A single three-parameter tilted fibre Bragg grating sensor to monitor the thermosetting composite curing process

Abstract The unique sensing features of the tilted Fibre Bragg Grating (TFBG) as a single three-parameter optical sensor are demonstrated in this work, to monitor the manufacturing process of composite materials produced using Vacuum Assisted Resin Transfer Moulding (VARTM) process. Each TFBG sensor can measure simultaneously and separately strain, temperature and refractive index (RI) of the material where the optical fibre is embedded. A TFBG embedded in a 2 mm glass-fibre/epoxy composite plate was used to measure the thermomechanical variations induced during the curing process. At the same time, the RI measurements, performed with the same TFBG sensor, can estimate the degree of cure of the resin. The TFBG sensor shows to be a valid and promising technology to improve the state of art of sensing and monitoring in composite material manufacturing. Graphical Abstract


Introduction
Composite materials are widely used in different engineering sectors such as aerospace, aeronautics, automotive, naval, wind turbines and railways [1] since their high strength to weight ratio and anisotropic nature brings several advantages compared to traditional engineering materials. In particular, the use of composite in primary structures has allowed significant weight reduction while maintaining the same mechanical performance in the aeronautic sector. In the last decades, a significant weight reduction has been achieved by replacing solid metallic parts with composite material [2]. Nevertheless, several defects can be introduced due to a poorly designed manufacturing practice. This can lead to unacceptable level of defects and rejection of the part, which poses cost and sustainability issues [3,4]. Therefore, inspection techniques to investigate the internal state of composites during manufacturing have been developed. Since the first monitoring technologies appeared, Optical Fibre (OF) sensing immediately proved to be a valid option [5], and specifically, Fibre Bragg Grating (FBG) sensors are the most used OF sensors for embedded real-time monitoring of composite structures. They provide accurate and reliable remote multi-parameter measurements (temperature, strain, pressure, etc.), low weight, minimal intrusiveness and high installation flexibility, and additionally, they are immune to electromagnetic interference [6]. Since the 2000s, the embedding of OF sensors has been performed for composites manufacturing quality control [7][8][9][10], as it is a fundamental step to obtain the best mechanical performance [11]. Due to dominant heat transfer phenomena and low thermal conductivity through thickness, some defects such as voids, temperature overshoots, residual stresses and part deformation can occur during composite manufacturing process, which can significantly lower the mechanical performances of the components. A significant effort on the optimisation of the composite manufacturing processes has been undertaken to minimise the occurrence of such defects [12][13][14][15][16]. Moreover, the monitoring of a specific manufacturing process stage is necessary to avoid unexpected defects to arise. At the same time, multiple sensors strategy is not desirable as their presence can generate defects or influence the material mechanical performance, and it would increase the cost and, hardware and software complexity. Some solutions have been developed by considering the benefits of the OFs using a combination of similar and/or hybrid sensors in the same waveguide able to compensate or perform dual-parameter measurements [17][18][19][20][21][22][23][24]. However, they suffer from several drawbacks which make them unattractive, such as nonlocalized measurements, poor spatial resolution and accuracy, use of an intrusive capsule and time consuming. TFBGs are characterised to have a Bragg grating structure tilted with respect to the optical axis of the OF. This special imposition allows a spectrum composed of several well-defined resonance peaks to be obtained [25]. By exploiting these peaks in the TFBG, after a calibration step, each single TFBG sensor can be used to measure, simultaneously and separately, the variation of strain, temperature and Refractive Index (RI) at the point where it is embedded [26]. Specifically, the RI variation of the resin during its curing stage can be monitored and associated with its degree of cure [27]. However, there are no works proving that the TFBG can perform simultaneous three parameter measurements when embedded in a composite and that a single sensor is sufficient to monitor the thermomechanical and curing state of the material.
In the present paper a single three-parameter optical sensor based on a weakly Tilted FBG (TFBG) is demonstrated to monitor the thermomechanical state and the degree of cure of a thermosetting composite during the Vacuum Assisted Resin Transfer Moulding (VARTM) manufacturing process. The simultaneous temperature, strain and RI measurements with a single embedded TFBG during the curing stage of a 2 mm composite, are reported with a section dedicated to the comparison between the degree of cure and the RI trend during the curing time.

