- Email: [email protected]
Surface temperature of carbon composite samples during thermal degradation
International Journal of Thermal Sciences 112 (2017) 427e438
Contents lists available at ScienceDirect
International Journal of Thermal Sciences journal homepage: www.elsevier.com/locate/ijts
Surface temperature of carbon composite samples during thermal degradation Zoubir Acem a, b, Damien Brissinger a, b, Anthony Collin a, b, Gilles Parent a, b, Pascal Boulet a, b, *, Thi Hay Yen Quach c, Benjamin Batiot c, Franck Richard c, Thomas Rogaume c a b c
Universit e de Lorraine, LEMTA, UMR 7563, Vandoeuvre-l es-Nancy, F-54500, France CNRS, LEMTA, UMR 7563, Vandoeuvre-l es-Nancy, F-54500, France PPRIME Institute, CNRS, Universit e de Poitiers, ISAE ENSMA, BP 40109, F86961, Futuroscope cedex, France
a r t i c l e i n f o
a b s t r a c t
Article history: Received 15 September 2016 Received in revised form 11 October 2016 Accepted 7 November 2016 Available online 13 November 2016
An infrared camera and a Fourier transform infrared spectrometer were used simultaneously to observe the radiation from the surface of carbon composite samples during thermal degradation experiments. Surface temperatures were estimated from radiation measurements conducted during cone calorimeter tests. Infrared spectra and images were post-processed involving a subtraction method between successive images, to withdraw the high incident flux from the cone calorimeter partly reflected by the sample. The surface intensities were first evaluated as a function of time. Then, an identification step was implemented linking the rise in intensity between two time steps with the increase in sample temperature. Corrections were introduced for the reflection of the incident radiation from the heater and for the true emissivity of the material. Both experimental devices e camera and spectrometer e showed temperature evolutions in a good agreement. The sharp temperature increase of the sample and the auto-ignition of the pyrolysis gases on the surface were observed and evaluated for various incident fluxes from the cone between 20 and 65 kW/m2. Measurements carried out with a short time step down to 0.2 s further allowed an evaluation of the supplementary flux due to the auto-ignition with a peak up to 20 kW/m2 for an incident flux of 35 kW/m2 on a carbon composite sample with surface 50 mm 300 mm and thickness 5 mm. © 2016 Elsevier Masson SAS. All rights reserved.
Keywords: Pyrolysis Cone calorimeter Infrared imaging Radiative transfer Temperature identification Heat flux Flame radiation
1. Introduction The study of the thermal degradation of materials provides some of the input parameters required for the prediction of their ignition and combustion. Standard characterization involves samples submitted to given external radiative fluxes, usually produced by a high temperature lamp in the Fire Propagation Apparatus [1], or a coil when using a cone calorimeter [2]. A calibration with a heat flux gauge ensures that the incident flux is well controlled and data are registered for the mass loss experienced by the sample as a function of time. This problem involves a complex coupling of heat and mass transfer. Therefore, temperature data are also required for
de Lorraine, LEMTA, UMR 7563, Vandoeuvre* Corresponding author. Universite s-Nancy, F-54500, France. le E-mail address: [email protected] (P. Boulet). http://dx.doi.org/10.1016/j.ijthermalsci.2016.11.007 1290-0729/© 2016 Elsevier Masson SAS. All rights reserved.
a better understanding of the phenomena and for validation purpose. These data are then compared to results obtained with degradation and combustion models, which are needed in fire simulations for prescribing the mass loss rate and for predicting the heat release or the heat transfer inside the materials. The knowledge of the sample surface temperature is of particular interest for the description of the heat transfer inside the sample and for the definition of reliable boundary conditions. Thermocouples may be used to measure this surface temperature, but the confidence in the acquired data is questionable as direct radiation to the thermocouple may occur from the source or from the flame developing on the surface after sufficient heating. In addition, this is a high temperature measurement possibly disturbed by a non perfect contact (a specific problem which would be encountered in the present study devoted to a carbon composite material with a low conductivity) or also because of soot deposit. To avoid these problems,
428
Z. Acem et al. / International Journal of Thermal Sciences 112 (2017) 427e438
measurements inside the sample can be done with embedded thermocouples, extrapolating the values to the surface. Stoliarov et al. [3] presented temperature data measured with thermocouples embedded as close as possible to the surface (1.5 mm below it) in order to characterize the top layer temperature. Pizzo et al. [4] on PMMA and Hidalgo et al. [5] on a carbon composite sample from the same series as the one studied here, to name some recent studies, presented temperature measurements with embedded thermocouples at various positions inside the materials. Such measurements are usually used as validation data for models. Temperature identification could be also sought based on the observation of the back surface, which is protected from direct irradiation. However, other difficulties are raised owing to the nonstationary conditions and to the problem complexity (including the required model for heat transfer, through a complex medium with varying properties and the difficulty to do the measurement itself). Li et al. carried out such measurements at the back face and examined the accuracy of the surface temperature [6,7]. Data were then used for pyrolysis model validation purpose. The present work was conducted to test another method based on the observation of the flux emitted by the front surface with a Fourier Transform InfraRed (FTIR) spectrometer and an infrared (IR) camera. Our objectives were twofold: (i) to observe the radiation sent by the irradiated surface during a cone calorimeter experiment in order to follow the temperature increase and to check for the temperature homogeneity on the surface; (ii) to develop a reliable method to identify the temperature despite the difficulty raised by the experimental setup (limited accessibility and strong background irradiation because of the cone itself). IR thermography was already used in the past to observe solid surfaces through a flame. For example, De Vries et al. [8] evaluated surface temperatures in rack-storage fires with a long wave camera (expecting limited perturbations by smoke and particles). They observed perturbations in the measurement because of the flame, evaluating the related uncertainty in the measured temperature to approximately 50 K (in a case of a large scale fire). Melendez et al. [9] also measured surface temperature in the case of a plate heated by a burner flame. They used a combination of an optical filter and an image processing involving the flame image subtraction to withdraw the perturbation by the flame. Similar ideas were used here, but in different conditions because the heat source was not a burner and the flame ignition or extinction could not be controlled. The FTIR spectrometer and the IR camera were already used by present authors on several applications related to fire [10e13], for the characterization of sources or flames in particular. Intensities from flames and radiative sources were especially measured in various situations. Based on past experiences [14,15], spectra can be analyzed to identify characteristic temperatures in some wavenumber ranges. Filters can be also added to the camera in order to limit the perturbation of the signal by the flame and to isolate the surface radiation, which is then analyzed to identify its temperature. These methods were used here when registering the radiation from the sample surface. For the present study, composite carbon samples were degraded under a cone calorimeter, varying the incident heat flux between 20 and 65 kW/m2. The surface emission of the samples was simultaneously registered with the spectrometer and the camera. Then, several identification methods were tested in order to obtain the temperature and a satisfactory confidence in its time-varying evolution. In the following sections, the experimental setup will be described first. Then, the radiative transfer model and the identification methods will be presented. Finally, results will be discussed for the evolution of the surface temperature of the carbon composite.
2. Experimental setup 2.1. The composite sample The material studied in the present work is made of carbon fibers in an epoxy matrix, a composite used for high pressure hydrogen storage tanks. Samples were prepared in the frame of the FIRECOMP European project aimed at characterizing the degradation of such materials when submitted to fire sources. The composite is mainly composed of carbon (85% of the total weight according to the available elementary analysis conducted in the frame of the project [16]). Specimens were cut in large composite cylinders, providing quasi-flat samples for degradation tests under cone calorimeter. 2.2. The cone calorimeter combined with infrared metrology A picture of the setup is presented in Fig. 1. In this case, a standard cone calorimeter [1] was used in vertical configuration for the degradation of a 100 mm 100 mm x 5 mm sample. The composite sample was placed in a sample holder specially developed for this application. This one was manufactured into a block of aluminum, and a cover was used in order to maintain the sample in the right position. The opened surface of the cover was 88.4 cm2 in accordance with ISO 5660 [1]. It is important to note that no insulation was used in the back surface of the sample. The heat source was in fact a truncated conical heater and the coil of the cone can be seen in the picture. Radiation was provided from this coil toward the sample surface, leading to its thermal degradation. The picture was taken after auto-ignition of the gas mixing as shown by the flame developing along the sample surface. The observation in the infrared range was done along a path centered on the cone axis through its aperture, using a camera and a spectrometer which are described below. A silicon beam splitter divided the radiation signal toward the infrared camera and the spectrometer for a simultaneous observation of the same area with both devices. A typical test started when the sample was suddenly submitted to the radiation emitted by the cone at t ¼ 0 s (removing a shield used between the heater and the sample to allow the heating of the cone up to the stationary regime while protecting the sample before the experiment starts). Four different incident fluxes were considered: 20, 35, 50 and 65 kW/m2 (values obtained from heat flux gauge measurements carried out at the sample position in standard use of the cone [1]). Experiments were repeated twice to check the repeatability.
Fig. 1. Picture of the setup with the cone calorimeter, the spectrometer and the IR camera.
Z. Acem et al. / International Journal of Thermal Sciences 112 (2017) 427e438
429
2.3. FTIR spectroscopy The spectrometer (Matrix by Bruker) was used in a configuration involving an external source (the sample surface) in the spectral range between 800 and 5000 cm1 in the present study. Background measurements were done in a preliminary step, for an evaluation of the surrounding influence. The contribution of the heater itself was taken into account in the post-processing step based on the analysis of successive spectra analysis as described below. Each spectrum was based on a series of 3e10 scans, as a usual technique used to reduce the signal-to-noise ratio. This number of scans is small (as compared to usual FTIR measurements), but it warranted a quite instantaneous measurement in order to avoid a too strong temperature variation on the surface during the spectrum registration. Figs. 2 and 3 show typical spectra registered during experiments conducted with incident fluxes equal to 35 and 50 kW/m2 respectively. Only a limited number of the numerous acquisitions are plotted here in order to keep a good visibility of the spectrum evolution. Data were converted into equivalent intensities after comparison with calibration data obtained using a high temperature blackbody (M330 by Mikron, at 1500 C). Spectral intensities are plotted as a function of wavenumber for different times. They correspond to the signal received by the spectrometer from the surface, meaning that both proper emission and radiation reflection are involved. During the early step of the experiment, reflection was dominating the intensity, but emission became progressively dominant as the sample was heated. The intensity level increased with time as a consequence of the surface temperature increase. For an incident flux equal to 35 kW/m2 (Fig. 2), the spectrum shapes were found similar, with an increasing level with time. A standard spectral evolution was obtained, as for a usual surface emission, affected by the sample emissivity, with some sharp variations in definite bands due to the gas influence (mainly absorption by atmospheric water vapor and carbon dioxide). No spontaneous ignition was observed, explaining the monotonic increase in the intensity, without emission peak. For an incident flux equal to 50 kW/m2 (Fig. 3), the first increasing phase was followed
Fig. 2. Radiative intensity emitted and reflected by the sample surface at different times when irradiated by a cone calorimeter with a 35 kW/m2 flux.
