Fatigue Crack Length Sizing Using a Novel Flexible Eddy Current Sensor Array Ruifang Xie *, Dixiang Chen, Mengchun Pan, Wugang Tian, Xuezhong Wu, Weihong Zhou and Ying Tang Received: 22 October 2015; Accepted: 16 December 2015; Published: 21 December 2015 Academic Editor: Vittorio M. N. Passaro College of Mechatronics Engineering and Automation, National University of Defense Technology, Changsha 410073, China; [email protected]
(D.C.); [email protected]
(M.P.); [email protected]
(W.T.); [email protected]
(X.W.); [email protected]
(W.Z.); [email protected]
(Y.T.) * Correspondence: [email protected]
; Tel.: +86-731-8457-6479 (ext. 123); Fax: +86-731-8457-6479
Abstract: The eddy current probe, which is flexible, array typed, highly sensitive and capable of quantitative inspection is one practical requirement in nondestructive testing and also a research hotspot. A novel flexible planar eddy current sensor array for the inspection of microcrack presentation in critical parts of airplanes is developed in this paper. Both exciting and sensing coils are etched on polyimide films using a flexible printed circuit board technique, thus conforming the sensor to complex geometric structures. In order to serve the needs of condition-based maintenance (CBM), the proposed sensor array is comprised of 64 elements. Its spatial resolution is only 0.8 mm, and it is not only sensitive to shallow microcracks, but also capable of sizing the length of fatigue cracks. The details and advantages of our sensor design are introduced. The working principal and the crack responses are analyzed by finite element simulation, with which a crack length sizing algorithm is proposed. Experiments based on standard specimens are implemented to verify the validity of our simulation and the efficiency of the crack length sizing algorithm. Experimental results show that the sensor array is sensitive to microcracks, and is capable of crack length sizing with an accuracy within ˘0.2 mm. Keywords: flexible eddy current array; crack sizing; quantitative nondestructive evaluation; spatial resolution; high sensitive
1. Introduction The key components of aircraft such as engine blades are under the long-term effects of cyclic loads and thermal loads, suffering the danger of fatigue failure and the resulting risk of disastrous accidents. To ensure safety and prevent accidents, quantitative nondestructive evaluation (QNDE) systems have been proposed and developed against this background. Condition-Based Maintenance (CBM) is one kind of maintenance strategy based on device status evaluation, and the status information is obtained by using sensors and external test equipment. While the crack is small enough, the structure can keep working despite its “illness”. Sometimes the structure must be sifted out, which depends on the size of crack. A reasonable maintenance strategy reduces not only maintenance costs, but also the economic losses due to outage. CBM can ensure the reliability and safety of equipment, reduce the operation and support costs, decrease breakdown maintenance and preventive maintenance tasks, and prevent accidents. QNDE of microdefects is one of the key technologies of CBM, which helps to effectively prevent the generation and spread of various mechanical structural failures and eliminate failure in the embryonic stage. Eddy Current Non-Destructive Testing (ECNDT) [1–4] has been widely used in non-destructive testing of critical aircraft components, because of its convenience, speediness and high sensitivity in Sensors 2015, 15, 32138–32151; doi:10.3390/s151229911
Sensors 2015, 15, 32138–32151
the detection of metal component problems. Cha et al.  designed a micro-coil sensor with high detection sensitivity, which was fabricated using MEMS technology, whose dimension was only 400 µm ˆ 380 µm ˆ 35 µm. However due to its small size it had to rely on a motion platform, and the detection efficiency is relatively low. Chen et al. [6,7] designed an eddy current probe to test the nuclear power plant pipelines, and to reconstruct and classify defects, but its accuracy is only several millimeters. Flexible eddy current sensors and arrays [8,9] are becoming a hot research topic, aiming to solve the inspection problems of key components with complex geometry. A Meandering Winding Magnetometer (MWM) sensor array fabricated with flexible printed circuit board (FPCB) technology, has been developed to detect fatigue, corrosion, thermal barrier coatings, stress and so on [10,11]. The MWM sensor was used for monitoring the initiation and propagation of fatigue cracks, with a minimum detectable defect length of 200 µm, and depth of 50 µm, but no quantitative detection application was mentioned. Marchand, et al.  designed a flexible eddy current array probe comprising 96 elements, which consists of an exciting and sensing micro-coil for examining specimens with complex surfaces. The probe was very sensitive to micro-defects, yet no quantitative detection results were disclosed. Endo, et al.  designed a flexible array eddy current testing (ECT) probe, and a 12 decibel drop method was used to size fatigue crack length, with a sizing accuracy within ˘3 mm. In addition, other techniques have been used to perform quantitative testing of microcracks. Deng et al.  designed an impedance sensor film for detecting the initiation and expansion of fatigue cracks. A thin film layer of several tens of nanometers was plated on the surface of targets using an ion sputter plating method. The initiation and expansion of fatigue cracks can be monitored by measuring the impedance of the film, which varies because of fatigue crack initiation or expansion. The crack length can be measured with this method and the error is less than 0.068 mm. However, this method requires a metal film to be sputtered on the surface of the subject material, and a high level of surface cleanliness, which can hardly be achieved in many cases, such as in-service equipment like engine blades. Lim and Soh  monitored three stages of fatigue crack using an electro-mechanical impedance (EMI) technique which employed the PZT material as the excitation and vibration sensing sensor and they were able to monitor fatigue remotely with high sensitivity and precision. Nevertheless, the position of cracks is unknown and crack visualization is impossible. In general, although crack sizing has been achieved by researchers using different non-destructive testing techniques, the following disadvantages still exist : preknown defect locations, automation difficulties, operational outages, equipment dismantling, high cost and labor consumption, lack of precision or reliance on motion platforms, low adaptability to complex surfaces, etc. Therefore, a technology which can detect cracks in their early stages or even before formation and allows quantitative tracking of crack propagation and adoption of appropriate maintenance strategies based on crack type, size, etc. is urgently needed. According to a report by Fuchs and Stephens , up to 90% of structural failures are caused by fatigue cracks. Extensive experiments  have shown that fatigue cracks are characterized by small width and approximately rectangular shapes in crosssection, and they are not substantially bifurcated like stress corrosion cracks; besides, the gap between sections, which is filled with air, is about 10–50 µm, thus the defect area is not conductive. This means that fatigue cracks can be simulated using electro-discharge machined defects (EDM defects). Moreover, the length and depth of natural cracks approximately satisfy a 2:1 relationship, while the width of cracks is usually not of concern. In addition, fatigue cracks are mostly surface or sub-surface cracks. Based on the above facts, this paper present the design of a flexible planar eddy current sensor array for quantitative measurement of fatigue crack length in metal components. The novelty of this paper is that a quantitative algorithm named NCSF based on a novel flexible eddy current array (FECA) is developed for fatigue crack length measurement.
