Mechanical Characterization of Carbon Fibres Recycled by Steam Thermolysis: A Statistical Approach

The recent development of technologies for recycling carbon fibre reinforced plastics (CFRPs) leads to the need to evaluate the 
mechanical response of recycled carbon fibres. As these fibres are likely to be degraded during the recycling treatment, it is very 
important to determine their tensile residual properties so as to evaluate their ability as reinforcement for new composite 
materials. Carbon fibres reclaimed by a steam-thermal treatment applied to degrade the epoxy resin matrix of a CFRP are here 
analysed. Two conditions were chosen so as to reach two degradation efficiency levels of the steam thermolysis. Several carbon 
fibre samples were selected for mechanical testing carried out either on single filaments using single fibre tensile tests or on fibre 
tows using bundle tensile tests. It is shown that the single fibre tensile test leads to a wide variability of statistical parameters 
derived from the analysis. Bundle tensile tests results were able to indicate that fibre strength of recycled carbon fibre is similar to 
corresponding as-received carbon fibres thanks to a statistically relevant database. Wide number of tested filaments enabled 
indeed to obtain low scatters.


Introduction
Carbon fibre reinforced plastics (CFRPs) have been widely used these last years in many industrial, sportive, and transport applications, especially for their low weight and high strength.e global carbon fibre market is expected to reach high annual growth rates until the next few years.Although the current global demand for carbon fibre, 82,400 tons per year, is lower than expected in last year's market reports [1,2], it is still expected to grow at a minimal annual rate of 9.0%.Global demand in carbon fibres is expected to reach 116,000 tons per year in 2021 for the less optimistic scenario [3] whereas other projections estimate a 150,200 tons demand [4,5].In addition, the carbon fibre reinforced composites market obviously shows very similar growth trends.While in 2013 the global demand for this kind of material was 72,000 tons, recent reports expect this market to reach a 191,000 tons demand by 2022 [6].e high growth perspectives of wind turbines and aerospace industries can mainly explain the intensification in using CFRP as their recent introduction in the automotive industry.is dramatic increase in using carbon fibre means that the quantity of generated waste will also rise significantly, either as an offcut or as an end-of-life composite product.us, it appears to be critical to develop suitable composite recycling technologies that could offer interesting environmental and economic perspectives.If the environmental and social responsibilities are the first arguments for such development efforts, market economics is still a key factor.Considering that the carbon fibre market's potential is clearly affected by the high price of carbon fibre, although its production capacity is nowadays growing, there is a huge opportunity for future or existing recycled carbon fibre producers and processers to answer new needs.
Although landfilling is currently the main option to manage CFRP wastes, the high added value of carbon fibre associated with a restrictive European legislation [7,8] has driven researchers and engineers to look for new recycling technologies, especially as life cycle analysis already showed that the environmental benefit is much higher for a recycling scenario than for a classical incineration or landfilling [9][10][11].
ese last years, the main studied approach has been to degrade the organic matrix to leave clean the carbon fibres, these ones being valorised as reinforcement in secondgeneration composite products.Various technologies focused much effort in this way: solvolysis [12], pyrolysis [13], and steam thermolysis [14].
Solvolysis is a chemical process based on the organic matrix depolymerisation by means of a solvent.Most of the time, near-or supercritical conditions are required to obtain the best results and avoid the use of aggressive chemical solvents that make the treatment more complex.Methanol [15], propanol [16][17][18], water [19,20], or even a mixture of water and ethanol [21] in supercritical conditions were successfully used: the removal of an epoxy matrix can reach 100% without loss of tensile strength of reclaimed carbon fibres.Although more investigation efforts have been made in these methods, there is still no example of an industrial scale launch of this technology applied to the CFRP recycling: supercritical reactors are expensive as they have to be designed for high temperatures, high pressure, and a corrosive environment.
Pyrolysis is based on the organic matrix thermal degradation.It has been the most studied thermal process [22][23][24], and some variations can be found as the microwave heating pyrolysis [25,26].Depending on the matrix nature, and the considered variation, the efficiency of such a treatment is variable: from 80% to 99% of eliminated resin.Reclaimed carbon fibre tensile strength can be degraded due to the presence of char on the fibre surface that needs to be eliminated by an air posttreatment.However, in spite of lower results than what can be obtained in solvolysis, pyrolysis is a cost-efficient technology well suited to the relatively undeveloped composites recycling market.ese research efforts have even been commercially applied by European companies such as ELG Carbon Fibre (United Kingdom), Karborek (Italia), Reciclalia (Spain), or CFK Valley Stade Recycling (Germany) and American ones such as Adherent Technologies Inc or Carbon Conversions.
Finally, steam thermolysis is a thermochemical process using superheated steam at environmental pressure for degrading organic materials.It is a cost-efficient technology as no energy-consuming posttreatment of reclaimed fibres is needed, nor high pressure environment requiring discontinuous working flow and expensive reactors.It has been applied to the material recovery of circuit boards [27], to the degradation of polyimide [28], or to the production of oil from biomass [29].Only few studies focused on the steamthermolysis process applied to the recovery of carbon fibre from CFRP wastes [10,[30][31][32][33]. Steam thermolysis enables to efficiently degrade the organic matrix of the CFRP waste, which makes this technology a serious alternative.e aim here is to evaluate the efficiency of this technology by proposing a true mechanical characterization of the reclaimed carbon fibres considering two techniques: single fibre tensile test (SFTT) and bundle tensile test (BTT).Using the widely used SFTT technique, some inherent variability sources of the tensile strength determination can appear as the specimen selection, the damage of the fibres during the sampling operation, and the difficulty in getting a perfect alignment of the fibre with the tensile machine.Hence, the bundle tensile test can become an alternative for the tensile strength determination of recycled carbon fibres, as it has been successfully used to study virgin glass, ceramic, and carbon fibres.