Fig. 3. Radiative intensity emitted and reflected by the sample surface at different times when irradiated by a cone calorimeter with a 50 kW/m2 flux.
by a sudden increase in the intensity level and a sharp peak observed around 2300 cm1 as a consequence of a flame suddenly developing along the surface (see the legends “phase 1” and “ignition” in the Figure). Then, a further increase was observed (“phase 2”), before a decreasing phase and the emission peak disappearance as the flame extinguished (see the curves at 260 and 600 s, phase 3), most of the sample being degraded. 2.4. IR camera The camera (Orion SC7000 by FLIR) was used with filters in order to isolate images at given wavenumbers. For the present work, the filter at 2564 cm1 was especially used to observe the surface, since it operates in a spectral region where gases (H2O, CO and CO2 especially) do not contribute to the radiation, hence preventing from any perturbation of the image processing by possible combustion gases, but soot can still contribute to the emitted radiation as discussed later. On the contrary, specific filters around 2353 and 3500 cm1 were used in the same time to observe the flame through its combustion products CO2 and H2O, respectively. A filter wheel was used to change the filters automatically with a high frequency rotation speed resulting in a quite simultaneous observation. The time step between two acquisitions was first set to 1 or 3 s during a first test campaign. Then, it was decreased to 0.2 s for a second test campaign in order to catch faster temperature changes accurately. Both sets of results will be commented in the result section. For each acquisition, 12 images were acquired at a frequency of 50 Hz in order to allow successive instantaneous observations or the computation of average intensities and temperatures on a short period. Background radiation was also measured in a preliminary step and the main difficulty raised by the reflection of the incident flux sent by the cone calorimeter was addressed thanks to image subtraction as described in the next section. Camera images were studied in a first step in order to check the homogeneity of the radiation emitted by the surface. Two examples are given in Figs. 4 and 5, for the filter at 2564 cm1. Both images are presented in terms of intensity after a conversion of the camera signal based on reference data obtained in front of our reference
430
Z. Acem et al. / International Journal of Thermal Sciences 112 (2017) 427e438
Fig. 4. IR image at 2564 cm1, converted in spectral intensity (W/(m2 sr cm1)). Incident flux 35 kW/m2, time t ¼ 390 s.
emitted radiation. A calibration test was therefore conducted to evaluate the measurement bias in the temperature evaluation which can be expected because of the soot influence (see section 3). The flame was better emphasized when observing the surface through the filters at 2353 and 3500 cm1. Figs. 6 and 7 show images corresponding to the same case as in Fig. 5 (50 kW/m2, time t ¼ 80 s) but radiation involves here both a contribution from the sample surface and from H2O in the flame (Fig. 6) and CO2 in the flame (Fig. 7). As compared to Fig. 5, supplementary patterns are observed in the images, especially in Fig. 7, meaning that CO2 is the main contributor to the flame radiation. Indeed, H2O was not observed to contribute much to emission, probably because hydrogen content in the composite is weak (3% in weight for hydrogen, as compared to 85% for carbon as above-mentioned, according to tests conducted on the elementary composition of the composite sample [16]). This is in agreement with the spectra in Fig. 3, which showed a sharp peak in the range corresponding to CO2 emission, but no actual emission peak in the ranges associated to H2O (around 1600 and 3700 cm1).
2.5. Surface emissivity
Fig. 5. IR image at 2564 cm1, converted in spectral intensity (W/(m2 sr cm1)). Incident flux 50 kW/m2, time t ¼ 80 s.
blackbody. As for the spectrometer results, the intensity involves both proper emission and reflection. Note that the white circle at the center of the image features the area which was observed by the spectrometer (which received radiation in a limited solid angle and does not see the whole plate as the camera does). All data processing in the following sections (based on camera or spectrometer data) correspond to averages computed for the area inside this circle, with diameter 2.3 cm. On the images, some heterogeneities were seen, which correspond to the carbon fiber orientation and are probably due to the surface roughness affecting reflectivity and emissivity, but the temperature itself was not supposed to vary much along the surface. Fig. 4 is for a moderate incident flux of 35 kW/m2, which did not induce auto-ignition of the sample. On the contrary Fig. 5 was obtained for a higher incident flux (50 kW/ m2), after auto-ignition of the pyrolysis gases. The same scale was kept to emphasize the significant increase in intensity. The image shows some hot points which probably correspond to the flame developing along the surface in this case, despite the use of the filter at 2564 cm1, chosen to avoid the perturbation by the combustion gas, because soot, even in a weak concentration, probably
Results presented in the above sections are straightforward visualizations of the intensity sent by the sample. A further analysis of the true emission required the knowledge of its emissivity. It was evaluated based on transmissivity and reflectivity measurements of samples with a FTIR spectrometer in the spectral range of interest. The spectral absorptivity (equal to the spectral emissivity according to the Kirchhoff's law) was deduced as the complementary part to 1. In order to take into account the possible change in properties as the sample is degrading, measurements were repeated on virgin samples and on similar samples submitted to degradation tests for various incident flux and various durations. After degradation, samples were cooled and characterized with the spectrometer (no in situ measurements during degradation were possible here and it must be assumed that the measured emissivity also holds for the same sample at high temperature, during its exposure in front of the cone calorimeter). The measurements were described in detail in a recent paper [17]. The main results are presented here in terms of spectral absorptivity. Fig. 8 shows three selected curves for the virgin material and for two samples after significant duration of
Fig. 6. IR image at 3500 cm1, converted in spectral intensity. Incident flux 50 kW/m2, time t ¼ 80 s.
Z. Acem et al. / International Journal of Thermal Sciences 112 (2017) 427e438
431
2564 cm1 which was widely used in the followings, the average value (considering all the measurements performed for various fluxes and degradation durations) was finally evaluated to 0.91. For the analysis of the spectra which requires the emissivity over the whole wavenumber range, the exact curves for the properties after degradation were used. A specific sensitivity study was conducted on the influence of the emissivity on the surface temperature identification (see the corresponding section below), showing that the temperature was not strongly affected by slight variations or errors in the emissivity. 3. Model for temperature identification 3.1. Radiation model The basic idea was of course to link the temperature and the intensity emitted by the surface, using the Planck's law [18] for a black surface: Fig. 7. IR image at 2353 cm1, converted in spectral intensity. Incident flux 50 kW/m2, time t ¼ 80 s.
Ibh ¼
1 C1 h3 p eC2T h 1
(1)
where C1 ¼ 3.7418$1016 W m2, C2 ¼ 1.4388$102 m K and h stands for the wavenumber in m1. The measured intensity was then expected to allow identifying the surface temperature, simply inverting relationship (1). However, as the received flux is in fact a combination of the flux emitted by the surface, plus the reflection of the radiation coming from the cone calorimeter, the measured intensity corresponds to:
Ih ¼ εh Ibh ðTðtÞÞ þ rh Ibh ðTcone Þ
Fig. 8. Absorptivity of the composite (virgin sample and degraded sample under 35 and 65 kW/m2).
thermal degradation under incident fluxes equal to 35 and 65 kW/ m2 (as a whole, more than 20 samples were studied in order to investigate the effect of the incident flux and of the degradation duration). These cases were chosen for the present illustration because they represent two different situations: i.e. degradation without auto-ignition for the lowest flux and with ignition for the highest one. As can be seen, the absorptivity of the virgin material is relatively high (close to 0.88 in average), with a near grey behavior considering that spectral variations are kept in a narrow range between 0.86 and 0.91. The thermal degradation induces some changes, with a slightly higher average value (around 0.90) and a larger variation range between 0.84 and 0.95. These are still moderate changes, however. The influence of the incident heat flux is also weak, as detailed in Ref. [17]. At the specific wavenumber of
(2)
with T(t) standing for the surface temperature, Tcone for the cone temperature, εh is the spectral emissivity and rh ¼ 1εh is the surface reflectivity. This assumes a Lambertian irradiation from the cone (which emission is close to the one of a blackbody, as demonstrated in Ref. [13]) and a diffuse reflection for the sample. This reflection part is even the major contributor to the radiation received from the surface during the first instants of the experiment (note that radiation from the surroundings at ambient temperature is clearly negligible as compared to the cone radiation, it is not included in this relationship). Reflection cannot be simply removed after a rough evaluation of the incident flux, because any uncertainty in this value or in the reflectivity would result in systematic errors in the temperature. This error would be particularly strong during the first step of the sample heating, when the surface temperature is still moderate, because the main part of the received signal is due to the reflection and the “useful” part of the signal is weak. Considering that one of the main objectives of the present study was to provide temperature data as accurate as possible, especially during the heating step, for degradation model validation purpose, a fine correction must be introduced. In the second step of the temperature evolution, the emission becomes more and more influent in the global signal as the surface temperature increases. Then, the uncertainty due to the reflection decreases. This will be confirmed by section 4.2, when comparing our results with temperatures obtained thanks to usual methods of IR image processing. Finally, after ignition along the sample surface, the flame also provides an additional flux directly emitted toward the spectrometer or reflected by the sample surface, which must be taken into account. For the present analysis, these parasitic reflections were accounted for in the signal processing thanks to a subtraction method, which was applied on the IR images at 2564 cm1 and on the spectra. It was assumed that no significant change occurred in
432
Z. Acem et al. / International Journal of Thermal Sciences 112 (2017) 427e438
the incident radiation (and thus in the reflected intensity) during the experiment, since the main part was due to the cone calorimeter which was in stationary regime. The subtraction eliminated the reflection contribution and the variation in intensity was correlated to the temperature increase. The method can be summarized as follows: i. Evaluation of the instantaneous intensity from the direct signal acquisition thanks to the calibration data obtained with the high temperature blackbody ii. Calculation of the difference between the measured instantaneous intensity and the reference value at initial time (meaning the subtraction of the reflection from the cone): DIh0 ¼ Ih(t)Ih(t ¼ 0) iii. Evaluation of the temperature T(t) from an inversion based on the Planck's law applied to the intensities involved in step (ii):
131
0
C2h 1 6 B pDIh0 C7 TðtÞz4 ln@ þ eTðt¼0Þ A5 3 hC2 εh C1 h
(3)
The temperature evaluation was initialized at T(t ¼ 0 s) ¼ 293 K. The possible drawback of this method is that the reflection subtraction assumes no change during all the experiment, which is a questionable hypothesis especially when the flame appears along the sample, or because the emissivity was seen to slightly vary with time. Therefore a second method was tested, replacing step (ii) by a subtraction between two successive images, such that the reflection part was only assumed constant between two acquisition times (down to 1 s or 0.2 s in some cases for the camera, around 3 s for the spectrometer). The method becomes: i. Evaluation of the intensity from the direct signal acquisition thanks to the calibration data obtained with the high temperature blackbody ii. Calculation of the intensity difference between two successive acquisitions DIh ¼ Ih(t)Ih(tDt) iii. Evaluation of the temperature T(t) from the inversion of the Planck's law for radiation:
2
131
0
1 6 B pDIh TðtÞz4 ln@ þe hC2 εh C1 h3
C Tðt2hDtÞ
C7 A5
800
(4)
Initialization was still done with T(t ¼ 0 s) ¼ 293 K. The reflection correction is better taken into account with this second method, but any error in a temperature evaluation at time t is kept and cumulated with other possible errors during all the experiment, which could cause some deviations for the final result. Actually, both methods gave the same results in our case and the two devices (spectrometer and camera) also gave the same temperature evolutions (within a reasonable uncertainty range) as it will be seen later in the result section. This shows that the reflection correction is well controlled and thus gives some confidence in the results.