Sensors 2015, 15, 32138–32151
2. Sensor Design Sensors 2015, 15, page–page
Although single element eddy current sensors and multi-element sensor arrays are both used for defect sizing, sensor arrays are more widely used because of their advantages such as higher 2. Sensor Design detection efficiency and spatial resolution, and larger coverage [8,12]. Transmit/receive (T/R) type Although single element eddy current sensors and multi-element sensor arrays are both used coil configurations have beenarrays proven provide higher and better for defect sizing, sensor aretomore widely useddetection because ofsensitivity their advantages suchsignal-to-noise as higher ratio,detection and thusefficiency have been widely used in eddy current sensors . In addition, from perspective and spatial resolution, and larger coverage [8,12]. Transmit/receivethe (T/R) type of thecoil actual inspection needs keyproven aircraft to components, the eddy currentsensitivity sensors should meet the configurations have of been provide higher detection and better signal-to-noise ratio, and thus have been widely used in eddy current sensors . In addition, from following requirements: 1. 2. 3. 4. 5.
the perspective of the actual inspection needs of key aircraft components, the eddy current sensors
Flexibility, forfollowing the defect detection of complex surfaces; should meetfitthe requirements: High detection efficiency; no need for motion platforms; in-situ and in-service inspection; 1. Flexibility, fit for the defect detection of complex surfaces; High sensitivity, allowing detection 2. High detection efficiency; no need of formicrodefects; motion platforms; in-situ and in-service inspection; High spatial resolution; high-precision 3. High sensitivity, allowing detection ofmeasurement microdefects; of crack lengths; 4. need Highof spatial resolution; lengths; No preknowing thehigh-precision position andmeasurement orientation of crack defects. 5.
No need of preknowing the position and orientation of defects.
To meet the above requirements, a flexible eddy current sensor array which takes into To meet the above requirements, a flexible eddy current sensor array which takes into consideration both sensitivity and resolution and is able to detect microdefects and measure their consideration both sensitivity and resolution and is able to detect microdefects and measure their lengths is designed as shown in Figure 1. lengths is designed as shown in Figure 1.
(c) Figure 1. Novel sensor array (a) Schematic diagram of the exciting coil; (b) Schematic diagram of the
Figure 1. Novel sensor array (a) Schematic diagram of the exciting coil; (b) Schematic diagram of the sensing coil; (c) Actual sensor. sensing coil; (c) Actual sensor. 3
Sensors 2015, 15, 32138–32151
Overall, the array is in a T/R coil structure with etched coils on polyimide film, using flexible printed circuit board technology, comprising a large uniform exciting coil and 64 sensing coil elements, and has a spatial resolution of 0.8 mm and an effective coverage length of more than 50 mm. The sensor array has a four-layer structure, where the exciting coil is configured on the top and bottom layers and the sensing coils on two middle layers. Every coil is composed of two parts in different layers, which are connected in series through vias, respectively. The flexible design of the sensor resolves the problem of complex surface detection, and allows compact contact between sensors and materials. The two-layered exciting coil is designed in a periodic structure, where a racetrack-like loop coaxial disposition with the sensing coil elements forming a single period. The main design features and advantages of the sensor array are as follows: 1.
Redundant dummy elements are designed on both sides of the array to maintain the periodic structure, eliminate the influence of fringe effect, and ensure the consistency of array elements (labeled “1” in Figure 1c); Because of the flexible printed circuit board technology, the array has very high spatial resolution, thus creating favorable conditions for quantitative detection; The multilayer structure design maximizes the mutual inductance between sensing and exciting coils, thus improving the response of sensing elements, and thereby increases the signal-to-noise ratio; Because of the closed form and very small width of the exciting coil, defect sensitivity can be achieved in any direction with almost no blind zones, thus ensuring the detection efficiency and reliability; Lead portions of the exciting and sensing coils are all designed as upper-lower parallel wires, thereby eliminating the generation of non-periodic excited magnetic field and coupled magnetic field interference introduced by sensing coil lead portions (labeled “2, 3”); 3D finite element simulation can be simplified by exploiting the periodicity of the sensor, with which the response of the sensor can be predicted; Copper grid wires are clad on both sides of the array to make the surface smoother, and improve the reliability of contact between sensor and material (labeled “4”).