Bundle Model.
e theoretical model of dry bundle of fibres considers a discrete set of N parallel fibres with statistically distributed strength.When the bundle is loaded, fibres' mechanical behaviour is linear elastic until their failure at the applied stress σ i , i � 1, . . ., N. When a fibre breaks in the bundle, the supplementary load that was carried by the broken fibre is equally distributed.Two distribution cases can be differentiated.
e global load sharing (GLS) considers that the supplementary load is equally distributed among the survival fibres whereas the local load sharing (LLS) considers that the supplementary load is equally distributed among the neighbouring fibres.e first case is here considered but needs to fit assumptions, called Coleman's conditions: fibre length must be constant within the bundle, stress-strain relationship follows Hooke's law until failure, the released load at a fibre break is uniformly distributed among the surviving fibres, and no external phenomena should lead to a premature fracture of fibres.As a consequence, any friction phenomena between fibres within the bundle must be avoided as it would lead to a catastrophic fracture of the whole bundle.Specific cares are taken to avoid this effect.

Statistical Distribution of Fibre Strength.
Fracture of carbon fibres is likely to be caused by flaws within the gauge length.Flaws are randomly distributed and show a high heterogeneity in size, location, and severity.en, a wide variation in failure load is expected and the ultimate tensile strengths measured on specimens have a statistical distribution.Weibull analysis is a well-known method typically used for fracture statistics for brittle materials.For a single gauge length and uniform uniaxial tensile stresses, the Weibull equation of failure probability is given by: where V is the stressed volume and V 0 a reference volume, en, the statistical parameters are obtained by fitting (1) to the Weibull plot.
However, the validity of the normal distribution to describe the distribution of strengths has already been shown [34] and the statistical parameters that were derived from are considered to provide a better fit to the data.Besides, it was demonstrated that the statistical parameters derived from a Weibull distribution showed a wide variability due to the construction of Weibull plots using an estimator and the sample size generally too small that does not enable to take into account the natural variability of material properties.erefore, a statistically relevant database and normal distribution are used for the analysis of failure data.Equations of probability density function f(ε) and normal distribution P N (E ≤ ε) are as follows: with ε the strain, μ the mean of strain, and S its standard deviation.

Bundle Behaviour.
Assuming that the applied load is uniformly distributed among the surviving fibres in the tow and that fibres have a linear load-strain relationship up to breakage, the force-strain relation during a tensile test is given by [35]: where N 0 is the number of initially loaded fibres, A f is the cross-sectional area of each of the fibres, E f is their Young's modulus, ε is the applied strain, and P(ε) is the probability of failure of a fibre at a strain ε, given by (2).

