Integration of piezoelectric transducers (PZT and PVDF) within polymer-matrix composites for structural health monitoring applications: new success and challenges

ABSTRACT This article investigates the interest of using in-situ piezoelectric (PZT and PVDF) disks to perform real-time Structural Health Monitoring (SHM) of glass fiber-reinforced polymer composites submitted to various tensile loadings. The goal is to evaluate the working range and SHM potential of such embedded transducers for relatively simple mechanical loadings, with the long-term aim of using them to monitor complete 3D structures submitted to more complex loadings. SHM is performed acquiring the electrical capacitance variation of the embedded transducers. To study the potential links between the in-situ capacitance signal and the global response of the loaded host specimens, a multi-instrumentation composed of external Nondestructive Testing techniques was implemented on the surfaces of the specimens to search for multi-physical couplings between these external measurements and the capacitance curves. Results confirmed the non-intrusiveness of the embedded transducers, and allowed estimating their working domain. PZT capacitance signal follows well the mechanical loadings, but the piezoceramic transducer gets damaged after a determined relatively low strain level due to its brittleness. The limits of this working domain are extended by using a stretchable PolyVinylidene Fluoride (PVDF) polymer transducer, allowing this one to perform in-situ and real-time SHM of its host tensile specimens until failure. GRAPHICAL ABSTRACT


Introduction
Due to their attractive characteristics, such as lightness along with good mechanical properties, composite materials are widely used in different industries. However, as they are increasingly employed in high-tech industrial sectors whose requirements are very strict, such as aeronautics, aerospace, or wind power sector, and as their highly heterogeneous and anisotropic properties make their damage modes very complex, the need to study their mechanical behavior is primary. Polymer-Matrix Composites (PMC) structures can encounter severe environmental conditions such as temperature, humidity or radiation, and/or various mechanical loadings like tensile, shear, or torsion. Thus, it is of significant importance to have a precise idea of the stress/strain rates they are experiencing, and also of their damage initiation and propagation to decide when to put them out of service for maintenance or replace them if their damage is too severe. Structural Health Monitoring (SHM) allows doing this, using many surface and/or volume Nondestructive Testing (NDT) techniques such as Strain Gages (SG) [1,2], Ultrasound (US) [3,4], Infrared Thermography (IRT) [5,6], Acoustic Emission (AE) [7,8], electrical [9][10][11]/electromagnetic [12,13] testing and others. These methods can either be used alone or combined [14], to confirm the obtained measurements (redundancy) and to make them more accurate thanks to information coupling. However, even if these techniques can provide useful information, they remain external, and the corresponding instrumentations are often cumbersome and cannot be left on the structures permanently. PMC structures have thus to be regularly put out of service to be checked, which is time and money consuming. Furthermore, the data cannot provide information for exact damage occurrence and progression scenarios because they are gathered punctually and not in real-time. These issues are rising the interest in 'smart' composite materials nowadays, this interest going along with the actual trend for functionalized structures used for various applications such as energy harvesting, vibration control, or morphing. As the NDT devices are incorporated within PMC structures, it is possible to perform their real-time SHM without any of the previously cited disadvantages. NDT devices also being protected from external environmental conditions thanks to their host composite parts [15]. Another advantage of embedding NDT devices instead of letting them external is the potential increase in damage sensitivity [16][17][18][19][20][21][22][23]. In-situ monitoring, though, has some disadvantages, such as manufacturing/embedding difficulties [24], or intrusiveness risks as the integrated device may degrade the mechanical properties of the monitored part [25]. Different types of devices have already been embedded into composite structures to perform insitu SHM, such as optical fibers ( [26,27]: monitoring of cracks), or piezoresistive elements including conductive fiber reinforcement ( [28]: delamination identification), conductivecoated fibers ( [29]: strain monitoring under cyclic flexural loading), nanofillers ( [30]: relation electrical resistance vs. mechanical loading [31]: uniaxial tension-tension fatigue monitoring), Z-pins ( [32]: delamination crack surveillance), and tuft threads ( [33]: tufted 'Omega' stiffener monitoring). Piezoelectric devices are also particularly investigated in the in-situ SHM literature, with the use of PolyVinylidene Fluoride (PVDF) [34] or Lead Zirconate Titanate (LZT or PZT) [17] as primary base materials. They are used to create various types of in-situ transducers such as piezoceramic discs/patches, thin PVDF layers, piezoelectric fibers, and others, which are then inserted inside composite structures to perform SHM in active or passive mode using different measurements such as electric charge variations [35,36], impedance metre [37], Lamb waves [38] or even AE signatures [19][20][21][22][23]25,34,39].