TFBG sensing theory
In a short period tilted grating the periodic and permanent RI modulation of the OF core, is deliberately tilted with respect to the longitudinal axis of the waveguide [28]. The tilt angle (h) is super-imposed during the Bragg grating writing process through tilting parts of the FBG manufacturing machine [25]. This angle enhances the generation of multiple peaks in the TFBG spectrum [29]. Specifically, in this paper, only reflective weakly TFBGs (h < 15 ) are considered since, apart from the Bragg and the cladding resonance peaks, they have a further peak in their spectrum called the Ghost resonance, which allows the measurement of multiple parameters simultaneously [25][26][27][28][29][30][31]. All the resonance peaks in the spectrum are sensitive in different measure to thermomechanical external perturbations, which cause a shifting variation of the nominal peak wavelength. However, the Bragg and Ghost peaks are immune to external RI variations, while the cladding resonances undergo a wavelength shift and amplitude change when this is varying. Considering these aspects and by performing a preliminary calibration, both the Bragg and the Ghost peaks can be used to measure simultaneously and separately the temperature and the strain variation from a reference condition. At the same time, the area generated by the envelope of the lower and upper cladding resonance peaks can be exploited to measure independently the variations of the RI surrounding the TFBG sensor [8,31,32].
Once the sensitivity coefficients (K e , K T ) of the Bragg and Ghost peaks are obtained from the calibration, their wavelength shifting (Dk Bragg , Dk Ghost ) are used to measure separately the strain (De) along the OF axis and temperature (DT) variations using the K global thermomechanical sensing matrix [33]. This technique is based on the peaks wavelength shifts acquired through the FBG interrogator device, which can detect data points with a certain scanning wavelength resolution (swR). Consequently, swR also influences the sensor thermal resolution, which can be calculated through the ratio between the thermal sensitivity coefficients difference and swR. The measurements were performed by respecting this resolution constraint, which omission implies incorrect calculations.
Regarding the surrounding RI, the cladding resonance peaks envelope area is calculated using the Delaunay triangulation (D-T) demodulation technique [31]. During the refractometric calibration, the area values are calculated and correlated at each external RI change and normalised with respect to a reference area value. At this point, from the correlation points, a fitting function can be obtained in relation to the external RI range of interest. Hence, the surrounding RIs can be obtained by solving the fitting function with respect to the normalised envelope area of the TFBG immersed in a medium. This process and the D-T technique are treated in more detail in [31].

Experimental
In the following section 3.1 provides details on the reinforcement fibre and matrix, and the TFBG sensor used to perform the measurements inside the composites, while section 3.2 treats the cure kinetics and T g characterisation. Section 3.3 provides details on the VARTM process and sensor embedding stage.

Materials and sensors
The 2 mm thick composite plate (12 plies) was manufactured using Interglass TM Unidirectional (UD) Sglass fibre (220 gr/m 2 , 92145) plies and epoxy resin. The resin system was a low temperature curing Hexion Epikote TM 04908 epoxy resin mixed with Epikure TM Hardener 04908 (mixing ratio resin/hardener 100:30 parts by weight) [34]. The vacuum bag setup consisted of a nylon film WrightlonV R 7400, sealant tape Solvay LTS90B, infusion mesh Airtech Greenflow 75, peel ply Airtech Stitch Ply A and release perforated polyolefin foil WrightlonV R WL3700. The TFBG sensor embedded in the composite, was manufactured by FORC-Photonics in Fibercore PS1250/1500 OF with a 2 tilt angle, and is 4 mm long with a 10 mm uncoated length. The TFBG signal was acquired by using the FBG interrogator NI PXIe-4844, which has a swR ¼ 4 pm and a maximum sample frequency of 10 ± 0.1 Hz. A thin K-thermocouple (TC, Ø ¼ 0.250 mm) was also embedded close to the TFBG to provide a temperature measurement reference for comparison with the TFBG measurements. The TC has an accuracy of ±1 C. The integration details are reported in section 3.3. Before TFBG embedding, the thermomechanical calibration of the TFBG sensor was performed as in [33]. The strain and temperature sensitivity coefficients found are Ke ,Bragg ¼1.255 ± 0.004 pm/le and Ke ,Ghost ¼1.255 ± 0.006 pm/le while KT ,Bragg ¼8.686 ± 0.012 pm/ C and KT ,Ghost ¼9.2 ± 0.014 pm/ C.
The RI calibration procedure of the TFBG was performed as described in [35], with the room temperature kept at 20 ± 1.5 C. Usually the epoxy resins suitable for VARTM processes have a RI range 1.5-1.56 [36] when uncured, which is expected to increase during the curing stage [27]. For the resin used in this study, the RI of the uncured resin is around 1.54 at 25 C [34]. Hence, to obtain the best measurement accuracy, the curve branch in the range 1.46-1.7 was fitted by a fifth-order polynomial function with a fitting square error (R 2 ) of 0.9996, where the worst RI accuracy is 1Â10 À3 at 1.5.

