Determination of the elastic properties in CFRP composites: comparison of different approaches based on tensile tests and ultrasonic characterization

. The mechanical characterization of composite materials is nowadays a major interest due to their increasing use in the aeronautic industry. The design of most of these materials is based on their stiffness, which is mainly obtained by means of tensile tests with strain gauge measurement. For thin laminated composites, this classical method requires adequate samples with specific orientation and does not provide all the independent elastic constants. Regarding ultrasonic characterization, especially immersion technique, only one specimen is needed and the entire determination of the stiffness tensor is possible. This paper presents a study of different methods to determine the mechanical properties of transversely isotropic carbon fibre composite materials (gauge and correlation strain measurement during tensile tests, ultrasonic immersion technique). Results are compared to ISO standards and manufacturer data to evaluate the accuracy of these techniques.


Introduction
Fibre-reinforced composites are widely used for many structural applications.The primary benefits of the composite components are the reduction of weight and the simplification of assembly (Soutis 2005).The determination of the mechanical properties, specially the stiffness, is essential for ensuring performance to the composite structures.In addition, the knowledge of complete elastic stiffness matrix is important for modeling and evaluating the mechanical behavior of composite materials under loading conditions (El Bouazzaoui et al. 1996).
For now, tensile test with strain gauge measurements is the technique normalized by ISO standards to identify composites elastic properties.Such conventional method is destructive and provides only a part of elastic constants when thin plates like laminated composite structures are considered.Accordingly, ultrasonic techniques based on the measurement of ultrasonic wave velocities provide an interesting and non destructive way to address such issue.
First experimental studies based on ultrasonic technique were done by Zimmer and Cost (1970).They have used transmission contact technique to obtain dynamic elastic stiffness necessary to derive the unidirectional composite elastic behavior.Several ultrasonic bulk wave methods have also been developed to obtain phase velocities in anisotropic plates, especially when only one or two sides of the sample are accessible (Chu et al. 1994, Vishnuvardhan et al. 2007).
Immersion ultrasonic technique represents a particularly suitable method in the case of composite materials with small thickness (Munoz et al. 2014, Reddy et al. 2005).For a transversely isotropic composite, measurements in symmetry planes are sufficient to entirely determine the stiffness tensor (that is the five independents constants) (Baste and Hosten 1990).Various numerical methods and/or technical devices have then been implemented to interpret experimental results (Balasubramaniam andWhitney 1996, Kawashima et al. 1998).
Generally speaking, two experimental protocols can be used for such immersion technique to get velocities measurements, namely through-transmission and back-reflection techniques (Reddy et al. 2005).Through-transmission method requires two transducers, one to send the wave through the sample and the second one to receive the transmitted wave (Margueres 2000, Franco et al. 2010).As the ultrasonic wave travels through the test sample, the wave is reflected in part as it encounters a medium of different acoustic impedance.Then the transmitted wave is received by the transducer and displayed or stored for analyses.The difficulty of this method is that it requires a tracking of the arrival wave and therefore difficult to implement in the case of immersion.
In the back-reflection technique, a transducer working in pulse/echo mode is associated to a large flat reflector which is positioned parallel to the transmitter.The back-reflected wave travels exactly along the same path as the incident wave in the opposite direction (Rokhlin and Wang 1992).When the sample angle is changed, the position of the incident wave on the back reflector is modified.Compared to through-transmission, such method appears then much relevant in the immersion case since it is not necessary to move the reflector or the transmitter/receiver transducer due to the large dimensions of the reflector.
The aim of this paper is to compare the back-reflection immersion ultrasonic method with classical mechanical characterizations based on tensile tests.Strain gauges are used to measure the axial and lateral strains and stiffness constants are then estimated from the elastic parts of the stress-strain response.At the same time, a digital image correlation system is also implemented to corroborate the gauge data.Regarding immersion technique, a specific device including a rotation system has been set up to study the response of the material under various incident waves.These three results of stiffness tensor measurement (strain gauges, digital image correlation during tensile tests and ultrasonic characterization) are finally compared and discussed.