Considering the variety of literature on this domain even until recently [40], which proves its current scientific importance, extending research on real-time in-situ SHM of PMC specimens is of great interest. The present work focuses on the surveillance of a 'smart' (PMC) material incorporating a thin-connected piezoelectric disc (PZT or PVDF) transducer during various tensile tests equipped with an external NDT multiinstrumentation, composed of strain measurement devices (Stereo Digital Image Correlation (SDIC), Strain Gages (SG), or extensometry), and Acoustic Emission (AE). These multi-instrumented tests were realized to evaluate the intrusiveness of such an embedded transducer, but also to understand the in-situ capacitance signal coming from this transducer by comparing it with several ex-situ volume (AE) and surface (SDIC, SG, extensometer) signatures, with the aim of using it to evaluate the working domain of the transducer as well as its real-time and in-situ SHM ability.
The conducted experiments are detailed in the next section, followed by the exploitation of the obtained results. First, an intrusiveness study will be presented, conducted on both pristine and 'smart' (with in-situ PZT) PMC specimens during different tensile loadings, to confirm the possibility of using such transducers inside PMCs without compromising the intrinsic mechanical properties of the host material. Then, the working range and SHM potential of both embedded PZT and PVDF transducers will be studied and compared. To do so, correlations will be conducted between the different NDT signatures and the capacitance signal coming from the in-situ transducer. A concluding section, discussing the potential of both PZT and PVDF transducers as well as their possible improvements, will end this paper.

Piezoelectric and PMC materials
PZT and PVDF transducers of the same dimensions (Figure 1(a)) are used in this study. They are composed of a disk-shaped piezoelectric base material, either a ceramic, or a thermoplastic polymer, coated on each face with silver electrodes. Although PZT disks were bought with fixed dimensions, PVDF material was bought as a metallized piezo film sheet, which makes it possible to cut samples out of it with any wanted dimensions (increased versatility). Their choice was mainly motivated by their potential low intrusiveness due to their small dimensions (Table 1), and their possible use in passive (sensor) and/ or active (actuator) modes. These features made them perfect candidates to manufacture 'smart' specimens for in-situ and real-time SHM. Their static capacitance C 0 was measured with an LCR bridge to confirm the datasheet provided by the manufacturer before their embedding and wiring in the middle plane of the fibrous stack used to manufacture the specimens. Thus, PZTs (or PVDFs) were regularly spaced in line on the middle fiber ply of the stack, as displayed in Figure 1(b), to manufacture a 300*300*5 mm 3 plate containing six specimens, each one being pristine or 'smart' (embedding one transducer at its mass center). The resulting plate manufactured using Liquid Resin Infusion (LRI) technique as well as one 'smart' sample cut from the latter are shown in Figure 1(c), and the positioning and wiring of the transducer inside a 300*40*5 mm 3 specimen are presented in Figure 2(a, b). This positioning was chosen for several reasons. First, to have the embedded transducer located inside the 'useful zone' of the specimen, between the clamping jaws of the tensile machine, and where the strain field is homogeneous. Another reason for choosing the sample center as embedding area was to avoid any possible edge effects which could have interfered with the proper functioning of the transducer. Finally, this embedding spot gives maximum protection against the external environment, as a non-negligible thickness of PMC is encapsulating the transducer. The wiring of the embedded devices was done using tinned copper wires of 210 µm in diameter, and LRI manufacturing was performed at ambient temperature in an air-conditioned room, with a fixed hardener concentration. • For 1 plate:800 g (the excess used to expel as many air bubbles as possible from the infusion system) A decrease of the static capacitance C 0 of the transducers to a value named C init (different for PZTs and PVDFs) when embedded in the consolidated specimens was noticed, due to the hydrostatic pressure applied on them by the cured resin [49,50]. All details about the materials used to manufacture the samples can be found in Table 1.

