Analysing Fig.11, one may notice that ad- ding the transducers to the star configuration causes that the direction of the crack becomes more apparent and precise. Although on the map (Fig.10d) small additional increments of the amplitude appear (around points (x=0.4, y = 0.8) and (x = 0.8, y = 0.3)), which may be also noticed on the corre- sponding polar plot (Fig.11d), its amplitude is negligible. After this one-crack case, the plate with two cracks (25mm both) has been investigated. Fig. 11. Amplitude angular dependence for the case of: (a) 9, (b) 17, (c) 25, (d) 33 transducers in the star configuration for 25mm long crack localisation The same procedure has been repeated. Firstly, the damagemaps are plot- ted (Fig.12) and, as before for the 33-PZTcase (Fig.12d), themap is themost 928 W. Ostachowicz et al. Fig. 12. Damagemap for two 25mm long cracks localisation for the case of: (a) 9, (b) 17, (c) 25, (d) 33 transducers in the star configuration precise. Especially, the second damage (on the right) is very well indicated, what cannot be said about the maps plotted for configurations with a fewer number of transducers. Secondly, the amplitude angle dependence was inve- stigated. Figure 13 shows that only for the last configuration the direction is verywell indicated.What is also important, the value of the amplitude for the second (on the right) crack is significant only for this 33-transducer configura- tion. The lower value of the amplitude for the second crack is a result of the crack placement in relation to the configuration. The crack on the left reflects waves better in the configurationdirection than that on the right. If oneplaced a crack on the horizontal line for y=0.5 or close to it, the reflections would not be registered by the PZTs. It is because the surface of the crack that can reflect waves back to the PZTs is very small. Surely, this should be considered as one of the disadvantages of such a concentrated configuration for damage localisation. On the other hand, onemay use few concentrated configurations placed around themonitored structure to avoid areaswhich cannot be scanned by one configuration due to damage orientation. Another solution is the use of a distributed configuration. Then, single transducers are placed around the whole structure.This last approachneeds also a localisation algorithmbut this Elastic wave phased array for damage localisation 929 is not the subject of this paper. Last information carried by polar plots Fig.11 andFig.13 which have to be taken into consideration is themaximum level of the amplitude for each case. For one crack scenario, the amplitude decreases with the increasing number of transducers in the configuration. However, the area with this level of amplitude is wide and the amplitude drops near the direction of 45◦ (Fig.11). It does not happen when 33 PZTs are used. Consi- dering the two-crack case, the amplitude level is the highestwhen 33PZTsare used and, as it was mentioned, only then the second damage is significantly indicated. Fig. 13. Amplitude angular dependence for the case of: a) 9, (b) 17, (c) 25, (d) 33 transducers in the star configuration for two 25mm long cracks localisation 7. Conclusions Thephased array technique for damage localisationwas successfully used.The influence of different number of transducers on damage localisation was inve- stigated.Damagemaps andpolar plotswere used to determine the influence of the number of transducers on damage localisation ability of the star configu- 930 W. Ostachowicz et al. ration. Star configurations with 9, 17, 25 and 33 transducers were compared. The best results were obtained for the 33-PZT case. Decreasing the number of transducers, results in an imprecise crack localisation and, for two cracks, in a weak response on the damage map for the second crack. It was not attempted to continue with increasing the tnumber of ransdu- cers, because the increase from 9 to 33 is already a large one considering the cost of transducers and necessary equipment. Experimental verification of the proposed configuration and algorithm is planned.The size of thewhole configurationwill be then