Specimen
In view of its increasing use in aeronautics, a carbon fibre reinforced laminate is considered for this study.The M10R/38%/UD150/CHS composite is made of 14 unidirectional plies of prepreg leading to a thickness h=2 mm.Sample fabrication is carried out using the manual lay up technique.Then the sample is cured at 125°C during 90 minutes at a pressure of 2 bars as recommended by the material manufacturer Hexcel ® .The plate was checked using ultrasonic C-h&'&()012'031 34 '5& &627'08 9(39&('0&7 01 @ABC 83)9370'&7D 83)92(0731 34 E044&(&1' 299(3285&7... pFGH 1 Samples for tensile test.Plane 1-2 is the laminate plane Ian and was found to be homogeneous and defect free.
The dimensions of the single sample used for the ultrasonic characterization are 150×100×2 mm 3 .The density of this composite is measured experimentally, namely ρ=1449 kg/m 3 .This material is representative of a transversely isotropic composite.

Experimental procedure for the tensile test method
The elastic behavior determination based on mechanical tests requires to apply axial tensile load with different orientation regarding the symmetry axis of the material.In this way, laminated samples with three axis fibre directions (0°, 90° and 45°), (see Fig. 1) have been cut according to the standard ISO 527-5 for unidirectional composites (Standard NF EN ISO 527-5 2009) leading to sample dimensions of 250×20×2 mm 3 for 0° specimens and 250×25×2 mm 3 for 45° and 90° specimens.It is important to note here that samples were obtained from the same cured plate in order to limit data scattering.With those tests, only 4 elastics constants can then be determined since loading in the direction of axis 3 (orthogonal to the laminate plane) cannot be done for thin composites plates.
Mechanical tests are carried out at ambient temperature in a room thermoregulated at 25°C.Uniaxial tensile tests are performed by an electromechanical testing machine INSTRON 5500.According to the standard, the displacement rate is 2 mm/min for 0° samples, 1.5 mm/min for 45° samples and 1 mm/min for 90° samples.To check the reproducibility of the response, three specimens are considered for each loading direction.
Strain gauges of 350 Ohm (HBM K-LY41-6/350-3-2M, length of 6 mm) are connected to a data acquisition system HBM Spider 8.30 and the data processing is made with Catman Easy software.Digital Image Correlation (DIC) is also performed during tensile tests.Such technique aims at matching two digital images of a surface observed at two different states of load, generally in the reference state (unloaded) and in a deformed state (Sutton et al. 2000).The markers used for matching the images with the DIC system are done by painting white points on the sample surface (see Figs. 2-3); they form a rhombus of 23×20 mm 2 .The processed image with four markers allows then to obtain the strain during the load for each direction (axial and lateral).The DIC uses an Aramis sensor made of two CCD cameras 1392×1040 pixels definition (displacement resolution of 5 µm).However for plane structures, only one camera is needed for measuring the iqr stutvtnce image of DIC iwr xy yqt q σ rupture /2 2 DIC images of the 0° sample 3 Set-up for tensile characterization method defgf hif jf kekf hl jf hmnfe f ofokefo lhg pqrdfifd dq hl the markers (see Figs. 2-3).The markers quality has been checked in order to ensure recognition by the CCD sensor of different gray levels and to allow an accurate tracking of each pixel during jf oflhgh hef.The calibration of the system is easily done by taking images of a edmh fd kof ollff ffeif ifs; moreover, 35 mm focal length has been chosen lh jf egf df o jf oef mfsff jf etkh qfg o fegf uvv gg, fkdp d fhdkh hl mhk wv×80 µm 2 (pixel size).Images are acquired from the mfpp xt=0 s) to the end of the test; according to the important data involved and related ehgkhd hpf ollekdfy jf etkh lftkfeq lzfo h hf gpf f feho {dq imdf dpj j mffn used to avoid any perturbation due to temperature variations.Gauges and monitored markers are positioned on the same face of the specimen.

|}|
According to standards, elastic constants are calculated for strain states between P R /10 and P R /2 where P R is the yield strength of the specimen.With the 0° tensile test, Young modulus E 1 and Poisson ratio v 12 are classically deduced from the ratios of axial stress and axial and lateral strains.
Young modulus E 2 =E 3 is calculated in the same way from 90° tensile test.Finally, Eq. ( 1) provides the shear modulus G 12 by calculating Young modulus E 45°o btained with the 45° tensile test (1) As said before, this procedure does not allow to obtain the last elastic property, namely the Poisson ratio v 23 , that would entirely determine the elastic behavior of the laminated composite.