Experimental methods
The manufactured 'smart' and pristine samples were submitted to three kinds of tensile tests until failure (monotonic, load-unload, and incremental fatigue) to simulate as many real-life cases as possible. To ensure sample breaking in the useful zone, four PMC heels were glued on the four ends of each sample, using bi-components Araldite 2015 A/B glue. During tensile tests, the cross-head speed was 2 mm/min, and the holding times were conducted at a fixed cross-head displacement of the electromechanical 200kN load cell tensile machine. The three loading profiles are presented in Figure 3(a,b,c), with at least two samples tested for each loading profile.
During the tests, the electrical capacitance of the in-situ transducers was recorded and displayed in real-time using an LCR bridge connected to an acquisition chain. At the same time, a multi-technique experimental device acquired external NDT data, which were used to infer the intrusiveness of the transducer, and compared with the capacitance ones for SHM considerations. This experimental device was composed of: • Two AE sensors and a 2D planar strain gage or extensometer for the monotonic tensile test until failure, • Two AE sensors and two SDIC systems or an extensometer for the load-unload tensile test until failure, • Two AE sensors and an extensometer for the incremental fatigue test until failure.
AE sensors were glued on the same face of the sample using grease as a coupling agent, surrounding the embedding area of the transducer. Each of the two SDIC systems was composed of two high-resolution cameras, each pair being focused on a portion of either the face or the thickness of the tested sample. These portions, including the embedding area of the transducer, were covered with a black/white speckle pattern, used to compute strains maps in the three directions of space on the two speckled zones: 60*40 mm2 for the face and 60*5 mm2 for the thickness. The mechanical extensometer, as well as the 2D planar strain gages (rosettes), were positioned on the face of the sample, at the same level as the embedded transducer.
The aim of gathering all these information is dual. First, it allows to infer the intrusiveness of the transducer, comparing the intrinsic macroscopic mechanical properties (Young's modulus, mechanical resistance, etc.) of both pristine and embedded 'smart' specimens. In a second step, these collected external NDT data will be coupled with the electrical capacitance signal coming from the in-situ transducer during the test, so the latter is better understood and its working range and SHM potential well-evaluated. All the essential details about each used device can be found in Table 2, and examples of multi-instrumented tensile tests are illustrated in Figure 4(a,b).

Results and discussions
Results for intrusiveness and working range of the embedded transducers, as well as their SHM ability, are presented in this section. Comparisons are made between PZT and PVDF as an integrated working SHM device for parts submitted to tensile loadings. All AE data used in this study have been located inside the area framed by the two AE sensors using Noesis software, to get rid of any interference noise coming from the clamping jaws or other sources not being a PMC damage. Strain curves coming from the SDIC maps have been computed with Vic3D software, using the entire speckle patterns, either on the face (60*40 mm2) or the thickness (60*5 mm2) of the samples. glued with grease (coupling interface).

Intrusiveness study
This subsection evaluates the consequences of embedding a piezoelectric transducer inside a PMC tensile specimen, by measuring the variation of the intrinsic macroscopic mechanical properties of the material with or without an integrated transducer. The potential impact of the tensile loading mode on intrusiveness is presented and discussed.
Since PVDF material is more ductile (flexible) and compatible with the polymer resins, this section is focused on the intrusiveness of PZT transducers which are more fragile and incompatible with PMC matrices, as well as a little bit thicker than the PVDF transducers.

Monotonic tensile loading
The mechanical behavior of both pristine (P) and embedded 'smart' (S) specimens (1 insitu PZT) submitted to monotonic tensile loading until failure are compared in Figure 5, and their corresponding intrinsic mechanical properties are displayed in Table 3. The Young's modulus is computed using longitudinal strains between 0.05% and 0.25%, as recommended by standard NF EN ISO 527-4.