Composite Manufacturing.
Composite samples were made by liquid resin infusion.A low-viscosity bicomponent system was used: a Sicomin SR1710 Infusion epoxy resin mixed with a Sicomin SD8822 hardener.A twenty hours at room temperature plus sixteen hours at 60 °C polymerisation cycle were applied before removing the system from the mould.Details of procedure can be found in a work of Balea et al. [36].Carbon reinforcement was a carbon twill 2 × 2 (Hexcel 46285 U1200) made from AS4C carbon fibres.Sixteen 400 × 400 mm plies were stacked so as to obtain an approximately 800 grams plate, for a 4 mm final thickness.e average fibre mass fraction was 66% corresponding to a fibre volume fraction of 55.5%.ese plates were cut by the mean of a circular saw in order to get 50 × 120 mm samples able to be used in the steam-thermolysis reactor.

Recycling Carbon Fibres.
e recycling was conducted in a bench-scale reactor as shown in Figure 1.Previous in-house produced composite samples are treated by steam thermolysis so as to reclaim carbon fibres.e thermochemical process uses superheated steam at atmospheric pressure in order to degrade the organic matrix of the composite.
A removable crucible was made from a stainless-steel fabric (own design, 1000 mL). is crucible was coupled with a thermogravimetric analyser and placed within the heating zone.
e experimental reactor is provided with an easy opening chamber located on the top of the apparatus.Once the experimental parameters reached the desired level, the chamber that contains scrap composites samples (100 g) is opened so as to let them fall into the reactor.After epoxy resin was decomposed, and once the system cooled down, recycled carbon fibres were collected from the reactor.No cleaning of the surface is required before their use.ree categories of products are actually collected: a solid fraction that is constituted of recycled carbon fibres, a permanent gaseous fraction principally constituted of methane and carbon monoxides, and a last condensable gaseous fraction that is constituted of pyridines, benzene, and phenols [31].
A unit made up of a steam generator and a nitrogen input manages atmosphere control.e experimental device is designed to operate in a wide range of conditions: temperature from 100 to 1000 °C, steam flow rate from 0 to 1000 g•h −1 , and nitrogen flow rate from 0 to 20 L•min −1 .Experimental conditions and reclaimed samples are described in Table 1.Experiments were carried out under atmospheric pressure for two hours at two temperatures: 400 °C and 500 °C.In both cases, the nitrogen flow rate was set to 10.8 L•min −1 whereas the steam flow rate was 90 g•h −1 .Reclaimed carbon fibres from these treatments are, respectively, named RF400 and RF500.

Fibre Morphology.
Yields of eliminated resin were measured by dissolution of remaining resin with hot sulphuric acid according to the French standard NF EN 2564.
Environmental scanning electron microscopy (ESEM) was used to observe surface texture and morphology of the fibres as well as visual signs of residual resin impurities.Fibre bundles of each sample were randomly selected and mounted on an adhesive carbon layer stuck onto an aluminium stub.As carbon fibre is conductive, no other specific preparation was needed.e acceleration voltage was 20 kV.Diameters of the fibres were also measured using image analysis with ImageJ software.For each sample, an average diameter was determined by measuring a population of 200 fibres from an image database obtained during the corresponding ESEM analysis.

Single Fibre Tensile Test.
e most common technique, the single fibre tensile test (SFTT), measures the strength of individual fibres.By measuring many fibres, a wide population can be formed and used for stress analysis.is test was employed to determine the tensile strength of the three fibre types of the study.Method is based on international standards ISO 11566 [37].A single filament is bonded to Advances in Materials Science and Engineering a paper window with cyanoacrylate Loctite 409.en, the specimen is inserted into a tensile rig equipped with a 5 N load cell.e carbon bre has to be carefully aligned with the tensile testing machine axis.Each side of the paper window was cut before testing.e gauge length was 25 mm.e crosshead speed was set to 0.1 mm/min.Carbon bre specimens were loaded at room temperature until failure, and the force displacement curve was recorded.At least 40 laments were tested for each bre type, that is, VF, RF400, and RF500.