3. Principle and Simulation To understand the performance of the sensor, a finite element simulation is conducted on the sensor array presented in Section 2. The AC/DC module in the COMSOL multiphysics software is used for the simulation of a single period model due to its spatial periodicity. The principle of ECNDT is as follows: when an alternating current passes through the coil above the conductive material, an eddy current is generated on the conductor. The excited magnetic field and induced magnetic field are superimposed while the induced magnetic field inhibits the excited magnetic field. Sensing coils located above the material measure the superimposed magnetic fields passing through the planes of the coils, and the induced AC voltage can be calculated as shown in Equation (1): Vout “
dB dφ “ NS dt dt
where, Vout denotes induced AC voltage. φ is the magnetic flux which is the product of the number of turns of sensing coil N, area of sensing coil S and magnetic flux density B, and t is time. The AC voltage depends on several factors such as material conductivity, lift-off, magnetic permeability, geometrical discontinuity, frequency, excitation current, as well as spatial structure and size of the excitation and detection coils. Usually a transimpedance which equals the output voltage Vout divided by the input current Iin is used to represent the performance of a T/R eddy current sensor. As the magnetic flux density B relies on the amplitude and angular frequency of the exciting current, permeability, conductivity, geometric discontinuity of material and lift-of (distance between the
Sensors 2015, 15, page–page Sensors 2015, 15, 32138–32151
depict the relation between acquired transimpedance change and the unknown parameters as shown Equation sensorincoil plane (2): and the surface of tested material), an implicit expression can be established to depict the relation between acquired transimpedance change and the unknown parameters as shown V Z transimpedance out f ( , , , X , l ) in Equation (2): (2) I in Vout “ f pµ, σ, ω, X, lq (2) Ztransimpedance “ Iin where, Ztransimpedance stands for the transimpedance of the eddy current sensor, μ and σ are the where, Ztransimpedance stands for the material transimpedance of theω eddy µ and σ are the permeability and conductivity of the respectively, is thecurrent angularsensor, frequency of exciting permeability and conductivity of the material respectively, ω is the angular frequency of exciting current, X denotes the geometric discontinuity of material, and l stands for the lift-off. current, denotes thevoltage geometric of material, thetransimpedance lift-off. SinceXthe output anddiscontinuity input current are bothand ACl stands signals,forthe is a Since the output voltage and input current are both AC signals, the transimpedance is a complex complex value which can be expressed as either a real part and imaginary part or amplitude and value The which can coil be expressed as eitherwork a realatpart imaginary part frequency or amplitude phase. planar sensors normally 100 and kHz–5 MHz. High can and boostphase. the The planar coil sensors normally work at 100 kHz–5 MHz. High frequency can boost the induction induction voltage of sensing coils, improve the signal-to-noise ratio, yet it increases the complexity of sensing coils, improve the signal-to-noise ratio, yet it increases the complexity of the ofvoltage the electronic system. electronic system. model is simulated with a 3 MHz sine wave excitation while the sensor is A single-period A single-period is simulated with a 3 MHz sine with waveaexcitation while located located 0.05 mm abovemodel the surface of a defect-free material conductivity of the 18.5sensor MS/mis(31.9% 0.05 mm above the surface of a defect-free material with a conductivity of 18.5 MS/m (31.9% IACS). IACS). The instantaneous distribution of the magnetic field and current density of the sensor at the The instantaneous distribution of the magnetic field and current density of the sensor at the instant instant ωt = 0 is shown in Figure 2a. As can be seen, a racetrack-shaped eddy current loop similar to ωt exciting = 0 is shown in Figure 2a. beneath As can bethe seen, a racetrack-shaped eddySince current similar to is the the coil is produced surface of the material. theloop exciting coil exciting coil produced beneath the surface of the material. Since in thethe exciting coilportion, is superposed superposed byistwo 3/4 tracks at different layers, with overlapping straight eddy by two 3/4 tracks at different layers, with overlapping in the straight portion, eddy current intensity current intensity below the straight tracks is the strongest, while it is weaker below the annular side belowespecially the straight is the strongest, while The it is weaker belowresponse the annular side tracks,along especially tracks, ontracks the side with larger lift-off. eddy current is symmetric the on the side with larger lift-off. The eddy current response is symmetric along the YZ plane, while YZ plane, while it is asymmetric on the XZ plane, leading to an anisotropic response of defects in it is asymmetric on the XZ plane, to an anisotropic defects directions. different directions. When defectsleading are perpendicular to theresponse straight of track, thatinis,different when cracks are When defects are perpendicular to the straight track, that is, when cracks are parallel along the X-axis parallel along the X-axis direction, the impediment to eddy currents is greater, and therefore the direction, thesensitive; impediment to eddywhen currents greater, anddirection thereforeis the arraytoisthe more sensitive; array is more conversely, the is crack length parallel Y-axis, the conversely, when the crack length direction is parallel to the Y-axis, the sensitivity of the sensitivity of the array decreases, therefore, the optimal sensitive direction of the array is the array one decreases, therefore, sensitive of the array is the one parallel to the X-axis parallel to the X-axis the and optimal perpendicular to direction the straight track direction. Quantification of crack and perpendicular to the straight track direction. Quantification of crack length is also based length is also based on the optimal sensitive direction. Since the sensor array is anisotropic, theon the optimal sensitive Since thereby the sensor array is optimal anisotropic, the directions cracks2b are directions of cracks aredirection. determinable, achieving sensitive detection.ofFigure determinable, thereby achieving optimal sensitive detection. 2b shows shows the spatial distribution of magnetic fields generated byFigure the sensor array.the Asspatial can bedistribution seen, the of magnetic fields generated by the sensor array. As can be seen, the flux almost passes through the flux almost passes through the sensing coil, which means a high mutual inductance and sensitivity sensing coil, which means a high mutual inductance and sensitivity are obtained. are obtained.
Figure 2. (a) Distribution of eddy currents; (b) Distribution of magnetic fields. Figure 2. (a) Distribution of eddy currents; (b) Distribution of magnetic fields.
Sensors 2015, 15, 32138–32151
4. Crack Length Sizing Algorithm Sensors 2015, 15, page–page
Before determining the quantitative detection method, the characteristics of the sensor response 4. Crack Length Sizing Algorithm to defects are studied through simulation and experiment. As stated earlier, the goal of this paper determining quantitative the characteristics of the In sensor is to design Before a flexible sensor the array, and to detection quantifymethod, the length of microcracks. fact,response length, width to defects are studied through simulation and experiment. As stated earlier, the goal of paperon is sensor and depth of cracks all impact the sensor response. The influences of width andthis depth to design a flexible sensor array, and to quantify the length of microcracks. In fact, length, width and response should be studied first, and then the correlation between sensor response and crack length depth of cracks all impact the sensor response. The influences of width and depth on sensor is investigated. Two scanning i.e., vertical scanning modesensor (V mode) andand horizontal scanning response should be studiedmodes, first, and then the correlation between response crack length mode (H mode), areTwo employed thevertical simulation to mode study(Vthe correlation of sensor response is investigated. scanningduring modes, i.e., scanning mode) and horizontal scanning mode (H as mode), are in employed the simulation to study the correlation response to the with crack size shown Figure during 3. V mode means the scanning directionofis sensor perpendicular with crack size ason shown in Figuresensitivity 3. V mode means the scanning perpendicular to the crack length direction the optimal direction premise,direction while Hismode means the scanning crack length direction on the optimal sensitivity direction premise, while H mode means the direction is parallel to the crack length direction on the optimal sensitivity direction premise with the scanning direction is parallel to the crack length direction on the optimal sensitivity direction crack right below the sensor array. premise with the crack right below the sensor array. Vertical Scanning mode
Sensor array (a) Horizontal Scanning mode Crack
Sensor array (b) Figure 3. Two scanning modes: (a) V mode; (b) H mode. Figure 3. Two scanning modes: (a) V mode; (b) H mode.