Cure kinetics and T g characterisation
The cure kinetics characterisation of the epoxy system was carried out with a TA Instrument Differential Scanning Calorimeter (DSC) 2500. The DSC used liquid nitrogen with a flow rate of 10 ml/ min. Two isothermal tests at 80 C and 100 C and one dynamic test at 1 C/min were performed. The glass transition temperature evolution was characterised using Modulated Differential Scanning Calorimetry (MDSC), at a 3 C/min ramp rate with a modulation set at 1 cycle/min and an amplitude of 1 C, by observing the evolution of the reversible components of the specific heat. Next to the fully cured and fully uncured samples, four partially cured samples were manufactured by heating uncured resin samples at 1 C/min up to an increasing final temperature followed by a quick cool down at the fastest possible machine rate of about 50 C/ min to stop the reaction from progressing. To validate the cure kinetics model, two additional samples have been partially cured. The cure profiles used dictated a ramp-up to 80 C at 1 C/min followed by an isothermal dwell of 20 min for the first sample and 40 min for the second. After that, the samples have been quickly cooled down to stop the reaction. An MDSC analysis to identify the glass transition temperature of the sample has been subsequently performed. The resin cure kinetics and Di Benedetto equation were fitted to the experimental results as shown in the following section.

Degree of cure and Di benedetto equation
The degree of cure and the glass transition temperatures trend are obtained from the experimental data. These have been fitted with the following kinetics model proposed by Khoun et al. [37], the model proved to accurately describe similar epoxy resin systems [14]: where a is the degree of cure, a c , a T , are coefficients controlling the transition of the kinetics from chemical to diffusion; C governs the breadth of the transition into the diffusion controlled regime, m and n are reaction orders for the n-th order and autocatalytic terms, A is a pre-exponential Arrhenius factor, E is the activation energy of the Arrhenius functions, T is the absolute temperature, and R is the universal gas constant. Figure 1a shows the fitting of the experimental data with the proposed phenomenological kinetic model. The average relative error of the fitting is about 2%. The heat generated Q by the exothermic reaction can be calculated as follows: Where q r is the resin density, v r the volume resin fraction and H tot is the total enthalpy. The glass transition temperature model to fit the experimental data follows the Di Benedetto equation [38]: here T g1 and T go are the glass transition temperatures of the fully cured and uncured material respectively and k is a fitting parameter governing the convexity of the dependence. Figure 1b illustrates the fitting of the experimental data with the Di Benedetto equation.

VARTM process
The VARTM manufacturing process was used to produce the samples and consists of an infusion stage followed by a curing stage. The OF was placed parallel to the UD reinforcement fibres layers of the composite to measure longitudinal strain. The mould was a flat aluminium plate (50x50x1 cm). The resin was infused at room temperature. The TFBG was embedded using the same translation stage of the TFBG calibration, to hold the TFBG straight and in the centre position of the reinforcement layer inducing a small pre-tensile force to the OF. The TFBG was then spot-glued on the glassfibre foil using small drops of cyanoacrylate glue close to the edges of the composite layer ( Figure 2). Additionally, a TC was placed as close as possible to the glued TFBG sensor by using a special web adhesive which melts once the resin covers it and, in a way that does not interfere with the material surrounding the OF sensor. The TFBG and TC were embedded in the middle plane (6 th reinforcement layer), after this, the preform can be assembled. Figure 3 gives a schematic view of the sample with the sensors embedded. The OFs and the TC wires exiting from the composite were protected with vacuum bag sealant tape. Resin infusion was performed at 50 mbar to avoid the evaporation of volatiles. During resin flow, the TFBG sensor was used to provide information on the flow arrival time whose details are reported in section 4.1.
After completing the infusion, the inlet line was closed and the panel was cured in an oven provided with an access hole from which the infusion lines, the OFs and the TCs can be externally connected. The applied curing temperature profile is suggested by the resin manufacturer [34]. As no indications are given by the manufacturer regarding ramp rates, a heating-up of 1 C/min and cooling-down by natural convection were imposed to have a gradual temperature variation. The strain, temperature and RI measurements were performed by processing the TFBG spectra acquired during the curing time and these are reported in the results section.