Bundle Tensile Test.
Mechanical tests were also carried out on bre bundles using bundle tensile tests (BTTs) so as to quantify the tensile strength of the recycled carbon bres.It is based on the random and individual bre failure within the bundle.erefore, statistics laws are used for analysis.
is statistical data approach enables to take into consideration a wide single lament population.
One of the di culties is the measurement of a reliable bundle strain.An extensometer is placed on heat shrink tubes previously threaded on each tip of the bundle, as it is shown in Figure 2, to de ne the gauge length.Each tip is impregnated with Araldite 2015 resin and then polymerised at 70 °C for one hour.Impregnated tips are then inserted in metallic tubes and lled again with Araldite 2015 resin and polymerised at 70 °C for one hour.Metallic tubes enable a regular clamping by tensile grips.During any of these preparation steps, a speci c care must be taken to avoid any handling of the bre bundle within the gauge length.Before loading, the sample is lubricated by petroleum wetting, avoiding premature rupture due to friction phenomena between bres within the bundle.
is meticulous experimental procedure is also described in [38,39].e tensile tests were performed using a pneumatic testing machine with a 2 kN cell.ey were carried out at 4 Advances in Materials Science and Engineering room temperature under constant displacement rate of 0.06 mm/min on specimens prepared according to the previous procedure with a gauge length of 60 mm.Carbon bre bundles were loaded until failure, and the load displacement curve was recorded.For each bre type, about 3000 laments were tested in each tow.For RF500 bre, 3 tows were tested so as to make sure measurements are repeatable.

Methods of Failure Data Analysis.
For single bre analysis, the means of ultimate strengths are known by collecting individual data.Both normal and Weibull distributions are used.Weibull plots are constructed using an empirical distribution function P j (j − 0.5)/N with N the sample size and j the specimen number.
For bundle analysis, the mean of bre tensile strength and its standard deviation are obtained by tting an analytical curve based on (4) to the experimental data.Firstly, the load-strain curve of the bundle is determined by the tensile test as described in 3.4.2.en, the initial slope of the linear part of the analytical curve is tted to the experimental one.Equation (3) can also be written as: where R 0 is the initial slope of the (F − ε) curve.Finally, by tting the nonlinear parts of experimental and analytical curves, the mean of strains to failure μ and its standard deviation S are determined.Assuming Young's modulus of each type of bre is constant, the mean of ultimate tensile strength of each type of bre and its standard deviation can be determined.

E ciency of Steam-ermal Treatments.
e temperature is an important parameter on the degradation kinetic and thus on the e ciency of the treatment.Measurements of yields of eliminated resin are shown in Table 2.A 400 °C thermolysis did not enable the elimination of all the resin of the composite (yield of eliminated resin reached 95% in mass) whereas the 500 °C treatment was more e ective and enabled to degrade all the epoxy resin (yield of eliminated resin is higher than 99% in mass).
Figure 3 shows an ESEM image of the virgin bre VF and recycled carbon bre RF500.Examination of images of several bres from di erent batches clearly showed no visible alteration of the surface topography due to steam thermolysis.Similar regular and clean surfaces are observed, indicating the e ciency of the treatment that removed the most part of the resin of the composite material.Recycled RF400 bres are shown in Figure 4.A few small particles can be seen and are attributed to resin residues that stuck on the surface.
e 400 °C steam-thermal treatment left little quantities of residual resin on a smooth and regular surface (5% by mass of residual resin).
e particles have a size ranging from 2 to 20 micrometres, avoiding individual bres to be properly separated.ese observations obviously show the importance of temperature on the degradation kinetic.
e mean diameters were calculated as 7.1, 6.9, and 6.9 μm, respectively, for VF, RF400, and RF500 bres (Table 1). is is in good agreement with the value of 6.9 μm provided by the manufacturer [40].It may be inferred from the similarity of the mean and standard deviation values of the bres, with visual evidence from the ESEM, that there was no alternation to the bre morphology.

Mechanical Properties
4.2.1.Single Fibre Mechanical Analysis.Two statistical parameters are deduced from the analysis: the mean of strength μ and its standard deviation S. From the experimental data, the 95% con dence interval of mean value is also established as it is often used as an indicator of the precision of an estimate derived from an analysis.For a sample size N 40, μ the sample mean, and S the standard deviation, the 95% con dence interval of mean value (I c ) is given by: Statistical parameters of normal distribution of strength deduced from SFTT are reported in Table 3. e average tensile strength of RF500 bre is slightly di erent from that of the corresponding virgin bre VF.A 4% decrease was observed.However, the result shows a high degree of variability with a standard deviation of about 540 MPa for a tensile strength of 3610 MPa.Looking at the frame given by these 95% con dence intervals, it appears to be di cult to obtain reliable results.Indeed, there is no statistically representative di erence between the two samples.us, it could be premature to a rm that the tensile strength loss is really signi cant or not, although it could be negligible regarding the low decrease of only 4%.Nevertheless, it can be stated that reclaimed RF400 bre showed substantial strength degradation relatively to the virgin bre.Even taking into account the high degree of variability of measurements, tensile strength loss of RF400 bre is likely to be