700
Temperature [K]
2
according toh a representative exponential law such as i t TðtÞ ¼ Tð0Þ 1 e =t , with T(0) ¼ 293 K and t ¼ 100 s. This relationship was chosen as it looks like the temperature increase observed in present degradation tests, as will be seen later. Then, the theoretical intensity sent by the surface was predicted using relationship (2), considering an emissivity close to the actual value (εh ¼ 0.90) and diffuse reflection (with reflectivity rh ¼ 0.10) of an incident intensity coming from a blackbody at temperature Tcone ¼ 992 K (the set temperature of the cone for an incident flux of 35 kW/m2 in our experiments). This “synthetic” intensity was then used for a mock test of temperature evaluation from intensity data. This test is a kind of inverse crime, since the same model is used in a forward and a reverse application for the data generation and the identification. Therefore, it cannot be considered as a full validation. However, it allows investigating the sensitivity of the results to a hypothetic error in the emissivity (value 0.85 tested), in the initialization (error tested in initial temperature 303 K instead of 293 K), or due to a noisy acquisition (intensity acquired with a random noise up to 5% of the signal). Results are presented in Figs. 9 and 10 for the different temperature evaluations. Only the results based on the subtraction of the data obtained at two successive steps are presented, as no difference could be seen between the results based on relationships (3) or (4). All curves (basic evaluation and data disturbed by error in initial temperature, emissivity or noise) seemed to overlap and there was no significant discrepancy or uncertainty visible with this temperature scale (consequently, only the exact “synthetic” temperature and the identification in nominal conditions with the exact emissivity and initial temperature are shown in Fig. 9 for the sake of visibility). Discrepancies in the evaluation appear when the absolute error (true temperature minus identified temperature) is plotted as a function of time (Fig. 10). Evaluation at the first instants may present uncertainties if initial conditions are not well known, but results converge toward the same values for larger times, which are very close to the exact result if the emissivity is well known. In this case, long time evaluation can be taken with confidence, error being less than 1.5 K for an absolute temperature of 800 K. If an error occurs in the emissivity evaluation, the temperature may shift from the exact value, but the discrepancy is still in a range of 10 K here, as
600
500 Exact Identified 400
3.2. Temperature identification and sensitivity analysis Before using the method seen in the previous paragraph on real experimental data, a numerical test was carried out first, in order to check our ability to identify a given temperature evolution in an accurate manner. A typical evolution of the temperature was considered during 300 s in the range between 293 K and 800 K
300
0
50
100
150
200
250
300
Time [s] Fig. 9. Numerical test on the surface temperature evaluation. Exact temperature (black solid line) compared with identified data (triangles).
Z. Acem et al. / International Journal of Thermal Sciences 112 (2017) 427e438
an experiment conducted during 50 s. Higher instantaneous peaks were observed, but they were rather interpreted as noise and the error fluctuated in a range between 0 and 16 K for 90% of the measurement step. We even think that this provides an upper limit of the error since the present measurement was done throughout a fully developed flame, while the flame along the sample during degradation tests is not perfectly covering the measurement area in our experiments. However, this possible error, around 7 K, must be kept in mind in the analysis presented in the next section.
1 0 -1 -2
Absolute error [K]
-3 -4 -5 -6
4. Results
-7
4.1. First experimental campaign
-8 Identified temperature
-9
Error by 10 K in the initial temp.
-10
Emissivity set to 0.85
-11 -12
433
Random noise (5%) in the intensity difference 0
50
100
150
200
250
300
Time [s] Fig. 10. Absolute error between exact and identified data, with supplementary data obtained when introducing an error by 10 K in the initial temperature, or an error in the emissivity (0.85 instead of 0.90) or random noise in the input data (±5% on the instantaneous intensity computed from reference data).
compared to the 800 K of the exact value. As a consequence, present results are obtained with moderate or weak uncertainty. This is especially due to the fact that the possible error due to reflection is well controlled thanks to the image subtraction process. Without such a method, approximation in the reflection evaluation would result in dramatic uncertainties during the first instants. 3.3. Measurement bias due to the flame The optical filter avoids any possibility of measurement error due to radiation emission by the combustion gases, but soot may still affect the radiation measurement, inducing a possible temperature overestimation when auto-ignition occurs. Despite the fact that the flame was supposed to be optically thin (low soot concentration and flame thickness observed to be of the order of one centimeter, hence resulting in a weak influence) the related effect was evaluated through a dedicated calibration experiment. The idea was to generate a realistic flame in front of a surface with well known temperature and to evaluate this temperature with the above-presented methods. For that, an extended blackbody ECN100 by HGH Infrared Systems (30 cm 35 cm) was used, with surface temperature set to 823 K. A carbon composite sample was set along the path line, in vertical position, between the camera and the blackbody and it was ignited using a burner. After sufficient heating, the measurement started when the burner was withdrawn, the flame being maintained by the sample combustion itself. A comparison was done with and without perturbation by the flame. The chosen sample was one 100 mm by 100 mm, 5 mm thick, carbon composite piece of the same series than the one studied in the next section (in order to expect a flame similar to the one encountered during the auto-ignition step, even if the heating step was not the same). Then, the blackbody was observed throughout the flame, just above the composite sample, and its surface temperature was evaluated in the conditions of the present study using both methods of image subtraction. Results show a systematic overestimation of the surface temperature, because soot induce a supplementary intensity which is interpreted as a higher emission level from the surface. The error was evaluated to 7 K in average, for
As above-explained, the present method was used for the study of carbon composite samples under incident fluxes equal to 20, 35, 50 and 65 kW/m2. A first campaign was conducted on samples with surface 100 mm 100 mm, 5 mm thick. The corresponding temperature evolutions with time are presented in Figs. 11 and 12. Fig. 11 gives an idea of the confidence that could be given to the results for two chosen fluxes (20 and 50 kW/m2). First of all, tests were repeated twice in order to check the repeatability (see the two sets of data per fluxes very close one from the other). Then, the two above-discussed methods, namely reflection subtraction based on two successive images or using the initial image as a reference, were compared (triangles vs continuous lines) with no apparent difference. Finally, the two devices (spectrometer and camera) were also compared, providing the same results (crosses for the spectrometer, lines and triangles for the camera). These tests reported in Fig. 11 give an idea of the reliability of the whole experimental campaign. Then, Fig. 12 allows comparing the sample evolution as a function of the incident flux, still presenting two tests in each case with the same incident flux. All curves show a sharp increase in temperature (several hundreds of degrees during the first minute) and a plateau always reached between 300 s and 400 s as a consequence of an energy balance for the sample. The highest the incident flux is, the sharpest the temperature increase
Fig. 11. Time evolution of the surface temperature for two incident fluxes (20 and 50 kW/m2). Continuous lines and triangles for the camera data, crosses for the spectrometer data. Two sets of results for each flux (repeatability test).
Z. Acem et al. / International Journal of Thermal Sciences 112 (2017) 427e438
Fig. 12. Time evolution of the surface temperature for various incident fluxes (camera data). One particular curve at 35 kW/m2 was obtained using piloted ignition after 100 s of degradation (shifted from the two other curves between 100 and 300 s because of the flame contribution).
is. Some fluctuations appeared for the highest fluxes after 100 s as a consequence of ignition at the sample surface, which provided a supplementary incident flux for the degradation. It did not happen in standard conditions for the incident fluxes of 20 and 35 kW/m2, but it was forced using a piloted ignition device (a spark igniter) in one special test at 35 kW/m2. This explains why one of the three curves for 35 kW/m2 indicates a higher temperature level as compared to the two others between 100 and 300 s. This further increase in temperature is due to the radiative feedback from the flame. The gap between this particular curve and the two others shows the influence of the flame developing along the sample. When the flame disappeared after 300 s, the curve tended toward the same plateau than the two others. This was confirmed by both apparatuses with independent measurements. These data were found useful for people working on thermal degradation models, but the sharp increase during the first step showed that further experiments should be conducted decreasing the acquisition time, which is discussed in one of the next sections. 4.2. Comparison of the method with standard temperature evaluations based on IR images Our method based on image subtraction was especially developed because of the expected difficulty to evaluate the part of the radiative flux, due to the reflection by the sample, of the incident flux emitted by the cone. This contribution is dominating the measured radiative flux during the first step of the sample heating. In order to compare our results with data obtained with other usual methods, we conducted dedicated evaluations of the surface temperature, considering what could be a basic use of the camera images: (1) a basic straightforward evaluation of the equivalent blackbody temperature simply inverting the Planck's law given by relationship (1) (providing the temperature of a blackbody emitting the same radiative flux than the sample), (2) a corrected temperature involving the radiative properties of the sample: emissivity 0.91 and reflectivity 0.09, assuming that the incident flux from the cone is the value measured by a heat flux gauge located at the
sample position, with the spectral distribution given by the Planck's law for a blackbody at the temperature set for the cone (measured with three thermocouples on the coil). The case of the 35 kW/m2 incident flux with piloted ignition was chosen for an illustration of the discrepancy between these basic evaluations and our method. Fig. 13 presents the different temperature estimates and Fig. 14 evaluates the discrepancy between the different methods when compared with our evaluation (temperature obtained with one of the usual evaluations minus temperature according to our method). The equivalent blackbody temperature strongly overestimates the true temperature during the first step of heating (discrepancy close to 300 K at the beginning of the test), since the sample is still cold whereas the reflection from the cone is interpreted as if it was proper emission. The gap decreases because true emission by the sample progressively becomes the dominating contribution to the radiative flux. When the stationary step is reached, the overestimation is between 5 and 10 K. The effect is limited because the cone temperature is not really far from the sample one, the gap could be larger in different conditions of irradiation. The temperature is still overestimated because the cone has a higher temperature (992 K) than the sample. The second evaluation which aims at considering the flux from the cone reflected by the sample is better performing, but the first step of heating is still approximated because the correction cannot be perfect. The gap is close to 40 K at the beginning, finally decreasing to a few kelvins during the stationary step. One can also note some fluctuations during the flaming period (mainly in the range between ± 50 K for times ranging between 150 s and 250 s). Note that the discrepancy at the beginning of the test could be higher for larger incident fluxes. The accuracy of the correction for the reflected flux strongly depends on the ability to measure the sample emissivity and the cone temperature (any error would result in a rise in discrepancy), while our method is little dependent on the emissivity and does not depend at all on any reflection evaluation which is directly eliminated with the subtraction process.
1000 -2
65 kW.m 900
800
Temperature [K]
434
-2
35 kW.m
700
600
500
400
300
Present method Equivalent blackbody temperature Corrected temperature involving the cone radiation
0
200
400
600
Time [s] Fig. 13. Temperature evaluation with the present method and using two basic evaluations: an equivalent blackbody temperature and a corrected estimate involving the reflection of the radiation from the cone.
Z. Acem et al. / International Journal of Thermal Sciences 112 (2017) 427e438
435
800
Temperature [K]
700
600
Test 1 500
Test 2 Test 3 Test 4
400
Test 5 Test 6
300
0
50
100
150
200
Time [s]
Fig. 14. Absolute discrepancy between the basic evaluations of Fig. 13 and our method.