As it can be noticed in Figure 5 and Table 3, the principal mechanical properties of PMC 'smart' specimens embedding PZT transducers in their mass center are very similar to those of the tested pristine specimens. Looking at the data, the insertion of the PZT gives a 6% increase in mechanical resistance, an 11% increase in maximum longitudinal strain, a 5% decrease in Young's modulus, and a 26% decrease in Poisson's ratio. The variations in Young's modulus and mechanical resistance can be considered negligible, and the decrease in Poisson's ratio for the 'S' samples can be explained by the presence of a very stiff body inside the samples, lowering their lateral (ε xx along x axis) contraction during the tests and thus the Poisson's ratio (-ε xx /ε yy ). The increase in ε yy was unexpected,  considering the small elongation potential of the PZT ceramic, and also contributed in the Poisson's ratio decrease. An explanation could be the PZT being able to oppose lateral contraction better than the longitudinal extension, as the values for the last one are much higher. However, the gap between maximum mean values of ε yy for pristine and smart specimens is only 0.17% according to Table 3. As the PZT embedding does not have major impact on the mechanical properties of the specimens, this transducer can thus be considered non-intrusive for this type of loading.

Load-unload tensile loading
The same kind of pristine and smart specimens, embedding the same PZTs, were submitted to load-unload tensile loading until failure. Intrusiveness was once again evaluated, comparing the Damage Indexes (D E = 1-E i /E 0 ) calculated using stiffness evolution during the tests. The stiffness values for each loading block i were taken during the unloading phases, so that the whole loading history is influencing Young's modulus evolution for each block. This could not be ensured if the stiffness values were taken during the loading phases. All the used D E will be computed this way.
The D E for both pristine and embedded 'smart' specimens are summed up in Figure 6. Figure 6 allows to confirm the non-intrusiveness of the PZT transducer during loadunload tensile loading as the D E curves for pristine and smart specimens are very close to each other, with a stiffness loss around 20% at 250 MPa for both configurations, meaning that the embedding of the PZT inside a specimen will not lead to a faster or higher decrease in its mechanical properties during load-unload tests.

Incremental tensile fatigue loading
The intrusiveness of the PZT is studied for this last but different kind of tensile loading until failure, as it is a dynamic one. D E have been used once again, and are presented in Figure 7 for both pristine and embedded (1 in-situ PZT) samples. As for the previous D E , the values of the Young's modulus have been taken during the unloading phases at the end of each fatigue block, after having undergone 5000 loading cycles in each one. Looking at Figure 7, it is clear that D E curves for pristine and smart specimens are very close to each other, reaching the same loss in stiffness (around 20%) after having undergone 11 incremental fatigue blocks of 5000 loading cycles each. This allows to affirm the non-intrusiveness of the PZT for this type of loading.
The non-intrusiveness of the PZT has thus been established for three kinds of tensile loadings until failure, being quasi-static or dynamic. There is no impact of the type of tensile loading on intrusiveness. The non-intrusiveness of the PVDF transducer can thus be deducted from the non-intrusiveness of the PZT, as it is of smaller dimensions as well as more flexible and compatible with the polymer matrix, as stated at the beginning of this subsection. This intrusiveness was a very important point to check, as the novel functionality provided by the insertion of these transducers would be useless if the intrinsic mechanical properties of the PMC had been compromised by the insertion of such as foreign body. As intrusiveness is a big worry for industrials willing to use smart materials for their structures, due to the harsh certifications to take, this study will confirm the industrial interest of such materials. The SHM potential of both transducers is thus worth studying, with the associated results presented in the next subsection.

SHM ability of the embedded piezoelectric transducers
Considering the negligible intrusiveness of the embedded transducers when smart PMC specimens are loaded in tension, as demonstrated in 3.1, their SHM potential is worth studying. Thus, in this subsection, the electrical capacitance signature coming from the transducers (PZT and PVDF) during the quasi-static tensile tests performed in 3.1 is valued as an SHM indicator, by coupling it with the external NDT measurements realized during the mechanical tests. First, the working range of the transducers and their ability to follow mechanical loadings in real-time are evaluated, then their use to assess damage occurring to the PMC samples is discussed. PZT and PVDF will be compared in this subsection.