Experimental procedure for ultrasonic characterization 2.3.1 General principle
The elastic stiffness matrix for a transversely isotropic material is given by the following tensor written in the coordinate axes described in Fig. 1 (2) Ultrasonic characterization aims at providing five components representative of the elastic behavior of such material, namely C 1111 , C 2222 , C 1122 , C 2323 and C 1212 (C 2233 can be derived from C 2222 and C 2323 based on the elastic relations).This determination relies on the resolution of the Christoffel equation (Rose 1999) (3) where V(n) is the wave velocity related to the propagation direction of unit vector n, ρ is the material density and I is identity second-order tensor.Accordingly, this requires the measurement of ultrasonic velocities for different orientations of wave propagation in the sample.For a transversely isotropic composite, measurements in planes of symmetry are sufficient to determine all five independent elastic constants (Aristégui andBaste 1997, Chu et al. 1994).
For the immersion ultrasonic technique, water acts as a couplant that transfers the wave from the transducer to the sample under inspection (Castagnède et al. 1990, Rokhlin andWang 1992).The transducer is not directly connected with the sample and hence consistent coupling is ensured.In this way, it is possible to measure the wave velocities at different angles of propagation either by adjusting transducer orientation or by rotating the sample (Reddy et al. 2005).Here the sample is held and moved using a rotation system (see Fig. 4).As ultrasonic generator, an Omniscan 32: 128 PR is used with a mono-element transducer connected to it.The transducer acts at the same time as ultrasonic source and receiver (reflection mode).Classical value of frequency is used, namely 5 MHz, which allows mainly to avoid inside reflection; back-reflector and specimen were checked for alignment at normal incidence.
The measurements are performed for different angles Ф between plane 2-3 and incident plane (see Fig. 5).For a given Ф, the sample is then rotated in the incident plane for different incident angles θ i between the transducer and axis 3 of the sample.The wave passes through the sample in ¦ §¨4 Back-reflection set-up ¡¢£ Coordinated system ¡¤£ ©ª«¬¢®¯¦°±¢²³ ´¢«µ ¥¦ §¨5 Wave path description.Ф: angle between plane 2-3 and incident plan direction.θ ¶ • incident ¢¯ §ª³¨θ r : refraction angle ¸¹º »¼½º¾¸¼¿À ¿Á n with angle θ r , so θ r is function of Ф and θ i .For each set of angles (Ф and θ i ) the time of flight is measured.