4.3. Second experimental campaign A second experimental campaign was conducted on samples with a surface 50 mm 300 mm, 5 mm thick (this particular sample size was designed to allow a simultaneous study of thermal degradation and structure modification under mechanical stress, results are only commented regarding heat transfer here). Despite an incident flux equal to 35 kW/m2, auto-ignition was systematically observed here, whereas this flux was not high enough to allow auto-ignition in the previous series. This can be explained by the difference in size for the samples, the use of a different sampleholder which can affect the heat balance for the sample, or other phenomena influencing the concentration of pyrolysis gases (which is close to the inflammability limit in both cases but can be below or above the lower limit depending on the exact set up configuration). Auto-ignition process is not yet satisfactorily described and is beyond the scope of the present study rather focused on the surface temperature evaluation. In this second campaign, the studied case was a sample progressively heated, until auto-ignition appeared along the surface. The same method of IR imaging was applied, using the IR camera with the filter at 2564 cm1 and applying the image subtraction method for data processing (subtraction of two successive images following relation (4)). The time step of 0.2 s between two images allowed expecting a better description of the transient heating of the sample. The test was focused on the first step of heating (between 100 and 200 s after the beginning of the irradiation by the cone). Fig. 15 shows the time evolution of the sample temperature (average value in the center of the sample as for the first experimental campaign). Six tests are presented, showing the quite good repeatability of the experiments, especially during the first 100 s. The same trends were observed but the auto-ignition did not occur exactly at the same instant. It appeared through the temperature jump in the different tests, between 85 and 120 s, always when the sample surface temperature reached a threshold value around 650 K. Such composite samples made of carbon fibers and epoxy resin present some heterogeneities in their structure and contents, explaining these discrepancies, which are still moderate here. Moreover, as above-discussed, this case is probably close
Fig. 15. Surface temperature under a flux of 35 kW/m2, second experimental campaign, six repeatability tests.
to the conditions limiting the two behaviors with or without autoignition, also explaining the discrepancies. The method itself showed its ability to capture the sharp temperature increase with a quite good repeatability and the short time step used for the image acquisition allowed tracking the temperature with a finer resolution. 4.4. Evaluation of the heat flux induced by auto-ignition Beside the observation of the degradation mechanisms and the measurement of the surface temperature, present data were further used to evaluate the additional incident flux due to the flame, which affects the sample in addition to the initial radiation from the cone apparatus. This is a crucial problem as the standard use of the cone calorimeter for the characterization of materials considers the heat flux calibrated at the beginning of the experiment as the only incident flux, while it is obvious that the autoignition produces an additional flux toward the sample due to radiative feedback and convection between the flame and the surface. Some attempts were reported in the past to measure this flux with heat flux gauges embedded in the sample (see Refs. [19e22] for example) but this is a tricky measurement in a harsh environment. The values found in the literature are clearly depending on the sample type, its orientation, its size and the incident flux. Most of the presented values are in a range between 10 and 20 kW/m2, with some higher fluxes reported for PMMA in Ref. [22] for example, but for larger samples. We started here with the observation that a significant rise of the surface temperature occurs due to the auto-ignition of the pyrolysis gases, which is actually due to the heat provided by the flame to the surface. This led us to the conclusion that a model for the temperature evolution at the sample surface can provide the required supplementary flux in order to explain the temperature gap. This was the guiding idea of our investigation. Recalling that an overestimation of the temperature is suspected when the flame is developed, the present test must be considered as a feasibility test and a quantitative confirmation will be required in a future work. In a first step, a simple 1D transient conduction model was implemented for the sample, considering the ambient temperature
Z. Acem et al. / International Journal of Thermal Sciences 112 (2017) 427e438
as the initial condition for the whole domain, with boundary conditions accounting for the incident flux measured from the heat flux gauge during the calibration step, plus radiative and convective exchanges on both faces. Let us consider a medium with thickness E. The face submitted to the incident radiation is at position x ¼ E, the other face (at x ¼ 0) being only submitted to radiative and convective exchanges with ambient air. The governing equation for temperature prediction inside the medium is the following 1D transient balance energy for a solid opaque medium:
rC
vT v2 T ¼ l 2 þ q_ vt vx
(5)
where r, C and l stand for the thermo-physical properties of the medium (density, heat capacity and thermal conductivity) and q_ is a sink term aimed at describing the pyrolysis process and the corresponding latent heat). Actually, the mass loss rate was not available for the present tests and the term q_ was not included in the model. This is of course an obvious shortcoming of the present analysis. Note that the present analysis is restricted to the first three minutes of the degradation test. Hence, the actual mass loss and the corresponding source term are expected to be moderate. Moreover, the emphasis is put on the required supplementary flux in order to explain the temperature jump at the surface, which is expected to be affected by the flame feedback at the surface above all. Boundary conditions involve external convection and radiation as follows:
vT Irradiated surface ðx ¼ EÞ : l ¼hE ðTðEÞ T∞E Þ vx E 4 aqinc þ εs TðEÞ4 T∞E (6) vT Back surface ðx ¼ 0Þ : þl ¼h0 ðTð0Þ T∞0 Þ vx 0 4 þ εs Tð0Þ4 T∞0
(7)
where qinc is the incident flux to the sample, a is the absorptivity assumed to be equal to the emissivity ε, h0 and hE stand for the convective heat transfer coefficient at the two boundaries and T∞E or T∞0 stand for the temperature of the surrounding medium. At the beginning of the experiment, qinc is equal to the incident flux from the cone (qinc,0). After auto-ignition, an additional flux (qflame) is affecting the sample and qinc ¼ qinc,0 þ qflame. This latter value qflame was sought by our numerical routine. A value of 5 W/(m2 K) was set for the heat transfer coefficients (corresponding to a typical free convection situation). This is a choice guided by a standard evaluation of the Nusselt number for such a configuration. Some sensitivity tests varying this value showed very weak influence on the evaluation of the heat flux from the flame, as the order of magnitude of the incident flux and of the flux from the flame is obviously larger than convection at the boundaries. The emissivity was set to 0.91. A straightforward numerical scheme using a Finite Volume Method was used to solve the problem (not detailed here). One supplementary difficulty was the lack of knowledge of the thermal properties. Measurements raised problems because this composite material is heterogeneous and because properties may change along the degradation process. The chosen solution was to evaluate the equivalent average properties by optimization, identifying the set of thermal properties which give the best fit of the measured temperature evolution during the time period before auto-ignition. Then, the simulation was conducted further, with these properties as input data, in order
to search for the additional flux which explains the time evolution of the temperature after auto-ignition. This simulation in two steps was therefore carried out as follows: (i) A first step of simulation during the first 80 s in order to evaluate the thermal conductivity l and the product rC (density heat capacity) which allows the best reconstruction of the temperature, using a particle swarm optimization algorithm (see Refs. [23,24] for an introduction to the method). Main numerical parameters were set to 50 particles and 50 loops (with a control of the convergence thanks to a cost function based on a second order norm evaluating the discrepancy between experimental and reconstructed data). The search space was set to [5.105e5.106 J K1 m3] for the product rC and [0.1e10 W m1 K1] for the thermal conductivity. (ii) A second step of simulation was carried out between 80 s and the end of the experiment, keeping the thermal properties identified in step (i) and searching for the additional flux required to explain the time-evolution of the temperature. A dichotomy scheme was used at each time step to track this flux corresponding to the auto-ignition influence. The convergence criterion was set to 0.1 K between the experimental and reconstructed temperatures. This method was applied on the six tests shown in Fig. 15. The step (i) provided average thermal properties as follows: 2.50 106 J K1 m3 for the product rC and 0.4 W m1 K1 for l. These are equivalent properties, which are in a very good agreement with what was predicted by Hidalgo et al. [5] who suggested an average value of 1.035 106 USI for the inertia product l r C on similar samples, based on the analysis of cone calorimeter tests. Fig. 16 shows a typical temperature reconstruction based on the two steps above-presented for test 6 (similar results were obtained for the other tests). The discrepancy between numerical and experimental data is below 6 K during the optimization step. This is a quite good agreement considering that identified properties are equivalent average data, kept constant during the process. Then, the discrepancy never exceeds 0.1 K in the second step, since this is the convergence criterion used in the dichotomy scheme.
800 Experimental Numerical 700
Temperature [K]
436
600
500
400
300
0
50
100
150
200
Time [s] Fig. 16. Temperature reconstruction in Test 6 (dotted line) versus experimental data (continuous line).
Z. Acem et al. / International Journal of Thermal Sciences 112 (2017) 427e438
437
the reflection of the incident radiation by the sample. This nonintrusive method gave the time varying evolution of the temperature during a series of degradation experiments for incident fluxes between 20 and 65 kW/m2. Both apparatuses independently provided results showing the same sharp temperature increase when the sample was irradiated, before reaching a plateau. Incident fluxes above 35 kW/m2 systematically induced a self-ignition of the pyrolysis gases after significant sample degradation. The temperature levels were observed to increase with the incident flux, between 740 K at 20 kW/m2 and 1100 K at 65 kW/m2. The flame produced by the auto-ignition itself was observed to result in a temperature jump around 200 K. Further analysis of the sample temperature in case of autoignition of the sample allowed the identification of the heat feedback from the flame. Peak values in the range between 15 and 20 kW/m2 were found for samples with surface 50 mm 300 mm, 5 mm thick. This supplementary flux is obviously affected by the sample size and the experimental conditions, but the method seems to be promising and the identification model could be refined for a systematic application. Acknowledgements Fig. 17. Evaluation of the additional flux due to auto-ignition for the 6 cases presented in Fig. 10.