PZT
Typical curves obtained during a multi-instrumented monotonic (a) and load-unload (b) tensile tests until failure are shown in Figure 8. The PZT capacitance variation (ΔC =C-C init ) is plotted in red, the stress curve is blue, the strains are detailed in green, and finally the acoustic activity located between the two AE sensors is represented in pink using the Cumulative Absolute Energy (CAE) of the AE hits.
The analysis of Figure 8(a,b) shows that the PZT capacitance signal seems able to follow the mechanical loadings (stress and strains rises and decreases), as well as the stress relaxation during the holding times for each block of the load-unload test. This allows to infer a link between these signatures having the same trends. It is observed that the capacitance varies linearly with the strains in the three main directions and for the two types of loadings, as displayed in Figures 9 and 10. Results are shown repeatable, with a maximum standard deviation of 7.2% between the slopes. What can be inferred is that if one of the strains evolves in a significant way due to damage of the composite specimen (here no linear evolution anymore), in-situ capacitance should be affected. PZT transducers seem thus able to give information on strain evolution in the three directions of space using electrical capacitance measurement, when placed inside a specimen submitted to monotonic or load-unload tensile loading. However, Figures 9 and 10 only show the observed relationship between strains and capacitance until a particular strain threshold around 0.3% (for ε yy ). But looking at the entire strain range reached during the tests in the three directions of space (example in Figure 11), a repeatable linearity loss between PZT capacitance and mechanical strain is observed beyond this limit. Thus, going beyond this so-called Non-Working Threshold (NWT) could compromise the reliability of the signal coming from the embedded transducer, as no major damage can have occurred at such a small strain level considering the mechanical properties of the PMC specimens. This is the case for the three studied strains, and the two kinds of tensile loadings.
This NWT is associated to a repeatable PZT ΔC slope change around an ε yy of 0.3%, which can be observed in Figure 8(a,b), this value corresponding to the maximum longitudinal strain level acceptable by such a ceramic material before it starts cracking. The damage threshold (NWT) of the ceramic was of course expected, and was precisely evaluated thanks to these tests, as no information on its maximum elongation was provided by the PZT supplier.  However, another reason for this slope change could be the formation of a cavity at the PZT/composite interface, due to the very different strain behaviors of these two bonded materials. The birth of this cavity would modify the stress state around the embedded PZT in its environment, provoking the ΔC slope change. This possibility has been investigated below.
Several methods were employed to assess the continuity of this PZT/composite interface. First of all, postmortem imaging was conducted on the cut and polished smart sample interfaces. The sample preparation method is displayed in Figure 12(a). These images, realized for the two types of tests (monotonic and load-unload), were then compared to a reference image taken on the interface of an unloaded smart sample. To check the continuity of the material in the interface area, an elementary analysis was conducted on all the samples using Scanning Electron Microscopy (SEM) imaging. This analysis showing all the chemical elements forming the matter, each one indexed by a color, the void associated to a potential cavity would appear in black. Looking at Figure  12(b,c), corresponding to damaged interfaces of smart samples submitted to monotonic and load-unload tensile tests, no black zones are observed around the interface. This ensures the continuity of the matter in this area, confirming that any damaged sample has an interface as healthy as the reference unloaded sample (Figure 12(d)). Considering these results, it was decided to overturn the formation of a cavity at the PZT/composite interface and to focus on the PZT cracking.