4.1. Effect Analysis of Crack Width
4.1. Effect Analysis of Crack Width
On the basis of the model shown in Figure 2, the lift-off is set to be 0.05 mm, 1 mm-deep cracks
are added andinthe central2,location of cracks obtain the sensor cracks Onwith the varying basis ofwidths the model shown Figure the lift-off is setare to altered be 0.05tomm, 1 mm-deep transimpedance in V scanning mode. Due to its periodic boundary conditions, crack length is setthe to sensor with varying widths are added and the central location of cracks are altered to obtain be infinitely long in this case, and the response of the sensing element above the center of a crack is transimpedance in V scanning mode. Due to its periodic boundary conditions, crack length is set to acquired. The relationship between the imaginary and real parts of the sensor transimpedance be infinitely in this case,isand theinresponse of the sensing Itelement above the center of a crack versus long the crack position shown Figure 4a,b, respectively. shows that the transimpedance is acquired. relationship between and thereal imaginary anda real the the sensor transimpedance outputThe curves of both the imaginary parts present singleparts peak,of where imaginary part presents crest curve the real part exhibits trough curve. It Nevertheless, to the versus the cracka position is and shown in Figure 4a,b, arespectively. shows thatcompared the transimpedance imaginary part is much and the signal-to-noise is where poorer. the Moreover, as output curves of part, boththe thereal imaginary andsmaller real parts present a single ratio peak, imaginary part the essence of an eddy current sensor, mutual inductance is mainly manifested in the imaginary presents a crest curve and the real part exhibits a trough curve. Nevertheless, compared to the part of the transimpedance. Therefore, in the subsequent simulation and experiments, the crack imaginary part, the real part is much smaller and the signal-to-noise ratio is poorer. Moreover, as the response curve of the transimpedance imaginary part is mainly studied. As seen from Figure 4a, essence different of an eddy sensor, mutual mainlycurve. manifested inwidth the imaginary crackcurrent width influences the peakinductance value of theisresponse As crack grows, the part of the transimpedance. Therefore, in the subsequent simulation and experiments, the crackthe response peak value increases. However, the waveforms are similar. Defect response begins to rise when curve of the transimpedance imaginary part is mainly studied. As seen from Figure 4a, different 6 crack width influences the peak value of the response curve. As crack width grows, the peak value increases. However, the waveforms are similar. Defect response begins to rise when the crack is near the bottom edge of the exciting coil, reaches a climax when the crack center coincides with the
Sensors 2015, 15, 32138–32151 Sensors 2015, 15, page–page Sensors 2015, 15, page–page
crack is near the bottom edge of the exciting coil, reaches a climax when the crack center coincides center of the sensing coils, and ends when crack leaves the other edge of exciting coil, suggesting with thenear center of the sensing and ends when crack climax leaves when the other edge of exciting coil, crack bottom edge ofcoils, the exciting coil, reaches the crack coincides that the is sensor the array is only sensitive to the coverage area,awhile insensitive to thecenter peripheral areas. suggesting that the sensor array is only sensitive to the coverage area, while insensitive to the with the center of the sensing coils, and ends when crack leaves the other edge of exciting coil, Thus, the sensor array can be used for measuring the discontinuous change of material properties peripheral areas. Thus, the sensor array can be used for measuring the discontinuous change of suggesting sensorwhich array are is only sensitive to the coverageofarea, while insensitive the and detecting that edgethe defects, precisely the insufficiencies the conventional eddytocurrent material properties andthe detecting edge can defects, which are precisely insufficiencies of the peripheral areas. Thus, sensor array be used thethe discontinuous change of sensors. Crack response is not centrosymmetric, andfor thismeasuring is associated with the aforementioned conventional eddy current sensors. Crack response is not centrosymmetric, and this is associated material properties and detecting edge defects, which are precisely the insufficiencies of the two-layered structure of exciting coil. Based on the analysis, the influence of crack width with the aforementioned structure of above exciting coil. Based on the above thecan conventional eddy currenttwo-layered sensors. Crack response is not centrosymmetric, and this analysis, is associated be influence eliminated by normalization processing of the crack response, thus the transimpedance data are of crack width can be eliminated by normalization processing of the with the aforementioned two-layered structure of exciting coil. Based on thecrack aboveresponse, analysis,thus the uncorrelated with the data crackare width. the transimpedance the crack width. influence of crack width can uncorrelated be eliminatedwith by normalization processing of the crack response, thus the transimpedance data are uncorrelated with the crack width.
(a) part; (b) Real part. Of sensor transimpedance in vertical (b) scanning mode for Figure 4. (a) Imaginary Figure 4. (a) Imaginary part; (b) Real part. Of sensor transimpedance in vertical scanning mode for cracks with different width. Figure 4. (a) Imaginary part; (b) Real part. Of sensor transimpedance in vertical scanning mode for cracks with different width. cracks with different width.
4.2. Effect Analysis of Crack Depth 4.2. Effect Analysis of Crack Depth 4.2. Effect of Crack Next Analysis the influence of Depth crack depth on transimpedance was studied. Crack width and lift-off are Next the influence of crack depth on transimpedance was studied. Crack and lift-off fixed at 0.2 mm and 0.05 mm, respectively, while the depth between 0–2 width mm 0.1 mm Next the influence of crack depth on transimpedance was varies studied. Crack width andin lift-off are areintervals. fixed at Cracks 0.2 mmare and 0.05 mm, respectively, while the depth varies where between 0–2 mm in 0.1 mm beneath the center of the the fixed at 0.2 mm andlocated 0.05 mm, respectively, while thesensing depth element, varies between 0–2 sensing mm in element 0.1 mm intervals. Cracksoutput. are located beneath the center of the sensing part element, where the sensing element has maximum The graph of transimpedance imaginary versus crackthe depth is shown in intervals. Cracks are located beneath the center of the sensing element, where sensing element hasFigure maximum output. The graph of transimpedance imaginary part versus crack depth is shown in 5. has maximum output. The graph of transimpedance imaginary part versus crack depth is shown in Figure 5. Figure 5.
Figure 5. Sensor response of cracks with different depth. Figure 5. Sensor response of cracks with different depth.