Regarding the evaluation of the additional flux explaining the temperature jump due to the flame, Fig. 17 presents the flux result in all cases. A sudden peak occurs at the ignition instant (a significant concentration of pyrolysis gas is available and quickly burns), then a decrease is observed. We chose to stop this evaluation after a reasonably short time because the mass loss effect is not involved (as above-discussed) and also because constant equivalent thermal properties are assumed. This latter assumption could not be kept after significant degradation, as the delamination of the sample involves strong heterogeneities and modifications of the properties. As can be seen in Fig. 15, the duration of these experiments was quite short after ignition, and further comparison was not allowed anyway. There is some variability in the flux evaluation, which is logical when considering the temperature curves in Fig. 15. They are not perfectly super-imposed and the sudden rise in temperature due to the auto-ignition does not occur exactly at the same time. Consequently, the additional fluxes presented in Fig. 17 have the same global shape and the same order of magnitude but with a significant variability. Despite the well-controlled operating conditions under the cone, there are some variations in the sample composition degradation, the degradation process itself, the ignition conditions (local concentration and temperature) which may explain this observation. At this stage, the order of magnitude of an additional flux with a peak around 15e20 kW/m2 seems to be confirmed when the auto-ignition occurs. However, the abovediscussed warnings regarding the accuracy of the temperature measurement must be kept in mind. The method itself is promising and further tests and developments are encouraged, in particular refining the experiment, extending the model with the pyrolysis and mass loss effects, addressing the property modifications and perhaps the structure evolution of the sample. 5. Conclusion The surface temperature of a carbon composite sample was identified during its degradation under a cone calorimeter, based on infrared observation with an IR camera and a FTIR spectrometer simultaneously. A special subtraction process was applied on spectra and images in order to eliminate the parasitic signal due to
The research leading to these results has received funding from the European Union's Seventh Framework Programme (FP7/ 2007e2013) for the Fuel Cells and Hydrogen Joint Technology Initiative under grant agreement n 325329. This work also pertains to the French Government program “Investissements d’Avenir” (LABEX INTERACTIFS, reference ANR11-LABX-0017-01). References [1] ISO 12136:2011. Reaction to fire tests e measurements of material properties using a fire propagation apparatus. 2011. [2] ISO 5660. Reaction to fire tests - heat release, smoke production and mass loss rate - Part 1: heat release rate (cone calorimeter method). 2015. [3] Stoliarov SI, Crawley S, Lyon RE, Linteris GT. Predictions of the burning rates of non-charring polymers. Combust Flame 2009;156:1068e83. [4] Kacem A, Mense M, Pizzo Y, Boyer G, Suard S, Boulet P, Parent G, Porterie B. A fully coupled fluid/solid model for open air combustion of horizontallyoriented PMMA samples. Combust Flame 2016;170:135e47. [5] Hidalgo JP, Pironi P, Hadden RM, Welch S. A Framework for evaluating the thermal behaviour of carbon fibre composite materials. In: Proc.2nd european symp. Fire safety science; 2015. [6] Li J, Gong J, Stoliarov SI. Gasification experiments for pyrolysis model parametrization and validation. Int J Heat Mass Transf 2014;77:738e44. [7] Li J, Gong J, Stoliarov SI. Development of pyrolysis models for charring polymers. Polym Degrad Stab 2015;115:138e52. [8] De Vries J, Ren N, Chaos M. Temperature measurements on solid surfaces in rack-storage fires using IR thermography. In: Proceedings of SPIE - the international society for optical engineering, 9485, 94850H; 2015. [9] Melendez J, Foronda A, Aranda JM, Lopez F, Lopez del Cerro FJ. Infrared thermography of solid surfaces in a fire. Meas Sci Technol 2010;21. 105504 (10pp). [10] Boulet P, Parent G, Collin A, Acem Z, Porterie B, Clerc JP, et al. Spectral emission of flames from laboratory-scale vegetation fires. Int J Wildland Fire 2009;18(7):875e84. [11] Boulet P, Parent G, Acem Z, Kaiss A, Billaud Y, Porterie P, et al. Experimental investigation of radiation emitted by optically thin to optically thick wildland flames. J Combust 2011;2011. Article ID 137437. € rsth M, Bal N, et al. Radiation emission [12] Boulet P, Parent G, Acem Z, Collin A, Fo from a heating coil or a halogen lamp on a semitransparent sample. Int J Therm Sci 2014;77:223e32. [13] Boulet P, Parent G, Acem Z, Rogaume T, Fateh T, Zaida J, et al. Characterization of the radiative exchanges when using a cone calorimeter for the study of the plywood pyrolysis. Fire Saf J 2012;51:53e60. [14] Billaud Y, Boulet P, Pizzo Y, Parent G, Acem Z, Kaiss A, et al. Determination of woody fuel flame properties by means of emission spectroscopy using a genetic algorithm. Comb Sci Technol 2012;185(4):579e99. [15] Pouplin J, Collin A, Acem Z, Parent G, Boulet P, Vena P, et al. Study of a V-shape flame based on IR spectroscopy and IR imaging. J Phys Conf Ser 2016;676: 012018. [16] Batiot B, Quach THY, Acem Z, Brissinger D, Collin A, Richard F, et al. Correlation between degradation of an epoxy-carbon fiber material and surface radiation
438
Z. Acem et al. / International Journal of Thermal Sciences 112 (2017) 427e438
during the thermal exposure. 2016. Submitted to Fire Tech. [17] Boulet P, Brissinger D, Collin A, Acem Z, Parent G. On the influence of the sample absorptivity when studying the thermal degradation of materials. Materials 2015;8:5398e413. [18] Modest MF. Radiative heat transfer. third ed. Academic Press; 2013. [19] Rhodes BT, Quintiere J. Burning rate and flame heat flux for PMMA in a cone calorimeter. Fire Saf J 1996;16(3):221e40. [20] Beaulieu PA, Dembsey NA. Flammability characteristics at applied heat flux levels up to 200 kW/m2. Fire Mater 2008;32:61e86. [21] Li K, Pau DSW, Wang J, Ji J. Modelling pyrolysis of charing materials: determining flame heat flux using bench-scale experiments of medium density
fiberboard. Chem Eng Sci 2015;123:39e48. [22] Kacem A, Mense M, Pizzo Y, Boyer G, Suard S, Boulet P, et al. A fully coupled fluid/solid model for open air combustion of horizontally-oriented PMMA samples. Combust Flame 2016;170:135e47. [23] Kennedy J, Eberhart RC. Particle swarm optimization. In: Proceedings of IEEE international conference on neural networks, Piscataway, NJ; 1995. p. 1942e8. [24] Shi Y, Eberhart RC. A modified particle swarm optimizer. In: Proceedings of the IEEE congress on evolutionary computation (CEC 1998), Piscataway, NJ; 1998. p. 69e73.
Contents lists available at ScienceDirect
International Journal of Thermal Sciences journal homepage: www.elsevier.com/locate/ijts
Surface temperature of carbon composite samples during thermal degradation Zoubir Acem a, b, Damien Brissinger a, b, Anthony Collin a, b, Gilles Parent a, b, Pascal Boulet a, b, *, Thi Hay Yen Quach c, Benjamin Batiot c, Franck Richard c, Thomas Rogaume c a b c
Universit e de Lorraine, LEMTA, UMR 7563, Vandoeuvre-l es-Nancy, F-54500, France CNRS, LEMTA, UMR 7563, Vandoeuvre-l es-Nancy, F-54500, France PPRIME Institute, CNRS, Universit e de Poitiers, ISAE ENSMA, BP 40109, F86961, Futuroscope cedex, France
a r t i c l e i n f o
a b s t r a c t
Article history: Received 15 September 2016 Received in revised form 11 October 2016 Accepted 7 November 2016 Available online 13 November 2016
An infrared camera and a Fourier transform infrared spectrometer were used simultaneously to observe the radiation from the surface of carbon composite samples during thermal degradation experiments. Surface temperatures were estimated from radiation measurements conducted during cone calorimeter tests. Infrared spectra and images were post-processed involving a subtraction method between successive images, to withdraw the high incident flux from the cone calorimeter partly reflected by the sample. The surface intensities were first evaluated as a function of time. Then, an identification step was implemented linking the rise in intensity between two time steps with the increase in sample temperature. Corrections were introduced for the reflection of the incident radiation from the heater and for the true emissivity of the material. Both experimental devices e camera and spectrometer e showed temperature evolutions in a good agreement. The sharp temperature increase of the sample and the auto-ignition of the pyrolysis gases on the surface were observed and evaluated for various incident fluxes from the cone between 20 and 65 kW/m2. Measurements carried out with a short time step down to 0.2 s further allowed an evaluation of the supplementary flux due to the auto-ignition with a peak up to 20 kW/m2 for an incident flux of 35 kW/m2 on a carbon composite sample with surface 50 mm 300 mm and thickness 5 mm. © 2016 Elsevier Masson SAS. All rights reserved.
Keywords: Pyrolysis Cone calorimeter Infrared imaging Radiative transfer Temperature identification Heat flux Flame radiation
1. Introduction The study of the thermal degradation of materials provides some of the input parameters required for the prediction of their ignition and combustion. Standard characterization involves samples submitted to given external radiative fluxes, usually produced by a high temperature lamp in the Fire Propagation Apparatus [1], or a coil when using a cone calorimeter [2]. A calibration with a heat flux gauge ensures that the incident flux is well controlled and data are registered for the mass loss experienced by the sample as a function of time. This problem involves a complex coupling of heat and mass transfer. Therefore, temperature data are also required for
de Lorraine, LEMTA, UMR 7563, Vandoeuvre* Corresponding author. Universite s-Nancy, F-54500, France. le E-mail address: [email protected] (P. Boulet). http://dx.doi.org/10.1016/j.ijthermalsci.2016.11.007 1290-0729/© 2016 Elsevier Masson SAS. All rights reserved.
a better understanding of the phenomena and for validation purpose. These data are then compared to results obtained with degradation and combustion models, which are needed in fire simulations for prescribing the mass loss rate and for predicting the heat release or the heat transfer inside the materials. The knowledge of the sample surface temperature is of particular interest for the description of the heat transfer inside the sample and for the definition of reliable boundary conditions. Thermocouples may be used to measure this surface temperature, but the confidence in the acquired data is questionable as direct radiation to the thermocouple may occur from the source or from the flame developing on the surface after sufficient heating. In addition, this is a high temperature measurement possibly disturbed by a non perfect contact (a specific problem which would be encountered in the present study devoted to a carbon composite material with a low conductivity) or also because of soot deposit. To avoid these problems,
428
Z. Acem et al. / International Journal of Thermal Sciences 112 (2017) 427e438
measurements inside the sample can be done with embedded thermocouples, extrapolating the values to the surface. Stoliarov et al. [3] presented temperature data measured with thermocouples embedded as close as possible to the surface (1.5 mm below it) in order to characterize the top layer temperature. Pizzo et al. [4] on PMMA and Hidalgo et al. [5] on a carbon composite sample from the same series as the one studied here, to name some recent studies, presented temperature measurements with embedded thermocouples at various positions inside the materials. Such measurements are usually used as validation data for models. Temperature identification could be also sought based on the observation of the back surface, which is protected from direct irradiation. However, other difficulties are raised owing to the nonstationary conditions and to the problem complexity (including the required model for heat transfer, through a complex medium with varying properties and the difficulty to do the measurement itself). Li et al. carried out such measurements at the back face and examined the accuracy of the surface temperature [6,7]. Data were then used for pyrolysis model validation purpose. The present work was conducted to test another method based on the observation of the flux emitted by the front surface with a Fourier Transform InfraRed (FTIR) spectrometer and an infrared (IR) camera. Our objectives were twofold: (i) to observe the radiation sent by the irradiated surface during a cone calorimeter experiment in order to follow the temperature increase and to check for the temperature homogeneity on the surface; (ii) to develop a reliable method to identify the temperature despite the difficulty raised by the experimental setup (limited accessibility and strong background irradiation because of the cone itself). IR thermography was already used in the past to observe solid surfaces through a flame. For example, De Vries et al. [8] evaluated surface temperatures in rack-storage fires with a long wave camera (expecting limited perturbations by smoke and particles). They observed perturbations in the measurement because of the flame, evaluating the related uncertainty in the measured temperature to approximately 50 K (in a case of a large scale fire). Melendez et al. [9] also measured surface temperature in the case of a plate heated by a burner flame. They used a combination of an optical filter and an image processing involving the flame image subtraction to withdraw the perturbation by the flame. Similar ideas were used here, but in different conditions because the heat source was not a burner and the flame ignition or extinction could not be controlled. The FTIR spectrometer and the IR camera were already used by present authors on several applications related to fire [10e13], for the characterization of sources or flames in particular. Intensities from flames and radiative sources were especially measured in various situations. Based on past experiences [14,15], spectra can be analyzed to identify characteristic temperatures in some wavenumber ranges. Filters can be also added to the camera in order to limit the perturbation of the signal by the flame and to isolate the surface radiation, which is then analyzed to identify its temperature. These methods were used here when registering the radiation from the sample surface. For the present study, composite carbon samples were degraded under a cone calorimeter, varying the incident heat flux between 20 and 65 kW/m2. The surface emission of the samples was simultaneously registered with the spectrometer and the camera. Then, several identification methods were tested in order to obtain the temperature and a satisfactory confidence in its time-varying evolution. In the following sections, the experimental setup will be described first. Then, the radiative transfer model and the identification methods will be presented. Finally, results will be discussed for the evolution of the surface temperature of the carbon composite.
2. Experimental setup 2.1. The composite sample The material studied in the present work is made of carbon fibers in an epoxy matrix, a composite used for high pressure hydrogen storage tanks. Samples were prepared in the frame of the FIRECOMP European project aimed at characterizing the degradation of such materials when submitted to fire sources. The composite is mainly composed of carbon (85% of the total weight according to the available elementary analysis conducted in the frame of the project [16]). Specimens were cut in large composite cylinders, providing quasi-flat samples for degradation tests under cone calorimeter. 2.2. The cone calorimeter combined with infrared metrology A picture of the setup is presented in Fig. 1. In this case, a standard cone calorimeter [1] was used in vertical configuration for the degradation of a 100 mm 100 mm x 5 mm sample. The composite sample was placed in a sample holder specially developed for this application. This one was manufactured into a block of aluminum, and a cover was used in order to maintain the sample in the right position. The opened surface of the cover was 88.4 cm2 in accordance with ISO 5660 [1]. It is important to note that no insulation was used in the back surface of the sample. The heat source was in fact a truncated conical heater and the coil of the cone can be seen in the picture. Radiation was provided from this coil toward the sample surface, leading to its thermal degradation. The picture was taken after auto-ignition of the gas mixing as shown by the flame developing along the sample surface. The observation in the infrared range was done along a path centered on the cone axis through its aperture, using a camera and a spectrometer which are described below. A silicon beam splitter divided the radiation signal toward the infrared camera and the spectrometer for a simultaneous observation of the same area with both devices. A typical test started when the sample was suddenly submitted to the radiation emitted by the cone at t ¼ 0 s (removing a shield used between the heater and the sample to allow the heating of the cone up to the stationary regime while protecting the sample before the experiment starts). Four different incident fluxes were considered: 20, 35, 50 and 65 kW/m2 (values obtained from heat flux gauge measurements carried out at the sample position in standard use of the cone [1]). Experiments were repeated twice to check the repeatability.