Results and discussion
This section reports the outcomes of the TFBG measurements from the spectra acquired during the different stages of the VARTM process. In section 4.1, the TFBG spectra are analysed during the resin infusion stage. The results during the curing stage of the composite plate are addressed in section 4.2.

TFBG detection during resin infusion stage
During the infusion stage, the TFBG spectra can be used to monitor the resin flow front arrival. At this stage, the TFBG spectra were acquired every second, while the TC was recorded every 3 s. When the resin starts to wet the sensors, the depth between the upper and lower cladding peaks decreases uniformly until they reach a stable condition when the TFBG is fully immersed. In this last condition, the normalised envelope area returns the resin RI using the fitting correlation function obtained from the calibration. A detailed analysis of this behaviour can be found in [35]. Figure 4 shows the TFBG spectra before and after the resin flows in the thin composite. In this case, the required time for the resin to contact the sensor from the start of the infusion was $50 s whilst to have a stable spectrum at the relative area, $9 s were needed. The latter can be considered as the interval needed for the resin flow front to fully cover the TFBG after reaching it. However, this time should not be confused with the absolute speed in a single direction of the composite. In fact, due to the presence of the flow media on top, the flow front propagation is non uniform through the thickness as its permeation from the flow media is superposed to the in-plane flow. Moreover, small oscillations can be present in the spectrum due to pores migration flowing along the OF surface where the TFBG is present and influencing its coupling mode system locally. In this context, the TFBG spectral signals may be used to locally identify possible defects deriving from poor or incomplete wetting. This can occur on-line since the operations to calculate the envelope area are usually fast enough to be completed within the minimum refresh time of the device used to interrogate the TFBG sensors. Furthermore, as the time interval to obtain the spectrum of the single TFBG from dry to fully wet condition, can be calculated with an accuracy of 0.33 s (depending on the interrogator device), these sensors may be used to investigate the local permeability of a fibre reinforcement layer along the embedding direction.