As can be seen, theFigure imaginary part of the transimpedance increasesdepth. rapidly at the beginning and 5. Sensor response of cracks with different then As gradually slowsthe down with increasing depth. When the crack depthatreaches 1 mm, and the can be seen, imaginary part of thecrack transimpedance increases rapidly the beginning transimpedance stabilizes. The following corollary can be derived: the sensor array has high then slows down with increasing crack depth. When the crack depth reaches 1 mm, the Asgradually can be seen, the imaginary part of the transimpedance increases rapidly at the beginning sensitivity to small shallow defects, which can detect only 0.1 deep cracks; when array the crack transimpedance Thewith following corollary bemm derived: the sensor hasdepth high and then graduallystabilizes. slows down increasing crackcan depth. When the crack depth reaches 1 mm, exceeds 1 mm, the depth becomes irrelevant the crack response signals. The reason iscrack that depth when to small shallow defects, cantodetect onlycan 0.1 mm deep cracks; thearray thesensitivity transimpedance stabilizes. The which following corollary be derived: the when sensor has high the sensors work under a 3 MHz excitation, the eddy current is concentrated on the surface of exceeds 1tomm, theshallow depth becomes irrelevant the crack signals. The reason is that when sensitivity small defects, which canto detect onlyresponse 0.1 mm deep cracks; when the crack depth the sensors work under a 3 MHz excitation, the7 eddy current is concentrated on the surface of 7 32144
Sensors 2015, 15, 32138–32151
exceeds 1 2015, mm,15,the depth becomes irrelevant to the crack response signals. The reason is that when the Sensors page–page sensors work under a 3 MHz excitation, the eddy current is concentrated on the surface of material, material, hence, even very shallow surface cause great changes in transimpedance. Forcracks, hence, even very shallow surface cracks causecracks great changes in transimpedance. For shallow shallow cracks, the eddy currents tend to bypass them from the bottom of crack, and when cracks the eddy currents tend to bypass them from the bottom of crack, and when cracks are deeper than arethe deeper 1 mm,will thebe eddy current benot blocked uppass and not able to pass up from the crack. bottomCrack 1 mm, eddythan current blocked upwill and able to up from the bottom of the of the crack. Crack impediment on eddy current is consistent, so the response does not change along impediment on eddy current is consistent, so the response does not change along with crack depth. with crack depth.
4.3. Quantitative Analysis of Crack Length
4.3. Quantitative Analysis of Crack Length
Figure 6a6ashows C-scanimage image a together crack together with thewhose position, whose Figure shows aa C-scan of a of crack with the position, size (length × size (length ˆ width ˆ depth) is 5 ˆ 0.2 ˆ 0.5 mm, while the normalized response of the crack width × depth) is 5 × 0.2 × 0.5 mm, while the normalized response of the crack in H mode scanning in H mode scanning and is simulated and shown is simulated shown in Figure 6b. in Figure 6b. Crcack( length 5mm) Strong
(b) Figure C-scan image image of × 0.5ˆmm (b) Simulated peak output curve along curve length along Figure 6. 6. (a)(a)C-scan ofaa5 5× 0.2 ˆ 0.2 0.5defect; mm defect; (b) Simulated peak output direction. length direction.
As can be seen, the crack signal is substantially similar to an isosceles trapezoid, and the width Asthe can be seen,isthe crack signaltoisthe substantially to an isosceles trapezoid, and the of trapezoid closely related length of thesimilar crack. The longer the crack is, the wider thewidth of the trapezoid is closely the of theedges crack.of The longer the is, thelines, wider the trapezoid becomes. It canrelated be seen to that thelength two beveled the trapezoid arecrack not inerratic with inflection points in be theseen middle, correspond edge of the thus the trapezoid becomes. It can thatwhich the two beveledtoedges the crack, trapezoid are quantification not inerratic lines, of crack length lies in the identification of crack edge location. One simple length estimation with inflection points in the middle, which correspond to edge of the crack, thus the quantification of method is to set a threshold, that is, when the difference between the element response the is to crack length lies in the identification of crack edge location. One simple length estimationand method response signal in the absence of defects exceeds a certain value, the element is confirmed to be set a threshold, that is, when the difference between the element response and the response signal in located above the crack, and hence, crack length can be obtained by multiplying the number of the absence of defects exceeds a certain value, the element is confirmed to be located above the crack, crack response elements by the spatial resolution. Other main crack size reconstruction methods are and the hence, crack length can be obtained by neural multiplying theoptimal number of crack response elements by prebuilt crack response database-based network, problem-solving method, etc. the spatial resolution. Other main crack size are the prebuilt crack response . These approaches require simulation of reconstruction variously sized methods cracks in advance and calculation of database-based neural network, optimal problem-solving method, etc. . These approaches require crack response signals, which bring about a significant computational burden; besides, accuracy simulation variously sized cracks in advance calculation crack the response which bring dependsofon the database size and the matchingand algorithm. What of is more, length signals, of the response signal is 6.8 mm versus a real crackburden; length ofbesides, 5 mm. This means that the response a single element about a significant computational accuracy depends on theofdatabase size and the is sensitive to not only theis area below element, but also the area belowisthe element. In crack matching algorithm. What more, thethelength of the response signal 6.8adjacent mm versus a real actual inspection, due to the limited spatial resolution of the array, which is 0.8 mm, the measured length of 5 mm. This means that the response of a single element is sensitive to not only the area below crack response curve is equivalent to the sampling on a continuous curve at a 0.8 mm interval. The the element, but also the area below the adjacent element. In actual inspection, due to the limited crack length sizing method proposed in this paper is named crack edge determination method
spatial resolution of the array, which is 0.8 mm, the measured crack response curve is equivalent to the sampling on a continuous curve at a 0.8 mm 8interval. The crack length sizing method proposed 32145
Sensors 2015, 15, 32138–32151
in this paper is named crack edge determination method using simulation based normalized crack 2015, 15, page–page signalSensors fitting (NCSF). The principle of the NCSF method is to pinpoint the location of crack edges through fitting the discrete measured crack data signal derived from the sensor array based the method normalized using simulation based normalized fitting (NCSF). The principle of theon NCSF simulated crack signals. By embedding the fitting algorithm into the real-time scanning imaging is to pinpoint the location of crack edges through fitting the discrete measured data derived from software, and combining with an encoder, positioning, real-time imaging and length sizing of cracks the sensor array based on the normalized simulated crack signals. By embedding the fitting can be achievedinto synchronously. algorithm the real-time scanning imaging software, and combining with an encoder, positioning, real-time imaging and length sizing of cracks can be achieved synchronously.