Fig. 1. Picture of the setup with the cone calorimeter, the spectrometer and the IR camera.
Z. Acem et al. / International Journal of Thermal Sciences 112 (2017) 427e438
429
2.3. FTIR spectroscopy The spectrometer (Matrix by Bruker) was used in a configuration involving an external source (the sample surface) in the spectral range between 800 and 5000 cm1 in the present study. Background measurements were done in a preliminary step, for an evaluation of the surrounding influence. The contribution of the heater itself was taken into account in the post-processing step based on the analysis of successive spectra analysis as described below. Each spectrum was based on a series of 3e10 scans, as a usual technique used to reduce the signal-to-noise ratio. This number of scans is small (as compared to usual FTIR measurements), but it warranted a quite instantaneous measurement in order to avoid a too strong temperature variation on the surface during the spectrum registration. Figs. 2 and 3 show typical spectra registered during experiments conducted with incident fluxes equal to 35 and 50 kW/m2 respectively. Only a limited number of the numerous acquisitions are plotted here in order to keep a good visibility of the spectrum evolution. Data were converted into equivalent intensities after comparison with calibration data obtained using a high temperature blackbody (M330 by Mikron, at 1500 C). Spectral intensities are plotted as a function of wavenumber for different times. They correspond to the signal received by the spectrometer from the surface, meaning that both proper emission and radiation reflection are involved. During the early step of the experiment, reflection was dominating the intensity, but emission became progressively dominant as the sample was heated. The intensity level increased with time as a consequence of the surface temperature increase. For an incident flux equal to 35 kW/m2 (Fig. 2), the spectrum shapes were found similar, with an increasing level with time. A standard spectral evolution was obtained, as for a usual surface emission, affected by the sample emissivity, with some sharp variations in definite bands due to the gas influence (mainly absorption by atmospheric water vapor and carbon dioxide). No spontaneous ignition was observed, explaining the monotonic increase in the intensity, without emission peak. For an incident flux equal to 50 kW/m2 (Fig. 3), the first increasing phase was followed
Fig. 2. Radiative intensity emitted and reflected by the sample surface at different times when irradiated by a cone calorimeter with a 35 kW/m2 flux.
Fig. 3. Radiative intensity emitted and reflected by the sample surface at different times when irradiated by a cone calorimeter with a 50 kW/m2 flux.
by a sudden increase in the intensity level and a sharp peak observed around 2300 cm1 as a consequence of a flame suddenly developing along the surface (see the legends “phase 1” and “ignition” in the Figure). Then, a further increase was observed (“phase 2”), before a decreasing phase and the emission peak disappearance as the flame extinguished (see the curves at 260 and 600 s, phase 3), most of the sample being degraded. 2.4. IR camera The camera (Orion SC7000 by FLIR) was used with filters in order to isolate images at given wavenumbers. For the present work, the filter at 2564 cm1 was especially used to observe the surface, since it operates in a spectral region where gases (H2O, CO and CO2 especially) do not contribute to the radiation, hence preventing from any perturbation of the image processing by possible combustion gases, but soot can still contribute to the emitted radiation as discussed later. On the contrary, specific filters around 2353 and 3500 cm1 were used in the same time to observe the flame through its combustion products CO2 and H2O, respectively. A filter wheel was used to change the filters automatically with a high frequency rotation speed resulting in a quite simultaneous observation. The time step between two acquisitions was first set to 1 or 3 s during a first test campaign. Then, it was decreased to 0.2 s for a second test campaign in order to catch faster temperature changes accurately. Both sets of results will be commented in the result section. For each acquisition, 12 images were acquired at a frequency of 50 Hz in order to allow successive instantaneous observations or the computation of average intensities and temperatures on a short period. Background radiation was also measured in a preliminary step and the main difficulty raised by the reflection of the incident flux sent by the cone calorimeter was addressed thanks to image subtraction as described in the next section. Camera images were studied in a first step in order to check the homogeneity of the radiation emitted by the surface. Two examples are given in Figs. 4 and 5, for the filter at 2564 cm1. Both images are presented in terms of intensity after a conversion of the camera signal based on reference data obtained in front of our reference
430
Z. Acem et al. / International Journal of Thermal Sciences 112 (2017) 427e438
Fig. 4. IR image at 2564 cm1, converted in spectral intensity (W/(m2 sr cm1)). Incident flux 35 kW/m2, time t ¼ 390 s.
emitted radiation. A calibration test was therefore conducted to evaluate the measurement bias in the temperature evaluation which can be expected because of the soot influence (see section 3). The flame was better emphasized when observing the surface through the filters at 2353 and 3500 cm1. Figs. 6 and 7 show images corresponding to the same case as in Fig. 5 (50 kW/m2, time t ¼ 80 s) but radiation involves here both a contribution from the sample surface and from H2O in the flame (Fig. 6) and CO2 in the flame (Fig. 7). As compared to Fig. 5, supplementary patterns are observed in the images, especially in Fig. 7, meaning that CO2 is the main contributor to the flame radiation. Indeed, H2O was not observed to contribute much to emission, probably because hydrogen content in the composite is weak (3% in weight for hydrogen, as compared to 85% for carbon as above-mentioned, according to tests conducted on the elementary composition of the composite sample [16]). This is in agreement with the spectra in Fig. 3, which showed a sharp peak in the range corresponding to CO2 emission, but no actual emission peak in the ranges associated to H2O (around 1600 and 3700 cm1).
2.5. Surface emissivity
Fig. 5. IR image at 2564 cm1, converted in spectral intensity (W/(m2 sr cm1)). Incident flux 50 kW/m2, time t ¼ 80 s.
blackbody. As for the spectrometer results, the intensity involves both proper emission and reflection. Note that the white circle at the center of the image features the area which was observed by the spectrometer (which received radiation in a limited solid angle and does not see the whole plate as the camera does). All data processing in the following sections (based on camera or spectrometer data) correspond to averages computed for the area inside this circle, with diameter 2.3 cm. On the images, some heterogeneities were seen, which correspond to the carbon fiber orientation and are probably due to the surface roughness affecting reflectivity and emissivity, but the temperature itself was not supposed to vary much along the surface. Fig. 4 is for a moderate incident flux of 35 kW/m2, which did not induce auto-ignition of the sample. On the contrary Fig. 5 was obtained for a higher incident flux (50 kW/ m2), after auto-ignition of the pyrolysis gases. The same scale was kept to emphasize the significant increase in intensity. The image shows some hot points which probably correspond to the flame developing along the surface in this case, despite the use of the filter at 2564 cm1, chosen to avoid the perturbation by the combustion gas, because soot, even in a weak concentration, probably
Results presented in the above sections are straightforward visualizations of the intensity sent by the sample. A further analysis of the true emission required the knowledge of its emissivity. It was evaluated based on transmissivity and reflectivity measurements of samples with a FTIR spectrometer in the spectral range of interest. The spectral absorptivity (equal to the spectral emissivity according to the Kirchhoff's law) was deduced as the complementary part to 1. In order to take into account the possible change in properties as the sample is degrading, measurements were repeated on virgin samples and on similar samples submitted to degradation tests for various incident flux and various durations. After degradation, samples were cooled and characterized with the spectrometer (no in situ measurements during degradation were possible here and it must be assumed that the measured emissivity also holds for the same sample at high temperature, during its exposure in front of the cone calorimeter). The measurements were described in detail in a recent paper [17]. The main results are presented here in terms of spectral absorptivity. Fig. 8 shows three selected curves for the virgin material and for two samples after significant duration of
Fig. 6. IR image at 3500 cm1, converted in spectral intensity. Incident flux 50 kW/m2, time t ¼ 80 s.
Z. Acem et al. / International Journal of Thermal Sciences 112 (2017) 427e438
431
2564 cm1 which was widely used in the followings, the average value (considering all the measurements performed for various fluxes and degradation durations) was finally evaluated to 0.91. For the analysis of the spectra which requires the emissivity over the whole wavenumber range, the exact curves for the properties after degradation were used. A specific sensitivity study was conducted on the influence of the emissivity on the surface temperature identification (see the corresponding section below), showing that the temperature was not strongly affected by slight variations or errors in the emissivity. 3. Model for temperature identification 3.1. Radiation model The basic idea was of course to link the temperature and the intensity emitted by the surface, using the Planck's law [18] for a black surface: Fig. 7. IR image at 2353 cm1, converted in spectral intensity. Incident flux 50 kW/m2, time t ¼ 80 s.
Ibh ¼
1 C1 h3 p eC2T h 1
(1)
where C1 ¼ 3.7418$1016 W m2, C2 ¼ 1.4388$102 m K and h stands for the wavenumber in m1. The measured intensity was then expected to allow identifying the surface temperature, simply inverting relationship (1). However, as the received flux is in fact a combination of the flux emitted by the surface, plus the reflection of the radiation coming from the cone calorimeter, the measured intensity corresponds to:
Ih ¼ εh Ibh ðTðtÞÞ þ rh Ibh ðTcone Þ
Fig. 8. Absorptivity of the composite (virgin sample and degraded sample under 35 and 65 kW/m2).