Fibres
Yield of eliminated resin (%) Mean diameter (standard deviation) (μm) VF -7.1 (0.7) RF400 95 6.9 (0.7) RF500 >99 6.9 (0.7) Advances in Materials Science and Engineering signi cant. is could be explained by the presence of residual resin on the bre surface that could act as stress concentrators leading to a premature failure of the bre.

Fibre bundle
Figure 5 shows mean stress-displacement curves obtained from single bre testing and con dence interval on the mean value of tensile strength and displacement at failure.As it can be seen that average value of failure strain is lower than that of virgin and RF500 bres, it con rms that the single bre fails before reaching its maximum stress level.
Weibull diagrams derived from this analysis are also presented in Figure 6, as they are a usual approach for describing failure behaviour of brittle materials.ey are compared to log-log graphs of normal distributions of stress to failure for each sample.A good agreement between both distributions is obtained for RF400 bre.However, a clear discrepancy can be noticed at the low failure probabilities for VF and RF500 bres.e RF500 Weibull plot suggests the presence of two domains re ecting two distinct failure modes for this bre whereas it does not seem to be the case in Figure 7 showing the probability density function of this bre and the associated experimental points.Indeed, at the lower stress values, experimental data do not clearly show two distinct populations.Many other reasons can be advanced to explain discrepancies on Weibull plots and high scatters observed on SFTT results: the use of an empirical estimator, the selection and damage of the bres during the operation of sampling, or the low sample size leading to a low representativity in the case of brittle materials [35,41].A wide distribution in aw size is inevitable considering the selection of test specimens.While variability cannot be avoided until a relevant database is used for failure analysis,    6 Advances in Materials Science and Engineering there is no means to evaluate the validity of this selection and so to validate the full strength retention of recycled carbon bres.

Bundle Mechanical Analysis.
Figure 8 shows typical load-strain curves obtained from bundle mechanical analysis.It is easy to see that a good agreement is obtained between experiment and model: experimental curve and normal distribution-based curve were well tted.e load decrease beyond maximum ts well with that obtained experimentally.Maximum load also depends on the number of laments in each tested tow and is consequently not always the same for a same sample.Exact number of laments is determined from the initial slope of the load-strain curve.Statistical parameters of normal distribution of strength extracted from analysis of bundle tensile tests are listed in Table 4. e RF400 sample shows the lowest mean strength of 3657 MPa whereas the VF sample and RF500 sample show a quite similar mean strength of about 3860 MPa.Variability of the results rst seems to be as high as that obtained for single lament tensile tests.However, as the tested population is very wide (Table 4), con dence intervals are lower than those obtained by single bre tensile tests analysis.is enables to get more con dence on the precision of the estimate.us, no signi cant di erence can be noticed between RF500 bres and VF bres, indicating that steam   Advances in Materials Science and Engineering thermolysis enables to retain tensile strength of the reclaimed carbon bre RF500.It shows that steam-thermal process has only little e ect on carbon bres' mechanical properties, although the recycling was performed at 500 °C.On the contrary, a decrease of almost 200 MPa a ected RF400 bres.Resin nodules on the surface of RF400 bres could be a contribution to the increase of friction between laments during the tensile test.Friction in BTT leads to a premature failure of the neighbouring bres in the tow [42] contributing to a steep load decrease beyond the highest measured load.However, the curve seems to be smooth and does not show any signs of bre friction especially as analytical curve ts very well with experimental data.Indeed, analytical data are based on bundle theory that considers that bres are independent.As in single bre tensile tests, the RF400 tensile strength decrease rather suggests that resin nodules could act as stress concentrators that lead to premature failure of single laments in the tow.