5. Experiments and Discussion 5. Experiments and Discussion
5.1. Experimental System Setup 5.1. Experimental System Setup
Signal processing of the sensor array is achieved with a multi-channel transimpedance Signalsystem, processing of the array is achieved with a multi-channel transimpedance measurement which has sensor been described in earlier work . A block diagram of the measurement system, which has been described in earlier work . A block diagram of the measurement system is shown in Figure 7. Fast measurement of 64-channel transimpedance is measurement system is shown in Figure 7. Fast measurement of 64-channel transimpedance is achieved based on FPGA and AD9272 with multiplexers. Transimpedance data is calculated by achieved based on FPGA and AD9272 with multiplexers. Transimpedance data is calculated by FPGAFPGA and and transmitted to the computer port;then thendata datacalibration, calibration, real-time imaging, transmitted to the computervia viaaaserial serial port; 3D 3D real-time imaging, crackcrack localization and automatic crack length sizing are further completed by the computer. localization and automatic crack length sizing are further completed by the computer. Sensor Array
DDS Control Sine and cosine RS232 reference sequences
Inductive Voltages Multiplexer
Figure 7. Schematic diagram of the measurement system.
Figure 7. Schematic diagram of the measurement system.
5.2. Experimental Verification
5.2. Experimental Verification
To verify the reliability of the simulation results and the validity of the NCSF algorithm, two
To verifyspecimens the reliability of the results and validity of the NCSF 8. algorithm, standard machined withsimulation different sized cracks arethe tested, as shown in Figure The specimens specimens are produced by Shandong Ruixiang Mould Ltd. (Jining,are China), which is a professional two standard machined with different sized cracks tested, as shown in Figure 8. manufacturerare of produced NDT testing a member of National TechnicalChina), Committee on is a The specimens byspecimens, Shandongand Ruixiang Mould Ltd. (Jining, which Non-destructive Testing of Standardization Administration of China. Two specimens are both made professional manufacturer of NDT testing specimens, and a member of National Technical Committee of 7075 aluminum alloy, with an actual conductivity of 18.5 MS/m, which is measured with a on Non-destructive Testing of Standardization Administration of China. Two specimens are both SIGMATEST 2.069 US instrument (Foerster Instruments, Pittsburgh, PA. USA). made of 7075 aluminum alloy, with an actual conductivity of 18.5 MS/m, which is measured with a SIGMATEST 2.069 US instrument (Foerster Instruments, Pittsburgh, PA, USA).
Figure 8. (a) Specimen 1 with cracks of different width; (b) Specimen 2 with cracks of different length.
Figure 8. (a) Specimen 1 with cracks of different width; (b) Specimen 2 with cracks of different length. 9
Sensors 2015, 15, 32138–32151 Sensors 2015, 15, page–page
The size of specimen 1 is 250 ˆ 250 ˆ 20 mm, two different wide cracks sizing 5 ˆ 0.1 ˆ 1 mm and 5Sensors ˆ 0.2 ˆsize 115, mm are machined on× its are marked crack 1 and respectively, The of specimen 1 is 250 250surface, × 20 mm,which two different wide as cracks sizing 5 ×crack 0.1 × 12 mm and 2015, page–page 5 × 0.2in× Figure 1 mm are surface, which areˆmarked and five crackcracks 2 respectively, as as shown 8a.machined The sizeon ofits specimen 2 is 250 250 ˆ as 10crack mm,1and whose lengths The of specimen 1 isare 250 × 250 × 20 2mm, two sizing 5 × 0.1 × 1cracks mm and shown insize 8a.11 The size of machined specimen is 250 × different 250 × 10wide mm, and five cracks whose lengths ranging from 3Figure mm to mm on it, as is shown in cracks Figure 8b. The five on the ranging mm to 11 machined on it,are as marked is depth shownas Figure 8b. crack The five the 5 ×of 0.2specimen × from 1 mm3are machined onare itswidth surface, which 1 Specimen and 21respectively, as surface 2 have themm same (0.2 mm) and (1incrack mm). iscracks firstlyoninspected of Figure specimen 2 have same width mm). 1 islengths firstly shown in 8a. The size the of specimen 2 is (0.2 250 mm) × 250and × 10depth mm, and five cracks whose undersurface vertical scanning mode and the transimpedance output curve(1of the Specimen sensing element passing ranging from 3 mm to 11 mm are machined on the it, astransimpedance is shown in Figure 8b. The fiveofcracks on the inspected under vertical scanning mode and output curve the sensing over the center of the crack is obtained and shown in Figure 9. As can be seen, the experimentally surface of specimen theofsame width (0.2 mm) and andshown depth in(1Figure mm). 9. Specimen 1 seen, is firstly element passing over 2thehave center the crack is obtained As can be the measured curves are basically consistent withand theconsistent simulation results. Crack width influences the peak inspected undermeasured vertical curves scanning the transimpedance output curve of Crack the sensing experimentally aremode basically with the simulation results. width of theinfluences outputpassing curve. The theof width is, the greater the element over the the crack is obtained andpeak shown in be. Figure As can the peak of larger the center output curve. The larger the width is, thewill greater the9.peak willbe be.seen, the experimentally measured curves are basically consistent with the simulation results. Crack width influences the peak of the output curve. The larger the width is, the greater the peak will be.
Figure 9. Output curve transimpedance imaginary under vertical scanscan mode. Figure 9. Output curve ofoftransimpedance imaginarypart part under vertical mode.
The response crack 2curve on specimen 1 under H mode scanning is shown Figure Figure of 9. Output of transimpedance imaginary part under vertical in scan mode.10. As can be
The of crackresult 2 onisspecimen 1 under H mode scanning is shown Figure 10. As can seen,response the experimental basically consistent with the simulation result. It isinalso a trapezoid be seen, the response experimental is apical basically consistent with simulation result. Itcan is be also a The crackresult 2the on two specimen 1angles under H mode is shown Figure 10. As structure, peaking of around and flat scanning in the the middle. Dueinto the few sampling seen, the experimental result is basically consistent theand simulation result. It both is also a trapezoid points, non-linear curves cannot be the reflected at thewith rising falling edges sides. This is few trapezoid structure, peaking around two apical angles and flat in the on middle. Due to the structure, aroundcurves the two apical be angles and assistance flat in the middle. Due to mechanism, the few on sampling because thepeaking actual scanning is done manually without of any scanning so thesides. sampling points, non-linear cannot reflected at the rising and falling edges both experimental results fail to reflect all the details. However the comparison of the results reveals points, non-linear curves cannot be reflected at the rising and falling edges on both sides. This is This is because the actual scanning is done manually without assistance of any scanning mechanism, almost of crack signals, which further verifies the validity of the analysis of becauseidentical the actualwidth scanning is done manually without assistance of any scanning mechanism, so the so the experimental results fail to reflect all the details. However the comparison of the results reveals experimental results to response. reflect all the details. However the comparison of the results reveals spatial distribution of fail crack almost identical width of crack signals, which further verifies the validity of the analysis of the spatial almost identical width of crack signals, which further verifies the validity of the analysis of the distribution of crack response. spatial distribution of crack response.