The PZT cracking phenomenon has been characterized with several methods, the first one being the detection of the crack using its acoustic signature. To achieve this, the acoustic activity of both pristine and smart specimens, in terms of Cumulative Absolute Energy (CAE) of hits, have been compared for monotonic (Figure 14(a)) and load-unload (Figure 14(b)) tensile tests. For longitudinal strain (ε yy ) values taken around 0.25%-0.4%, a significant CAE rise is noticed for the smart specimens, while it is not the case for pristine specimens. These CAE rises are associated with the previously described PZT ΔC slope changes, and cannot result from damage of the PMC material itself as they do not appear for the pristine specimens. They are thus linked to a damage of the PZT transducer itself, because the PZT/composite interface integrity is preserved until the end of the test, as described above. In the same time, a non-supervised clustering procedure (K-means) of  the AE signals allowed to associate this particular CAE rise with an AE cluster (named 'PZT cracking') starting around the capacitance slope change for the monotonic tests ( Figure  14(c)), and from the first capacitance slope change happening at the fourth loading block for the load-unload tests (Figure 14(d)), with regular reoccurrences at each following block until the end, as the transducer is being damaged more and more. As displayed in Figure  14(c,d), the contribution of this 'PZT' cracking' cluster to the first jump of these CAE rises is almost full, meaning that this jump could be associated to the PZT cracking phenomenon, and not to another one related to damage of the PMC specimen itself. As a potential 'PZT cracking' cluster has been found for two types of tensile loadings, it was decided to compare them to confirm that it is the same occurring physical phenomenon. To do so, the main AE descriptors of both clusters have been compared one by one. These fundamental parameters, most of them being described in Figure 13, are the amplitude of the wave (maximum value of the wave with respect to 0, named AMPL in dB), its duration (period of time during which the signal is energetic enough to cross the detection threshold defined by the user, named DURA in μs), its number of counts (number of times the alternative signal cross the detection threshold, named CNTS), its number of counts to peak (same thing than the CNTS, but until the wave peak, named PCNT), its rise time (duration from the first threshold crossing to the wave peak, named RISE in μs) and its absolute energy (computed by integrating the squared value of the signal on its duration, named ABEN in aJ). Thus, these descriptors of both potential 'PZT cracking' clusters (values taken at the cluster center) have been compared for the two types of tensile tests, as displayed in Figure 15, and are found to be relatively close to each other one by one. This confirmed the existence of a repeatable physical phenomenon for smart samples submitted to either monotonic or load-unload tensile loadings, this phenomenon occurring at the same time as the PZT capacitance slope changes, and occurring around a strain level compatible with the cracking of a ceramic material.
SEM imaging of damaged and reference (not damaged) smart samples micro-cuts (PZT transducer embedding area) have been realized, with the same sample preparation as Figure 13. Some of the main descriptors describing an AE wave. described in Figure 12(a). The resulting micrographs are presented in Figure 16. As expected, whatever the tensile loading profile was, one or several cracks can be observed in the piezoceramic material of the damaged samples, whereas it is not the case for the reference smart sample, imaged on the whole length of the micro-cut (the overall PZT diameter is visible). To visualize the crack propagation within the whole PZT transducer, tomographic imaging was realized on a damaged load-unload smart specimen. A developed crack is clearly visible on the resulting tomographic image displayed in Figure 17 a, located in the PZT disc plane and oriented along the tensile direction. To confirm the initiation moment and associated strain level of the crack, another loadunload test was conducted on a similar sample, but stopped after the end of the fourth loading block, after which the first capacitance slope change was appearing in a repeatable way. Figure 17(b) shows one slice of the resulting tomographic scan,   where a crack beginning can be observed, oriented the same way as in Figure 17(a), confirming that the embedded piezoceramic material starts getting damaged after the fourth loading, meaning the first overshooting of the ε yy strain level compatible with ceramic cracking. The PZT cracking phenomenon is thus fully evaluated and associated to the NWT of the transducer, previously highlighted in the present paper.
Another consideration was the ability of the embedded PZT transducers to detect damage initiation and spreading inside the PMC specimens. When having a first look at Figure 8(a,b), the appearance and spreading of damage inside the PMC samples are noticeable thanks to the CAE curves changes, whereas the stress or strain curves do not show particular variations allowing to infer them. When comparing the CAE curve to the PZT ΔC one for each kind of test, before the NWT to be sure that the PZT output signal can be trusted, no specific CAE variations can be noticed either, which is not that surprising as the stress/strain levels are still relatively low, meaning very few damages. Damage detection and follow-up cannot thus be done for these loadings with this type of embedded transducer, as real damage occurs after the NWT of the PZT, for higher strain levels. However, the PZT is able to detect the final failure of the samples in a repeatable way, whatever the tensile loading mode was.