TFBG measurements during curing stage
The composite thermomechanical state can be obtained with the TFBG as described in section 2. From this procedure, De and DT can be obtained from the moment in which the oven was switchedon (i.e. 96 min). At this point, the strain variation is  assumed to be zero and the temperature is $21 C.
Here, the TFBG acquisition frequency is once per minute, while the TC measurements are every 3 s. Figure 5 shows the temperature (dashed red line) and strain (black continuous line) profiles measured by the embedded TFBG sensor. Though the TFBG measurements performance are constrained to its thermal resolution (7.8 C), here the temperature trends measured by the TC and TFBG are similar. Nevertheless, the TFBG thermal resolution generates an average temperature difference between the TC and TFBG temperature trends (red lines in Figure 5) of $0.78 C, while the maximum difference is $1.7 C at 198 min. As a consequence, since strains are calculated in isothermal conditions when the TFBG measures the maximum temperature, a mean and maximum strain deviation of $5 le and $12 le respectively, can be calculated through temperature differences. Considering the De and DT variations involved in Figure 5, the previous deviations can be considered negligible. The strain oscillations detected by the TFBG along the entire interval between 200 and 540 min, are caused by temperature fluctuations. During the cooling down step, the TFBG measurements were stopped when the temperature reached 28.8 C as lower temperatures were not detectable due to the TFBG thermal calibration. At this temperature the strain measured by the TFBG is around À157 le. However, the lowest detected temperature was 23.7 C. The additional 5.1 C correspond to about À23 le hence, the maximum De expected is around À180 le. The compressive strain measured is a combination of compressive strains due to matrix related shrinkage and contraction in the cool down phase due to coefficient of thermal expansion (CTE) [15].
The resin RI can be evaluated through the embedded TFBG at any moment of the manufacturing process, simultaneously with the strain and temperature measurements, after being immersed by the resin flow. In Figure 6, the envelope area and the resin RI evolution is shown with the TC temperature profile where both parameters follow the same trend. Both the curves were smoothed with fine averaging to remove the greater part of the oscillations due to possible TFBG signal fluctuations caused by high-heat transfer, transverse strains, OF bending, power detection accuracy and background noise. As the RI is obtained from the envelope area through the correlation fitting function, their trend is very similar. This means that both the envelope area and RI can provide information about the resin curing state as is demonstrated in Figure 7 for the RI variation trend. Hence, in an on-line monitoring application of the resin curing process, the envelope area trend provides enough information to determine the resin state without the RI calculation. However, since the RI has a physical meaning, it is preferred here to base the following discussion on its correlation with the resin cure degree.
In Figure 7, a and the resin RI curves are compared and three ranges were identified. In the first, the sensor measures the resin's RI monotone increasing due to the crosslinking of the polymeric chains. Though the initial RI variation is severe, the curve slope becomes milder quickly reflecting the slow resin crosslinking reaction occurring at room temperature. In the second range, the RI trend changes when the ramp-up is performed by switching-on the oven with the programmed temperature profile. The RI value detected from the TFBG at the start of the oven ignition is $1.552. This value deviates slightly from the one reported in the manufacturer data sheet for three main reasons: different reference temperature (here, $21 C), the resin was previously mixed with its designated hardener which influences the overall RI, and the resin has already undergone part of the cure stage at room temperature. Switching-on the oven, the RI tends to flatten, and then decreases as the overall expansion due to the CTE of the mostly uncured resin dominates the shrinkage effect due to the crosslinking. Therefore, the RI curve reaches a local minimum point, where the third range starts. Here, the RI trend reverses when the degree of cure is about 0.4 which indicates that the resin system might be approaching its gelation point, and starts to increase as shrinkage related effect starts to dominate over the CTE. The competing effects between shrinkage and CTE in composite manufacturing has been discussed and quantified by means of FEA in [15]. The resin's RI converges into a plateau in the last part of the curing stage (observed also in [27,39]), where the great part of the crosslinking reactions (a ¼ 92%) occurred. The oscillations in the signal are possibly caused by the low crosslinking density occurring in this range, whereby the resin RI changes are too small to be detected from the TFBG with respect to the RI variation caused by the small temperature fluctuations. Hence, the resin RI measurements during its curing (or cladding resonance peaks envelope area), can be used to detect the different cure dynamics even without the help of the a curve. Additionally, a correlation between the resin RI and its degree of cure can be establish [40]. Furthermore, since the RI is sensitive to the temperature variations, its measurement can identify also whether the curing occurs at room temperature or in an oven. Finally, the resin's T g is exceeding the curing temperature at about 330 min (Figure 7), leading to a vitrification of the matrix.

Conclusions
In this work, the TFBG sensor has been demonstrated to be able to act as a three-parameter OF sensor in monitoring the VARTM manufacturing process of a glass-fibre/epoxy composites. The TFBG proved to be useful in detecting the time needed by the resin flow front to reach the TFBG and the envelope area can be also used to obtain flow information as resin arrival time, infusion degree and poor resin wetting. The TFBG simultaneous strain-temperature measurements detected during the curing in the oven, indicated a relevant development of compressive strains during cooldown. The measured temperature matches well the one detected by the TC. However, the two profiles deviated due to the TFBG thermal resolution, which influences also the measured strains. This limitation can be easily overcome using a FBG interrogator with a higher swR. The maximum strain deviation calculated corresponds to 2.7% of the average strain for a short time interval of the curing step, and this value can be considered low. At the same time, The TFBG spectra provide the resin RI variation starting from the infusion step. The comparison of the RI trend with the resin degree of cure obtained from its cure kinetics, showed that the RI measurements  can detect the resin cure state throughout the process. The monitoring technique can in future be applied to carbon fibre composites made with Liquid Composite Moulding processes or in an autoclave, and in pre-pregs composites. Furthermore, the TFBGs may be suitable for the composite industry as they can improve the quality and health control of the products by providing information during the manufacturing process and in service. TFBG measurements can be also performed on thicker composites, and can be potentially obtained in real-time, and for all the three dimensions of the composite by embedding more TFBGs.
To conclude, a single minimal intrusive TFBG sensor was demonstrated as three-parameter OF sensor embedded in a composite, to monitor the thermomechanical and cure state of the composite during the several steps of its manufacturing process after performing a preliminary calibration of the sensor. This allow to improve the monitoring and sensing technology state of art and raise the concept of structural health monitoring of a product.