Figure 10. Response of crack 2 in H mode scanning.
10 2 in H mode scanning. Figure 10. Response of crack Figure 10. Response of crack 2 in H mode scanning. 10
Sensors 2015, 15, 32138–32151
5.3. Quantitative Experiment 5.3.1. Planar Objects Testing A crack length sizing experiment is done to validate the NCSF algorithm. Firstly, two defects on specimen 1 are tested separately and repeatedly, and the results are shown in Table 1. As can be seen, the algorithm is highly precise in measuring the length of cracks with different widths. The average measured crack length is around 4.94 mm for crack 1, and around 5.08 mm for crack 2, with measurement accuracy approximating ˘0.2 mm. The experimental results demonstrate that the NCSF algorithm is effective for cracks of different widths. Table 1. Crack length sizing results of specimen 1. Times
1 2 3 4 5
4.92 4.95 4.86 5.08 5.01
5.04 5.12 4.98 5.06 5.16
Cracks with different length but same width and depth (shown in Figure 8b) are tested too. The crack length sizing algorithm based on threshold and the NCSF algorithm are both applied separately for quantitative measurement of actual cracks, and the results are shown in Table 2. Every crack is measured and recorded three times. The results show that algorithm based on threshold is able to size the crack length, but not accurately, with resolution equaling a spatial resolution of 0.8 mm, and error of ˘0.8 mm. While using the NCSF algorithm, crack length sizing accuracy increases, presenting a measurement accuracy of ˘0.2 mm. These results indicate that the optimized flexible sensor array designed herein has a high sensitivity to microcracks, and a sizing accuracy within ˘0.2 mm is obtained using the proposed NCSF algorithm. Table 2. Quantitative results of specimen 2. Threshold Based Algorithm (mm)
NCSF Algorithm (mm)
1 2 3 4 5
2.4 4.8 6.4 8.8 10.4
3.2 5.6 7.2 8.8 11.2
2.4 4.8 7.2 9.6 11.2
3.08 4.89 7.02 9.05 11.18
3.14 5.06 7.09 9.11 11.02
2.95 5.1 7.1 8.86 11.15
5.3.2. Curved Surface Testing An aircraft engine blade with complicated geometry was scanned. An EDM crack with 5 mm length was measured. Thanks to the flexible probe design, the crack is detected well and the length is also measured. A C-Scan of the crack is shown in Figure 11. It is evident that the crack is scanned with good SNR. Therefore, the flexible array probe is suitable for objects with complicated geometry as well as crack length sizing.
Sensors 2015, 15, 32138–32151
Sensors 2015, 15, page–page Sensors 2015, 15, page–page
Figure 11. Scanning of engine blade. (a) Engine blade; (b) Image of crack. Figure 11. Scanning of engine blade. (a) Engine blade; (b) Image of crack. Figure 11. Scanning of engine blade. (a) Engine blade; (b) Image of crack.
5.3.3. Fatigue Crack Testing 5.3.3. Fatigue Fatigue Crack Crack Testing Testing 5.3.3. A fatigue crack on an aluminum plate which is produced by a tension and compression fatigue A fatigue fatigue crack crack on on an an aluminum aluminum plate plate which which is produced produced by by aa tension tension and and compression compression fatigue fatigue A testing machine is also scanned. Figure 12 showsisthe specimen and the result. The size of the crack is testing machine is also scanned. Figure 12 shows the specimen and the result. The size of the crack is testing is also×scanned. Figure 12 shows the result. Thetosize the crack 10.838machine mm (length) 0.042 mm (width). As canthe bespecimen seen fromand Figure 12b, due theofsmall widthisof 10.838 mm (length) ˆ 0.042 mm (width). As can be seen from Figure 12b, due to the small width of the 10.838 mm (length) 0.042peak mm value (width). can be seen Figureto 12b, to the small widthBoth of the fatigue crack,× the is As relatively low from compared thedue EDM crack signal. fatigue crack, the peak value is relatively low compared to the EDM crack signal. Both quantitative the fatigue crack, theof peak relatively low compared the EDM signal. quantitative results enginevalue bladeisand fatigue crack are listed intoTable 3. As iscrack shown in theBoth table, results of engine blade and fatigue crack are listed in Table 3. AsinisTable shown3.in the table, the result is a quantitative results of engine blade and fatigue crack are listed As is shown in the table, the result is a little smaller than the true value. This is probably because the fatigue crack is not in a littleresult smallerathan the true value. This is value. probably because the fatigue crack is not incrack a straight line. the smaller than the true is probably the shown fatigue is not in a straight is line.little There is a segment stray from theThis crack line at thebecause left corner, in Figure 12a. There is a segment stray from the crack line at the left corner, shown in Figure 12a. straight line. There is a segment stray from the crack line at the left corner, shown in Figure 12a.