thermal degradation under incident fluxes equal to 35 and 65 kW/ m2 (as a whole, more than 20 samples were studied in order to investigate the effect of the incident flux and of the degradation duration). These cases were chosen for the present illustration because they represent two different situations: i.e. degradation without auto-ignition for the lowest flux and with ignition for the highest one. As can be seen, the absorptivity of the virgin material is relatively high (close to 0.88 in average), with a near grey behavior considering that spectral variations are kept in a narrow range between 0.86 and 0.91. The thermal degradation induces some changes, with a slightly higher average value (around 0.90) and a larger variation range between 0.84 and 0.95. These are still moderate changes, however. The influence of the incident heat flux is also weak, as detailed in Ref. [17]. At the specific wavenumber of
(2)
with T(t) standing for the surface temperature, Tcone for the cone temperature, εh is the spectral emissivity and rh ¼ 1εh is the surface reflectivity. This assumes a Lambertian irradiation from the cone (which emission is close to the one of a blackbody, as demonstrated in Ref. [13]) and a diffuse reflection for the sample. This reflection part is even the major contributor to the radiation received from the surface during the first instants of the experiment (note that radiation from the surroundings at ambient temperature is clearly negligible as compared to the cone radiation, it is not included in this relationship). Reflection cannot be simply removed after a rough evaluation of the incident flux, because any uncertainty in this value or in the reflectivity would result in systematic errors in the temperature. This error would be particularly strong during the first step of the sample heating, when the surface temperature is still moderate, because the main part of the received signal is due to the reflection and the “useful” part of the signal is weak. Considering that one of the main objectives of the present study was to provide temperature data as accurate as possible, especially during the heating step, for degradation model validation purpose, a fine correction must be introduced. In the second step of the temperature evolution, the emission becomes more and more influent in the global signal as the surface temperature increases. Then, the uncertainty due to the reflection decreases. This will be confirmed by section 4.2, when comparing our results with temperatures obtained thanks to usual methods of IR image processing. Finally, after ignition along the sample surface, the flame also provides an additional flux directly emitted toward the spectrometer or reflected by the sample surface, which must be taken into account. For the present analysis, these parasitic reflections were accounted for in the signal processing thanks to a subtraction method, which was applied on the IR images at 2564 cm1 and on the spectra. It was assumed that no significant change occurred in
432
Z. Acem et al. / International Journal of Thermal Sciences 112 (2017) 427e438
the incident radiation (and thus in the reflected intensity) during the experiment, since the main part was due to the cone calorimeter which was in stationary regime. The subtraction eliminated the reflection contribution and the variation in intensity was correlated to the temperature increase. The method can be summarized as follows: i. Evaluation of the instantaneous intensity from the direct signal acquisition thanks to the calibration data obtained with the high temperature blackbody ii. Calculation of the difference between the measured instantaneous intensity and the reference value at initial time (meaning the subtraction of the reflection from the cone): DIh0 ¼ Ih(t)Ih(t ¼ 0) iii. Evaluation of the temperature T(t) from an inversion based on the Planck's law applied to the intensities involved in step (ii):
131
0
C2h 1 6 B pDIh0 C7 TðtÞz4 ln@ þ eTðt¼0Þ A5 3 hC2 εh C1 h
(3)
The temperature evaluation was initialized at T(t ¼ 0 s) ¼ 293 K. The possible drawback of this method is that the reflection subtraction assumes no change during all the experiment, which is a questionable hypothesis especially when the flame appears along the sample, or because the emissivity was seen to slightly vary with time. Therefore a second method was tested, replacing step (ii) by a subtraction between two successive images, such that the reflection part was only assumed constant between two acquisition times (down to 1 s or 0.2 s in some cases for the camera, around 3 s for the spectrometer). The method becomes: i. Evaluation of the intensity from the direct signal acquisition thanks to the calibration data obtained with the high temperature blackbody ii. Calculation of the intensity difference between two successive acquisitions DIh ¼ Ih(t)Ih(tDt) iii. Evaluation of the temperature T(t) from the inversion of the Planck's law for radiation:
2
131
0
1 6 B pDIh TðtÞz4 ln@ þe hC2 εh C1 h3
C Tðt2hDtÞ
C7 A5
800
(4)
Initialization was still done with T(t ¼ 0 s) ¼ 293 K. The reflection correction is better taken into account with this second method, but any error in a temperature evaluation at time t is kept and cumulated with other possible errors during all the experiment, which could cause some deviations for the final result. Actually, both methods gave the same results in our case and the two devices (spectrometer and camera) also gave the same temperature evolutions (within a reasonable uncertainty range) as it will be seen later in the result section. This shows that the reflection correction is well controlled and thus gives some confidence in the results.
700
Temperature [K]
2
according toh a representative exponential law such as i t TðtÞ ¼ Tð0Þ 1 e =t , with T(0) ¼ 293 K and t ¼ 100 s. This relationship was chosen as it looks like the temperature increase observed in present degradation tests, as will be seen later. Then, the theoretical intensity sent by the surface was predicted using relationship (2), considering an emissivity close to the actual value (εh ¼ 0.90) and diffuse reflection (with reflectivity rh ¼ 0.10) of an incident intensity coming from a blackbody at temperature Tcone ¼ 992 K (the set temperature of the cone for an incident flux of 35 kW/m2 in our experiments). This “synthetic” intensity was then used for a mock test of temperature evaluation from intensity data. This test is a kind of inverse crime, since the same model is used in a forward and a reverse application for the data generation and the identification. Therefore, it cannot be considered as a full validation. However, it allows investigating the sensitivity of the results to a hypothetic error in the emissivity (value 0.85 tested), in the initialization (error tested in initial temperature 303 K instead of 293 K), or due to a noisy acquisition (intensity acquired with a random noise up to 5% of the signal). Results are presented in Figs. 9 and 10 for the different temperature evaluations. Only the results based on the subtraction of the data obtained at two successive steps are presented, as no difference could be seen between the results based on relationships (3) or (4). All curves (basic evaluation and data disturbed by error in initial temperature, emissivity or noise) seemed to overlap and there was no significant discrepancy or uncertainty visible with this temperature scale (consequently, only the exact “synthetic” temperature and the identification in nominal conditions with the exact emissivity and initial temperature are shown in Fig. 9 for the sake of visibility). Discrepancies in the evaluation appear when the absolute error (true temperature minus identified temperature) is plotted as a function of time (Fig. 10). Evaluation at the first instants may present uncertainties if initial conditions are not well known, but results converge toward the same values for larger times, which are very close to the exact result if the emissivity is well known. In this case, long time evaluation can be taken with confidence, error being less than 1.5 K for an absolute temperature of 800 K. If an error occurs in the emissivity evaluation, the temperature may shift from the exact value, but the discrepancy is still in a range of 10 K here, as
600
500 Exact Identified 400
3.2. Temperature identification and sensitivity analysis Before using the method seen in the previous paragraph on real experimental data, a numerical test was carried out first, in order to check our ability to identify a given temperature evolution in an accurate manner. A typical evolution of the temperature was considered during 300 s in the range between 293 K and 800 K
300
0
50
100
150
200
250
300
Time [s] Fig. 9. Numerical test on the surface temperature evaluation. Exact temperature (black solid line) compared with identified data (triangles).
Z. Acem et al. / International Journal of Thermal Sciences 112 (2017) 427e438
an experiment conducted during 50 s. Higher instantaneous peaks were observed, but they were rather interpreted as noise and the error fluctuated in a range between 0 and 16 K for 90% of the measurement step. We even think that this provides an upper limit of the error since the present measurement was done throughout a fully developed flame, while the flame along the sample during degradation tests is not perfectly covering the measurement area in our experiments. However, this possible error, around 7 K, must be kept in mind in the analysis presented in the next section.
1 0 -1 -2
Absolute error [K]
-3 -4 -5 -6
4. Results
-7
4.1. First experimental campaign
-8 Identified temperature
-9
Error by 10 K in the initial temp.
-10
Emissivity set to 0.85
-11 -12
433
Random noise (5%) in the intensity difference 0
50
100
150
200
250
300
Time [s] Fig. 10. Absolute error between exact and identified data, with supplementary data obtained when introducing an error by 10 K in the initial temperature, or an error in the emissivity (0.85 instead of 0.90) or random noise in the input data (±5% on the instantaneous intensity computed from reference data).
compared to the 800 K of the exact value. As a consequence, present results are obtained with moderate or weak uncertainty. This is especially due to the fact that the possible error due to reflection is well controlled thanks to the image subtraction process. Without such a method, approximation in the reflection evaluation would result in dramatic uncertainties during the first instants. 3.3. Measurement bias due to the flame The optical filter avoids any possibility of measurement error due to radiation emission by the combustion gases, but soot may still affect the radiation measurement, inducing a possible temperature overestimation when auto-ignition occurs. Despite the fact that the flame was supposed to be optically thin (low soot concentration and flame thickness observed to be of the order of one centimeter, hence resulting in a weak influence) the related effect was evaluated through a dedicated calibration experiment. The idea was to generate a realistic flame in front of a surface with well known temperature and to evaluate this temperature with the above-presented methods. For that, an extended blackbody ECN100 by HGH Infrared Systems (30 cm 35 cm) was used, with surface temperature set to 823 K. A carbon composite sample was set along the path line, in vertical position, between the camera and the blackbody and it was ignited using a burner. After sufficient heating, the measurement started when the burner was withdrawn, the flame being maintained by the sample combustion itself. A comparison was done with and without perturbation by the flame. The chosen sample was one 100 mm by 100 mm, 5 mm thick, carbon composite piece of the same series than the one studied in the next section (in order to expect a flame similar to the one encountered during the auto-ignition step, even if the heating step was not the same). Then, the blackbody was observed throughout the flame, just above the composite sample, and its surface temperature was evaluated in the conditions of the present study using both methods of image subtraction. Results show a systematic overestimation of the surface temperature, because soot induce a supplementary intensity which is interpreted as a higher emission level from the surface. The error was evaluated to 7 K in average, for
As above-explained, the present method was used for the study of carbon composite samples under incident fluxes equal to 20, 35, 50 and 65 kW/m2. A first campaign was conducted on samples with surface 100 mm 100 mm, 5 mm thick. The corresponding temperature evolutions with time are presented in Figs. 11 and 12. Fig. 11 gives an idea of the confidence that could be given to the results for two chosen fluxes (20 and 50 kW/m2). First of all, tests were repeated twice in order to check the repeatability (see the two sets of data per fluxes very close one from the other). Then, the two above-discussed methods, namely reflection subtraction based on two successive images or using the initial image as a reference, were compared (triangles vs continuous lines) with no apparent difference. Finally, the two devices (spectrometer and camera) were also compared, providing the same results (crosses for the spectrometer, lines and triangles for the camera). These tests reported in Fig. 11 give an idea of the reliability of the whole experimental campaign. Then, Fig. 12 allows comparing the sample evolution as a function of the incident flux, still presenting two tests in each case with the same incident flux. All curves show a sharp increase in temperature (several hundreds of degrees during the first minute) and a plateau always reached between 300 s and 400 s as a consequence of an energy balance for the sample. The highest the incident flux is, the sharpest the temperature increase
Fig. 11. Time evolution of the surface temperature for two incident fluxes (20 and 50 kW/m2). Continuous lines and triangles for the camera data, crosses for the spectrometer data. Two sets of results for each flux (repeatability test).
Z. Acem et al. / International Journal of Thermal Sciences 112 (2017) 427e438
Fig. 12. Time evolution of the surface temperature for various incident fluxes (camera data). One particular curve at 35 kW/m2 was obtained using piloted ignition after 100 s of degradation (shifted from the two other curves between 100 and 300 s because of the flame contribution).
is. Some fluctuations appeared for the highest fluxes after 100 s as a consequence of ignition at the sample surface, which provided a supplementary incident flux for the degradation. It did not happen in standard conditions for the incident fluxes of 20 and 35 kW/m2, but it was forced using a piloted ignition device (a spark igniter) in one special test at 35 kW/m2. This explains why one of the three curves for 35 kW/m2 indicates a higher temperature level as compared to the two others between 100 and 300 s. This further increase in temperature is due to the radiative feedback from the flame. The gap between this particular curve and the two others shows the influence of the flame developing along the sample. When the flame disappeared after 300 s, the curve tended toward the same plateau than the two others. This was confirmed by both apparatuses with independent measurements. These data were found useful for people working on thermal degradation models, but the sharp increase during the first step showed that further experiments should be conducted decreasing the acquisition time, which is discussed in one of the next sections. 4.2. Comparison of the method with standard temperature evaluations based on IR images Our method based on image subtraction was especially developed because of the expected difficulty to evaluate the part of the radiative flux, due to the reflection by the sample, of the incident flux emitted by the cone. This contribution is dominating the measured radiative flux during the first step of the sample heating. In order to compare our results with data obtained with other usual methods, we conducted dedicated evaluations of the surface temperature, considering what could be a basic use of the camera images: (1) a basic straightforward evaluation of the equivalent blackbody temperature simply inverting the Planck's law given by relationship (1) (providing the temperature of a blackbody emitting the same radiative flux than the sample), (2) a corrected temperature involving the radiative properties of the sample: emissivity 0.91 and reflectivity 0.09, assuming that the incident flux from the cone is the value measured by a heat flux gauge located at the
sample position, with the spectral distribution given by the Planck's law for a blackbody at the temperature set for the cone (measured with three thermocouples on the coil). The case of the 35 kW/m2 incident flux with piloted ignition was chosen for an illustration of the discrepancy between these basic evaluations and our method. Fig. 13 presents the different temperature estimates and Fig. 14 evaluates the discrepancy between the different methods when compared with our evaluation (temperature obtained with one of the usual evaluations minus temperature according to our method). The equivalent blackbody temperature strongly overestimates the true temperature during the first step of heating (discrepancy close to 300 K at the beginning of the test), since the sample is still cold whereas the reflection from the cone is interpreted as if it was proper emission. The gap decreases because true emission by the sample progressively becomes the dominating contribution to the radiative flux. When the stationary step is reached, the overestimation is between 5 and 10 K. The effect is limited because the cone temperature is not really far from the sample one, the gap could be larger in different conditions of irradiation. The temperature is still overestimated because the cone has a higher temperature (992 K) than the sample. The second evaluation which aims at considering the flux from the cone reflected by the sample is better performing, but the first step of heating is still approximated because the correction cannot be perfect. The gap is close to 40 K at the beginning, finally decreasing to a few kelvins during the stationary step. One can also note some fluctuations during the flaming period (mainly in the range between ± 50 K for times ranging between 150 s and 250 s). Note that the discrepancy at the beginning of the test could be higher for larger incident fluxes. The accuracy of the correction for the reflected flux strongly depends on the ability to measure the sample emissivity and the cone temperature (any error would result in a rise in discrepancy), while our method is little dependent on the emissivity and does not depend at all on any reflection evaluation which is directly eliminated with the subtraction process.