Procedure for the elastic constants estimation
The determination of the velocity of propagation in a medium requires the exact time of flight (TOF) in the specimen for the given angle of propagation.Such time is determined using the cross-correlation technique (Rao et al. 1993).It requires a reference signal, which corresponds here to the signal recorded without the sample (only water).The time to cross the material is obtained by subtracting the TOF along the reference path in the coupling medium (water) to the overall TOF along the path in the sample at a given incident angle.This time difference Δt gives then the ultrasonic velocity V(θ r ) [m/s] in the specimen for the refracted angle θ r by the following relation (Vishnuvardhan et al. 2007) (4) where V 0 is the velocity in water (without sample) and θ i the incidence angle (radians).V 0 depends on the water temperature; this velocity has been measured for each test to account for ÂÃÄ ÚÛÜÝ 6 Reference velocity V Þ for the different samples and angles Ф ßàá âãäãåãaeçã èÛÜaeàé àê Ф=0°ßëá ìÛÜaeàé àê Ф=0°and θ í =0°Ú ÛÜÝ 7 Determination of times of flight îïðñòóïôîïõö÷ ðöòñöõñóïø ùøîî úñûü ý) and it has been assumed to be constant during each test (duration of each test is around 30 minutes).The average value of V 0 is 1488.5 m/s.
Measurements were done by rotating the sample at 1°±0.05° step resolution.Fig. 7 illustrates reference signal at Ф=0°without sample (only water, Fig. 7(a)) and the signal at Ф=0°and θ i =0 from which Δt=t 1 −t 0 is derived (Fig. 7(b)).Note that the measured times t 0 and t 1 , and consequently the time difference Δt, are classically defined at the most important amplitude of the signal (Reddy et al. 2005); this point will be discussed later.
The transducer senses both longitudinal and quasi-shear waves.Yet, the longitudinal velocity V L of the first one is always more important than the quasi-shear velocity V QS of the latter; this allows to distinguish their own contribution in the velocity signal (Reddy et al. 2005).These both waves are generated when the incident wave encounters the sample.
In the present case, only two configurations of the incident plane are necessary to derive the elastic constants, namely Ф=0°and Ф=90°.Fig. 8 shows the ultrasonic velocities obtained in p 8 Ultrasonic velocity vs. incident angle in plane 2-3 (Ф=0°) p 9 Ultrasonic velocity vs. incident angle in plane 1-3 (Ф=90°) p !" #$%$ &' ()0 Ф=0°).The longitudinal velocity V L and quasi-shear velocity V QS are almost constant for all angles of propagation, which stands in agreement with the transverse isotropy of the material around axis 1.The slight deviation of velocity V L (only 8% for a range of 28° of incident angle) can be explained by some misalignment of plies during the manufacturing process.In contrast, regarding the ultrasonic velocities in plane 1-3 (that is for Ф=90°, Fig. 9), both longitudinal and shear velocities are clearly affected by the incident angle.This confirms the anisotropic character of the material in this plane (Vishnuvardhan et al. 2007).Note finally that, for incident angles bigger than 60°, no signal can be detected.
Ultrasonic measurements in plane 2-3 are directly related to the constants C 2222 and C 2323 .Component C 2222 is determined from normal incidence longitudinal velocity, that is for C 1111 , C 1122 and C 1212 are obtained from both longitudinal and shear velocity data measured in 1-3 plane.Experimental velocity data is fitted in the following two relations obtained from the Christoffel equation's solution (Eq. ( 3)).This allows then to determine the three unknown parameters.To solve undetermined Eqs. ( 7)-( 8), one can use the nonlinear least-square optimization technique which minimizes the deviations between the experimental and theoretical velocities for the considered angles of propagation (Reddy et al. 2005) In Eq. ( 11), n is the number of independent parameters to be extracted (here 5 elastic constants) and m is the number of measurements of velocities in different directions (here 137 experimental data).V e and V t are the experimental and theoretical phase velocities, respectively.
In a last step, component C 2233 is calculated using this relation between elastic stiffness components (12) Engineering moduli in the coordinate system (E 1 , E 2 =E 3 , v 12 , v 23 and G 12 ) can be deduced by inverting the stiffness tensor.The standard deviation σ is found by taking the square root of the average of the squared differences of the values x i from their average value ( 14)