Figure 12. (a) Specimen with fatigue crack; (b) Output of single element under V mode scanning. Figure Figure 12. 12. (a) (a) Specimen Specimen with with fatigue fatigue crack; crack; (b) (b) Output Output of of single single element element under under V V mode mode scanning. scanning. Table 3. Quantitative results of blade crack and fatigue crack. Table and fatigue fatigue crack. crack. Table 3. 3. Quantitative Quantitative results results of of blade blade crack crack and
Times Times 1 Times 12 1 23 2 3 43 4 54 5 5
Blade Crack (mm) Blade Crack 4.85(mm)
Blade Crack (mm)
4.85 4.82 4.824.96 4.96 5.08 4.96 5.08 5.08 5.01 5.01 5.01
Fatigue Crack (mm) Fatigue Crack (mm) Fatigue Crack 10.80 (mm) 10.80 10.69 10.80 10.69 10.69 10.52 10.93 10.5210.52 10.9310.93 10.65 10.6510.65
6. Conclusions 6. Conclusions A novel flexible eddy current sensor array for quantitative measurement of micro fatigue cracks A novel flexible eddy current sensor array for measurement of micro fatigue cracks in key components of aircraft, etc. is designed in quantitative this paper. The flexible design conforms the sensor into key components of aircraft, etc. is designed in this paper. The flexible design conforms the sensor the curved surface of engine blade. The principal 32149 of the sensor array and the response of cracks to the curved surface of engine blade. The principal of the sensor array and the response of cracks 12 12
Sensors 2015, 15, 32138–32151
6. Conclusions A novel flexible eddy current sensor array for quantitative measurement of micro fatigue cracks in key components of aircraft, etc. is designed in this paper. The flexible design conforms the sensor to the curved surface of engine blade. The principal of the sensor array and the response of cracks with different sizes are demonstrated by simulation. The NCSF algorithm is developed to size crack length with high accuracy. Experimental results are proved consistent with simulations. All results prove that the array is not only sensitive to microcracks, but also capable of crack length sizing. It has the following advantages: 1. 2. 3.
Flexible, and capable of inspecting complex geometric structures; Sensitive to microcracks, and capable of crack length sizing, with an accuracy within ˘0.2 mm; High spatial resolution, reaching 0.8 mm; array element of 64 mm, covering a length of over 50 mm; high detection efficiency; sensitive to microcracks with different directions.
Acknowledgments: This research is sponsored by the National Natural Science Fund of China (project number: 61171134). Author Contributions: Ruifang Xie proposed the idea. Ruifang Xie and Dixiang Chen arranged the methods and also contributed to the writing and editing the manuscript. Wugang Tian and Weihong Zhou ensured the realization of the experimental device and prepared the data. Mengchun Pan and Xuezhong Wu guided the research and were charged with the critical revisions of the manuscript. All of the authors analyzed the results and revised the paper. Conflicts of Interest: The authors declare no conflict of interest.
References 1. 2. 3. 4.
9. 10. 11.
He, Y.; Pan, M.; Luo, F.; Tian, G. Pulsed eddy current imaging and frequency spectrum analysis for hidden defect nondestructive testing and evaluation. NDT & E Int. 2011, 44, 344–352. Tian, G.Y.; He, Y.; Adewale, I.; Simm, A. Research on spectral response of pulsed eddy current and nde applications. Sens. Actuator A Phys. 2013, 189, 313–320. [CrossRef] Rosado, L.S.; Santos, T.G.; Ramos, P.M.; Pedro, V.; Piedade, M. A differential planar eddy currents probe: Fundamentals, modeling and experimental evaluation. NDT & E Int. 2012, 51, 85–93. Rosado, L.S.; Gonzalezd, J.C.; Santosd, T.G.; Ramos, P.M.; Piedade, M. Geometric optimization of a differential planar eddy currents probe for non-destructive testing. Sens. Actuator A Phys. 2013, 197, 96–105. [CrossRef] Cha, Y.-J.; Nam, B.; Kim, J.; Kim, K.H. Evaluation of the planar inductive magnetic field sensors for metallic crack detections. Sens. Actuator A Phys. 2010, 162, 13–19. [CrossRef] Chen, Z.; Janousek, L.; Yusa, N.; Miya, K. A nondestructive strategy for the distinction of natural fatigue and stress corrosion cracks based on signals from eddy current testing. J. Press. Vessel Technol. 2007, 129, 719–728. [CrossRef] Shejuan, X.; Zhenmao, C.; Hong-En, C.; Xiaowei, W.; Takagi, T.; Uchimoto, T. Sizing of wall thinning defects using pulsed eddy current testing signals based on a hybrid inverse analysis method. IEEE Trans. Magn. 2013, 49, 1653–1656. Marchand, B.; Decitre, J.-M.; Sergeeva-Chollet, N.; Skarlatos, A. Development of flexible array eddy current probes for complex geometries and inspection of magnetic parts using magnetic sensors. In Proceedings of the 39th Annual Review of Progress in Quantitative Nondestructive Evaluation, Denver, CO, USA, 15–22 July 2012; pp. 488–493. Lepage, B. Development of a flexible cross-wound eddy current array prob. Int. J. Appl. Electromagn. Mech. 2014, 45, 633–638. Rosado, L.S.; Santos, T.G.; Piedade, M.; Ramos, P.M.; Pedro, V. Advanced technique for non-destructive testing of friction stir welding of metals. Measurement 2010, 43, 1021–1030. [CrossRef] Sheiretov, Y.; Evans, L.; Schlicker, D.; Zilberstein, V.; Goldfine, N. TBC characterization using magnetic and electric field sensors. In Proceedings of the ASME Turbo Expo 2007, Montreal, QC, Canada, 14–17 May 2007; pp. 1–7.
Sensors 2015, 15, 32138–32151
13. 14. 15. 16. 17.
Endo, H.; Nishimizu, A.; Tooma, M.; Ouchi, H.; Yoshida, I.; Nonaka, Y.; Otani, K. Signal evaluation system of flexible array ect probes for inspecting complexly shaped surfaces. In Proceedings of the Review of Progress in Quantitative Nondestructive Evaluation, San Diego, CA, USA, 18–23 July 2010; pp. 597–603. Deng, G.; Sakanashi, Y.; Nakanishi, T. A practical method for fatigue crack initiation detection using an ion-sputtered film. J. Eng. Mater. Technol. 2009, 131, 71–76. [CrossRef] Lim, Y.Y.; Soh, C.K. Electro-mechanical impedance (EMI)-based incipient crack monitoring and critical crack identification of beam structures. Res. Nondestruct. Eval. 2014, 25, 82–98. [CrossRef] Muchs, H.O.; Stephens, R.I.; Saunders, H. Metal fatigue in engineering. Int. J. Fatigue 1981, 3, 346. Huang, H.; Sakurai, N.; Takagi, T.; Uchimoto, T. Design of an eddy-current array probe for crack sizing in steam generator tubes. NDT & E Int. 2003, 36, 515–522. Dixiang, C.; Ruifang, X.; Weihong, Z.; Hengjiang, H.; Mengchun, P. Multi-channel transimpedance measurement of a planar electromagnetic sensor array. Meas. Sci. Technol. 2015, 26, 025102–025108. © 2015 by the authors; licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons by Attribution (CC-BY) license (http://creativecommons.org/licenses/by/4.0/).