This subsection showed the ability of the in-situ PZT capacitance to follow various tensile loadings and highlighted a connection between capacitance and mechanical strains for a composite specimen submitted to uniaxial and quasi-static tensile loadings, which could be used as an SHM indicator as damage can have direct impact on strains evolution. However, this ability is limited by the NWT associated to the intrinsic brittleness of the piezoceramic material, preventing it to overtake elongation higher than 0.3% or 0.4% according to the conducted experiments. The studied PMC materials having maximum strain rates around 1.5% or 2%, in the longitudinal tensile direction, the in-situ PZT cannot be used for SHM in these configurations. It has to remain below its NWT to give reliable information on the health of its host specimen. That is why it is of major importance to employ a piezoelectric transducer made of a more flexible material, susceptible to accommodate such strain levels without being compromised. The PVDF was chosen to overtake this challenge, and the corresponding results obtained with it are presented in the next subsection.

PVDF
The PVDF is a thermoplastic polymer exhibiting piezoelectric properties and has the advantage to be relatively flexible compared to a PZT. As explained in the Materials and Methods section, PVDF discs with same dimensions as PZT transducers were embedded into the same PMC specimens previously tested. Results obtained for a typical multiinstrumented load-unload tensile test until failure are presented in Figure 18. The color code is the same as the one used for the previous subsection.
Several statements can be made when looking at Figure 18. First, the PVDF capacitance signal seems to have a linear relationship with measured longitudinal strain, similarly to the PZT transducer. But the most interesting point is the absence of capacitance slope changes during the whole test, allowing the ΔC signal to stay stable and linear until sample failure around 1.7% of longitudinal strain. This stable signal is explained by the nature of the used piezoelectric material, a flexible polymer that can support much higher strain levels than the ceramic do before starting to get damaged. The acoustic activity acquired during the test is compared to the one of a similar pristine PMC specimen submitted to the same load-unload test, as presented in Figure 19. Studying this figure allows to observe the start of the CAE rise for the pristine (Figure 19(a)) and smart ( Figure  19(b)) CAE curves around the same stress threshold (190 MPa), with no specific variations associated to the CAE of the smart specimen. This confirms that both specimens (pristine and smart) start getting damaged at the same stress level, meaning that the embedding of the PVDF disc inside the PMC sample does not change the specimen damage scenario during the mechanical test. This result ensures the non-intrusiveness of the PVDF and goes in the direction of its remaining integrity until the end of the test. To check that integrity, micrographs of both reference (not tested) and damaged smart samples embedding PVDFs were realized using a videomicroscope. Comparing these micrographs in Figure 20(a,b) allows ensuring the remaining integrity of the PVDF embedded inside the damaged sample, as it presents no signs of degradation, just like the reference.
As the PVDF integrity until sample failure is ensured, the study of its capacitance signature has been performed the same way it was for the PZT transducer. The results, displayed in Figure 21, allow confirming the previously observed linear relationship between PVDF ΔC and ε yy , this one being usable almost until sample failure on the contrary of the PZT due to the absence of capacitance slope change. The PVDF capacitance thus enables evaluating the longitudinal strain almost until the end of this type of test, 1.3% minimum according to the tested specimen, which is three or four times higher than allowed by the PZT. The observed linear relationship is also different from the one acquired with the PZT, as its electrical capacitance variation is much smaller: 0.7% at ε yy =0.25% against 30% at ε yy =0.25% for the PZT. The NWT has thus been highly raised by the change of transducer base material, allowing the PVDF to be considered a serious candidate for in-situ SHM inside materials reaching such strain levels before failure.
Its SHM potential has thus started to be studied during load-unload tensile tests, by comparing the evolution of the D E (measured during unloading phases) with the evolution of the capacitance decreasing slopes (see Figure 18) during these same unloading phases. The obtained results, displayed in Figure 22, show a strong correlation between the trends of the two evolution signals, opening the way for the use of the in-situ PVDF  capacitance variation rates as a tool to evaluate in real-time the progressive loss of mechanical properties of the loaded specimen in which it is embedded. Finally, as, the PZT, PVDF capacitance allows to detect the sample final failure.