increasedbecause the size of transducers will have to be taken into consideration, and it will not be possible to place them so close as the nodes in the used numerical model. The best choice seems to be to place them equally spaced. In Giurgiutiu and Bao (2004) the authors suggest that the distance should be chosen as an integer multiple of thehalf of thewavelength. Inanexperimentalwork, also the second propagating fundamental mode S0 and the mode conversion at defects and boundaries will have to be taken into consideration. References 1. DeutschW.A.K., ChengA., Achenbach J.D., 1997, Self-focusing of Ray- leigh waves and Lamb waves with a linear phased array, Research in Nonde- structive Evaluation, 9, 81-95 2. Fomme P., Wilcox P., Lowe M., Cawley P., 2004, Development of a permanently attached guided ultrasonic waves array for structural integrity monitoring,Proc. 16th World Conference on NDT, CD-ROM 3. Giurgiutiu V., Bao J., 2004, Embedded-ultrasonic structural radar for in- situ structural healthmonitoring of thin-wall structures, Structural HealthMo- nitoring, An International Journal, 3, 121-140 4. Greve D.W., Neumann J.J., Nieuwenhuis J.H., Oppenheim I.J. and Tyson N.L., 2005, Use of Lamb waves to monitor plates: experiments and simulations, Proc. Sensors and Smart Structures Technologies for Civil, Me- chanical, and Aerospace Systems, 5765, 281-292 5. Shi-Chang W., Yijun S., 1999, Optimum beam steering of linear phased arrays,Wave Motion, 29, 245-265 6. Sundararaman A., Adams D.E., Rigas E.J., 2005, Biologically inspired structural diagnostics through beamforming with phased transducers arrays, International Journal of Engineering Science, 43, 756-778 7. Peil U., Loppe S., 2006, Detection of structural changes by means of piezo disc,Proc. Third EuropeanWorkshop on StructuralHealthMonitoring, 133-140 Elastic wave phased array for damage localisation 931 8. Pena J., Melguizo C.P., Martinez-Ona R., Ullate Y.G., de Espi- nosa Freijo F.M., Kawiecki G., 2006, Advanced phased array system for structural damage detection, Proceedings of the Third European Workshop on Structural Health Monitoring, 244-250 9. Purekar A.S., Pines D.J., 2001, Interrogation of beam and plate structures using phased array concepts,Proc. of 12th International Conference on Adap- tive Structures and Technology, 275-288 10. Ye Lu, Lin Ye, Zhongqing Su, 2006, Crack identification in aluminium plates using Lambwave signals of a PZT sensor network, Smart Materials and Structures, 15, 839-849 11. YuL.,GiurgiutiuV., 2005,Advanced signalprocessing for enhanceddamage detectionwithpiezoelectricwafer active sensors,Smart Structures and Systems, 1, 185-215 12. Zak A., Krawczuk M., Ostachowicz W., Kudela P., Palacz M., 2006, Elastic wave propagation in a cracked isotropic plate,Proceedings of the Third European Workshop on Structural Health Monitoring, 316-323 13. Zhongqing Su, Lin Ye, Ye Lu, 2006,Guided Lambwaves for identification of damage in composite structures: A review, Journal of Sound and Vibration, 295, 753-780 Układ fazowy przetworników w lokalizacji uszkodzeń z wykorzystaniem metody propagacji fal sprężystych Sreszczenie W artykule przedstawiono propozycję użycia przetworników piezoelektrycznych oraz metody propagacji fal sprężystych do lokalizacji uszkodzeń w cienkich płytach wykonanych ze stopu aluminium. Generowane fale sprężyste za pośrenictwem prze- tworników piezoelektrycznych, propagującw płycie, odbijają się od jej krawędzi oraz wszelkich nieciągłości. Ze względu namałe wartości amplitud fal odbitych od niecią- głości zaproponowanoużycie układu fazowego przetwornikówdowzmocnienia sygna- łów fal odbitych. Układy fazowewykorzystują zjawisko interferencji fal sprężystych. Zaproponowa- na została konfiguracja przetwornikówwykorzystująca cztery liniowe układy fazowe. Mapa uszkodzeń tworzona jest poprzez połączenie map składowych otrzymanych dla poszczególnych układów liniowych. Dokonano również analizy wpływu ilości przetworników piezoelektrycznych na wyniki lokalizacji uszkodzeń. Przeprowadzone badaniamają charakter wyłącznie nu- meryczny. Manuscript received February 4, 2008; accepted for print February 25, 2008