Discussion
Figure 9 shows that tensile strength values obtained from bundle tensile tests are in good agreement with those obtained by SFTT although a slight di erence can be noticed.However, when taking into account the larger gauge length of tows (60 mm instead of 25 mm for single bres), tensile strengths should be much lower than those obtained by single bre testing.Indeed, carbon bre tensile strength is dependent on its length [43].More generally, the geometry of carbon bre plays an important role in its strength [44,45].e higher is its length, the larger is the number of aws and thus the probability to nd a severe aw that leads to fracture of the bre.Just as the bre diameter that is related to the bre volume that increases the probability to nd a severe aw. at is why higher gauge lengths should lead to lower strengths.For these reasons, experimental data must be statistically analysed.Taking into account a statistically signi cant sample size, bundle tensile test enables to overcome uncertainties that usually a ect single bre tensile tests analysis as the specimen selection, the damage of bres during sampling, or the sample size. is is why it is reasonable to consider that di erences observed between bundle and single lament testing results con rm that variability in single bre testing is high and inevitable and that results that are derived from could have likely been higher or lower if experiments were repeated.On this point, repeatability of the bundle tensile test was investigated on RF500 bre.Table 5 shows that only very slight di erence can be seen between average tensile strengths, lower than 1%.Most of all, the 95% con dence interval is quite the same from one experiment to another.It only changes a little on account of the change in the number of laments in the tow that directly has a consequence on this interval value.is test is a repeatable way to generate large databases in a reasonable amount of time in order to take into account the heterogeneity of carbon bres that naturally leads to high scatter in tensile strength results [45] if only a small population is considered.In this study, bundle tensile test enabled to characterize mechanical properties of recycled carbon bres with a good precision.However, the BTT needs very meticulous preparation and advanced statistics to be implemented.At the contrary, SFTT only needs an easy-tofollow procedure and data can be readily analysed.Also, geometry of most of CFRP recycling bench-scale reactors does not enable to reclaim recycled carbon bre lengths higher than 50 mm, which makes it di cult to determine their tensile properties by BTT.

Conclusion
Steam-thermal process was used in a bench-scale reactor to recycle carbon bre from epoxy resin/carbon bre composites.Properties of the recycled carbon bres were characterized using ESEM, single bre tensile test, and bundle tensile test.Carbon bres were properly separated from polymer matrix during the treatment, showing that a steam-thermal treatment is e cient and enables to reach high resin elimination levels.Two techniques were used for mechanical characterization of recycled carbon fibres.Single fibre tensile test did not allow to validate the full strength retention of recycled carbon fibres due to unavoidable high variability of the results.Bundle tensile tests were able to show that a 500 °C steam-thermal treatment enables to leave clean carbon fibres with no degradation of tensile properties.us, advantages of bundle tensile tests were highlighted: no selection of specimen and a relevant database that enabled to get reliable results.erefore, steam thermolysis not only degrades the whole part of matrix resin of the composite so as to leave perfectly clean carbon fibres but also enables to recover fibres with full tensile strength retention.Valorisation of these fibres could be possible.Properties of composites made from recycled carbon fibres should be measured so as to reveal the viability of such a process to produce recycled carbon fibres from epoxy-based composite materials.Recent works considered different ways to reintroduce them in structural components although potential applications are critical to identify [46][47][48].
e recycling of CFRP is acquiring a considerable importance due to legislative context, and the need to find sustainable solutions for waste processing.Steam-thermal process also demonstrated its abilities in this field.

Figure 6 :
Figure 6: Comparison of Weibull plot and normal distribution of stress to failure of RF400, RF500, and VF bres (log-log plots).

Figure 7 :Figure 8 :
Figure 7: Normal probability density function of RF500 bre and frequency histogram of failure events.

Figure 9 :
Figure 9: Mean strengths and their 95% con dence intervals obtained from single bre tensile tests and bundle tensile tests.

Table 1 :
Samples of the study and associated steam-thermolysis experimental conditions.

Table 2 :
Studied samples and associated steam-thermolysis experimental conditions.

Table 3 :
Statistical parameters of normal distributions obtained from single bre tensile tests analysis and related 95% con dence intervals.

Table 5 :
Statistical parameters of normal distributions obtained from three RF500 bundle tensile tests analysis and related 95% con dence intervals.

Table 4 :
Statistical parameters of normal distributions obtained from bundle tensile tests analysis and related 95% con dence intervals.