Conclusions and outlooks
In this innovative work, the interest of embedding thin piezoelectric (PZT and PVDF) disc transducers inside PMC specimens submitted to various tensile tests, to perform their in-situ and real-time SHM, was investigated. Several internal (transducer electrical capacitance) and external (MDIC, SG, extensometer, and AE) NDT instrumentations were implemented during the tests to make multi-physical correlations between the different obtained measurements. The first part of the study allowed to ensure the non-intrusiveness of the in-situ PZT and PVDF transducers for the three kinds of performed tensile tests: monotonic, load-unload, and fatigue. The following part investigated the working range of such embedded transducers, as well as their SHM potential during monotonic and load-unload tensile tests in terms of sensitivity to the mechanical loadings, as well as damage detection and evaluation. Thanks to the correlations obtained with the multi-instrumentation, it was observed that the  electrical capacitance of the PZT was following well the different tensile loading trends (loading, unloading, relaxation), and a linear relationship was observed between capacitance and strains in the three directions of space. However, its brittleness was found to be an obstacle to its use in SHM of such PMC specimens, the latter having maximum strains way higher than the evaluated strain threshold (said NWT), beyond which it cannot be trusted anymore due to its cracking. That is why the PZT gets deteriorated before the real damage occurs inside the PMC specimen, making the transducer unable to monitor it. This challenging issue was overtaken by changing the base material of the piezoelectric transducer, switching from a brittle ceramic to a flexible and more stretchable thermoplastic polymer called PVDF. Its use allowed to observe a linear relationship between the electrical capacitance of the transducer and the mechanical loadings (stress and strain) almost until final sample failure, multiplying the previous NWT by three or four without being intrusive. Consequently, its reliable signature was used to assess in real-time the progressive loss in stiffness of a PMC sample submitted to load-unload tensile loading until failure, and the results are found to be very promising. This study showed that the PVDF should be preferred to a PZT for in-situ and realtime SHM applications broadly speaking, as it does not get damaged that easily. In addition, as it is a stretchable material, it can be placed even in complex parts (curved areas, etc.) to perform SHM in any spot, and it can be cut off its initial sheet with desired shape and dimensions, which makes it of high versatility compared to the PZT. Finally, as it is a thermoplastic polymer, its insulating nature allows it to be embedded even in conductive PMC parts, for example, Carbon Fiber Reinforced Polymer (CFRP) structures used in high-tech industries. However, it is important to remember that it also has drawbacks, as it is not a very good actuator compared to a PZT, making it not easily usable for SHM applications using the ultrasonic technique. It also delivers relatively weak signals compared to the PZT (capacitance in pF and not in nF), necessitating an adapted measuring system to acquire them. It is also useful to remember that the PZT was found to be brittle in that case study, but embedding it in less strained spots so that its NWT (which is now determined) is not exceeded, for example on the neutral fiber of a specimen or structure loaded in bending, is highly practicable. Moreover, the NWT of the PZT could be monitored quite easily as the acoustic signature of the 'PZT cracking' cluster has been determined in this study. It is thus necessary to make a compromise for the choice of transducer, regarding the type of wanted measurements, the solicitation applied on the smart structure, and the area wherein the transducer will be embedded. The sensitivity/intrusiveness challenge must always be kept in mind and optimized as much as possible, by adapting the volume of the part wherein the transducer is embedded to minimize intrusiveness, also knowing that it has to stay reliable and sensitive enough to the surrounding damage.
This study is still ongoing, as repeatabilities must be made to evaluate the SHM potential of the PVDF submitted to tensile loadings, but the long-term goal remains in the application of both PZT and PVDF transducers to perform in-situ and real-time SHM of more complex 3D composite structures, such as stiffeners. The idea would be to combine the assets of both transducers, and perform SHM (damage detection, evaluation, location) as accurately as possible using them in passive (capacitance, Acoustic Emission) and/or active (Ultrasonic) mode. Other potential leads would be to investigate the manufacturing of versatile thinner and already wired piezoelectric or piezoresistive transducers for the same studied SHM purposes, using innovative and trendy manufacturing techniques such as electrospinning or spin coating.Glossary (by order of appearance within the text)

Disclosure statement
No potential conflict of interest was reported by the authors.