Results and discussions
vwxy 10 Calculated elastic constants de fg h results are compared to this method.The error on an elastic property P relative to its reference value is defined as follow (15) Indeed, we can see a good accuracy between the 0° tensile test with strain gauges and the only data provided by the manufacturer, namely the axial Young modulus E 1 .Even if UD composites usually exhibit a good reproducibility, especially on E 1 , dispersion obtained in our case in the stress-strain response can be linked to the manufacturing process (since the manual lay-up cannot warrant a perfect alignment of the fibres) and also to the misalignment of gauges.
Regarding DIC technique, we can note a very accurate determination of Young moduli E 1 (error of 5% on the mean) and E 2 (error of 2% on the mean).It should be noted that the deviation on these moduli is smaller than the one obtained with gauges.Since DIC method ensures the measurement of the strain along the vertical axis, this tends to confirm the influence of the sensors position in the reproducibility problem observed with gauges.Results are less precise for Poisson ratio v 12 (error of 15 % on the mean) and shear modulus G 12 (error of 8% on the mean).Latter moduli are determined from lateral strains measurement.For carbon fibre composite subjected to axial tension, these strains exhibit low amplitude and are then more affected by measurement inaccuracies.It is also important to note here that calculus of G 12 accumulates the errors derived US technique is the single one that provides Poisson ratio v 23 .Among elastic properties, the value of E 1 is the most accurate result in US characterization in regards to the normalized tensile test (error of 11% on the mean) and corresponds to a good estimation if the standard deviation is taken into account.Results on mean values of E 2 (error of 22% on the mean) and Poisson ratio v 12 (error of 23 % on the mean) seem less accurate but, again, they stand in the same range as the reference according to the standard deviation of both techniques.The higher error is found for G 12 (error of 76 % on the mean).All these tendencies (especially, better accuracy on E 1 compared to E 2 and important error on shear modulus G 12 ) stand in agreement with works of Reddy et al. (2005) andVishnuvardhan et al. (2007) on graphite-epoxy plates with the same ultrasonic technique (immersion back reflection technique).It should be noted that similar trend was also found by Reddy et al. (2005) with through-transmission.All these deviations appear then to be linked to the estimation of the TOF of the wave inside the material that can be difficult to determine with accuracy.In the present study, the device and measurement protocol of times t 0 and t 1 have been optimized to ensure the best reproducibility.As said before, a classical definition of the TOF based on the peak of amplitude has been used.Yet, for increasing angle θ i , the wave amplitude becomes weaker (Fig. 11(a)), so as the signal-to-noise ratio (Fig. 11(b)).In this latter case, it may be then hard to distinguish the peak amplitude and other definitions of TOF may be employed.As an example, some deviation is obtained between the measurement based on the real peak of amplitude and the measurement based on a theoretical peak for that signal derived from some polynomial fitting (Fig. 11(b)).

¯. Conclusions
This paper intended to compare different methodologies to establish the elastic behavior of composite materials.Transversely isotropic composites were considered here through the study of unidirectional carbon fibre/epoxy laminates.Two strain measurement techniques (gauge as standard, Digital Image Correlation) during tensile tests and the immersion ultrasonic method were analyzed.
Results highlight first the interest of DIC to determine the strain response and its good accuracy on the final determination of elastic constants.Yet, tensile tests do not allow to entirely provide the elastic stiffness and some indetermination remains for structures calculation.
This study has shown also the ability of the immersion ultrasonic technique to derive all elastic constants of a UD laminated composite through the measurement of time of flight inside a sample.The main advantage of this technique is that only one specimen is needed instead of cutting many samples with desired size and shape for a classical characterization tensile test.If elastic properties estimations are quite encouraging, some limitations on the precision have been noted that would require further investigations on the influence of the time of flight.This could be improved by adjusting the experimental set-up in such a way that the distance between transducer and reflector is fixed during all the acquisition or by using two transducers (through-transmission).Also, several signal data processing will be tested for weak amplitude signals in order to estimate the influence of the definition of the TOF on the elastic properties.
Given that, this immersion ultrasonic method could then be extended with the same principle to more complex anisotropic materials including orthotropic symmetry.This would need another degree of freedom to rotate the sample that will provide the possibility to propagate ultrasonic waves along more azimuthal angles/planes with respect to the fibre axis.

Fig. 10
Fig. 10 summarizes the elastic constants estimation obtained by means of the three different methods used to characterize the carbon fibre/epoxy composite.The arithmetic mean is calculated as the sum of the sampled values divided by the number N of tests (N=3 for each method) (13) Φ=0°and θ =81°.Weak amplitude -to-noise ratio for different 11 Signal at different measurements ¡¢£ ¤¥¦ §¨© §ª §©«¬ £¦¬®«¡¦£¦¨¤® ¢ E 1 , E 2 , v 12 and E 45 .However, elastic constants obtained by DIC clearly stand in the range of standard measurements if we consider the standard deviations of each technique.DIC appears then as an interesting alternative to strain gauge, especially as DIC avoids contact and is more practical to set-up.