1000 -2
65 kW.m 900
800
Temperature [K]
434
-2
35 kW.m
700
600
500
400
300
Present method Equivalent blackbody temperature Corrected temperature involving the cone radiation
0
200
400
600
Time [s] Fig. 13. Temperature evaluation with the present method and using two basic evaluations: an equivalent blackbody temperature and a corrected estimate involving the reflection of the radiation from the cone.
Z. Acem et al. / International Journal of Thermal Sciences 112 (2017) 427e438
435
800
Temperature [K]
700
600
Test 1 500
Test 2 Test 3 Test 4
400
Test 5 Test 6
300
0
50
100
150
200
Time [s]
Fig. 14. Absolute discrepancy between the basic evaluations of Fig. 13 and our method.
4.3. Second experimental campaign A second experimental campaign was conducted on samples with a surface 50 mm 300 mm, 5 mm thick (this particular sample size was designed to allow a simultaneous study of thermal degradation and structure modification under mechanical stress, results are only commented regarding heat transfer here). Despite an incident flux equal to 35 kW/m2, auto-ignition was systematically observed here, whereas this flux was not high enough to allow auto-ignition in the previous series. This can be explained by the difference in size for the samples, the use of a different sampleholder which can affect the heat balance for the sample, or other phenomena influencing the concentration of pyrolysis gases (which is close to the inflammability limit in both cases but can be below or above the lower limit depending on the exact set up configuration). Auto-ignition process is not yet satisfactorily described and is beyond the scope of the present study rather focused on the surface temperature evaluation. In this second campaign, the studied case was a sample progressively heated, until auto-ignition appeared along the surface. The same method of IR imaging was applied, using the IR camera with the filter at 2564 cm1 and applying the image subtraction method for data processing (subtraction of two successive images following relation (4)). The time step of 0.2 s between two images allowed expecting a better description of the transient heating of the sample. The test was focused on the first step of heating (between 100 and 200 s after the beginning of the irradiation by the cone). Fig. 15 shows the time evolution of the sample temperature (average value in the center of the sample as for the first experimental campaign). Six tests are presented, showing the quite good repeatability of the experiments, especially during the first 100 s. The same trends were observed but the auto-ignition did not occur exactly at the same instant. It appeared through the temperature jump in the different tests, between 85 and 120 s, always when the sample surface temperature reached a threshold value around 650 K. Such composite samples made of carbon fibers and epoxy resin present some heterogeneities in their structure and contents, explaining these discrepancies, which are still moderate here. Moreover, as above-discussed, this case is probably close
Fig. 15. Surface temperature under a flux of 35 kW/m2, second experimental campaign, six repeatability tests.
to the conditions limiting the two behaviors with or without autoignition, also explaining the discrepancies. The method itself showed its ability to capture the sharp temperature increase with a quite good repeatability and the short time step used for the image acquisition allowed tracking the temperature with a finer resolution. 4.4. Evaluation of the heat flux induced by auto-ignition Beside the observation of the degradation mechanisms and the measurement of the surface temperature, present data were further used to evaluate the additional incident flux due to the flame, which affects the sample in addition to the initial radiation from the cone apparatus. This is a crucial problem as the standard use of the cone calorimeter for the characterization of materials considers the heat flux calibrated at the beginning of the experiment as the only incident flux, while it is obvious that the autoignition produces an additional flux toward the sample due to radiative feedback and convection between the flame and the surface. Some attempts were reported in the past to measure this flux with heat flux gauges embedded in the sample (see Refs. [19e22] for example) but this is a tricky measurement in a harsh environment. The values found in the literature are clearly depending on the sample type, its orientation, its size and the incident flux. Most of the presented values are in a range between 10 and 20 kW/m2, with some higher fluxes reported for PMMA in Ref. [22] for example, but for larger samples. We started here with the observation that a significant rise of the surface temperature occurs due to the auto-ignition of the pyrolysis gases, which is actually due to the heat provided by the flame to the surface. This led us to the conclusion that a model for the temperature evolution at the sample surface can provide the required supplementary flux in order to explain the temperature gap. This was the guiding idea of our investigation. Recalling that an overestimation of the temperature is suspected when the flame is developed, the present test must be considered as a feasibility test and a quantitative confirmation will be required in a future work. In a first step, a simple 1D transient conduction model was implemented for the sample, considering the ambient temperature
Z. Acem et al. / International Journal of Thermal Sciences 112 (2017) 427e438
as the initial condition for the whole domain, with boundary conditions accounting for the incident flux measured from the heat flux gauge during the calibration step, plus radiative and convective exchanges on both faces. Let us consider a medium with thickness E. The face submitted to the incident radiation is at position x ¼ E, the other face (at x ¼ 0) being only submitted to radiative and convective exchanges with ambient air. The governing equation for temperature prediction inside the medium is the following 1D transient balance energy for a solid opaque medium:
rC
vT v2 T ¼ l 2 þ q_ vt vx
(5)
where r, C and l stand for the thermo-physical properties of the medium (density, heat capacity and thermal conductivity) and q_ is a sink term aimed at describing the pyrolysis process and the corresponding latent heat). Actually, the mass loss rate was not available for the present tests and the term q_ was not included in the model. This is of course an obvious shortcoming of the present analysis. Note that the present analysis is restricted to the first three minutes of the degradation test. Hence, the actual mass loss and the corresponding source term are expected to be moderate. Moreover, the emphasis is put on the required supplementary flux in order to explain the temperature jump at the surface, which is expected to be affected by the flame feedback at the surface above all. Boundary conditions involve external convection and radiation as follows:
vT Irradiated surface ðx ¼ EÞ : l ¼hE ðTðEÞ T∞E Þ vx E 4 aqinc þ εs TðEÞ4 T∞E (6) vT Back surface ðx ¼ 0Þ : þl ¼h0 ðTð0Þ T∞0 Þ vx 0 4 þ εs Tð0Þ4 T∞0
(7)
where qinc is the incident flux to the sample, a is the absorptivity assumed to be equal to the emissivity ε, h0 and hE stand for the convective heat transfer coefficient at the two boundaries and T∞E or T∞0 stand for the temperature of the surrounding medium. At the beginning of the experiment, qinc is equal to the incident flux from the cone (qinc,0). After auto-ignition, an additional flux (qflame) is affecting the sample and qinc ¼ qinc,0 þ qflame. This latter value qflame was sought by our numerical routine. A value of 5 W/(m2 K) was set for the heat transfer coefficients (corresponding to a typical free convection situation). This is a choice guided by a standard evaluation of the Nusselt number for such a configuration. Some sensitivity tests varying this value showed very weak influence on the evaluation of the heat flux from the flame, as the order of magnitude of the incident flux and of the flux from the flame is obviously larger than convection at the boundaries. The emissivity was set to 0.91. A straightforward numerical scheme using a Finite Volume Method was used to solve the problem (not detailed here). One supplementary difficulty was the lack of knowledge of the thermal properties. Measurements raised problems because this composite material is heterogeneous and because properties may change along the degradation process. The chosen solution was to evaluate the equivalent average properties by optimization, identifying the set of thermal properties which give the best fit of the measured temperature evolution during the time period before auto-ignition. Then, the simulation was conducted further, with these properties as input data, in order
to search for the additional flux which explains the time evolution of the temperature after auto-ignition. This simulation in two steps was therefore carried out as follows: (i) A first step of simulation during the first 80 s in order to evaluate the thermal conductivity l and the product rC (density heat capacity) which allows the best reconstruction of the temperature, using a particle swarm optimization algorithm (see Refs. [23,24] for an introduction to the method). Main numerical parameters were set to 50 particles and 50 loops (with a control of the convergence thanks to a cost function based on a second order norm evaluating the discrepancy between experimental and reconstructed data). The search space was set to [5.105e5.106 J K1 m3] for the product rC and [0.1e10 W m1 K1] for the thermal conductivity. (ii) A second step of simulation was carried out between 80 s and the end of the experiment, keeping the thermal properties identified in step (i) and searching for the additional flux required to explain the time-evolution of the temperature. A dichotomy scheme was used at each time step to track this flux corresponding to the auto-ignition influence. The convergence criterion was set to 0.1 K between the experimental and reconstructed temperatures. This method was applied on the six tests shown in Fig. 15. The step (i) provided average thermal properties as follows: 2.50 106 J K1 m3 for the product rC and 0.4 W m1 K1 for l. These are equivalent properties, which are in a very good agreement with what was predicted by Hidalgo et al. [5] who suggested an average value of 1.035 106 USI for the inertia product l r C on similar samples, based on the analysis of cone calorimeter tests. Fig. 16 shows a typical temperature reconstruction based on the two steps above-presented for test 6 (similar results were obtained for the other tests). The discrepancy between numerical and experimental data is below 6 K during the optimization step. This is a quite good agreement considering that identified properties are equivalent average data, kept constant during the process. Then, the discrepancy never exceeds 0.1 K in the second step, since this is the convergence criterion used in the dichotomy scheme.
800 Experimental Numerical 700
Temperature [K]
436
600
500
400
300
0
50
100
150
200
Time [s] Fig. 16. Temperature reconstruction in Test 6 (dotted line) versus experimental data (continuous line).
Z. Acem et al. / International Journal of Thermal Sciences 112 (2017) 427e438
437
the reflection of the incident radiation by the sample. This nonintrusive method gave the time varying evolution of the temperature during a series of degradation experiments for incident fluxes between 20 and 65 kW/m2. Both apparatuses independently provided results showing the same sharp temperature increase when the sample was irradiated, before reaching a plateau. Incident fluxes above 35 kW/m2 systematically induced a self-ignition of the pyrolysis gases after significant sample degradation. The temperature levels were observed to increase with the incident flux, between 740 K at 20 kW/m2 and 1100 K at 65 kW/m2. The flame produced by the auto-ignition itself was observed to result in a temperature jump around 200 K. Further analysis of the sample temperature in case of autoignition of the sample allowed the identification of the heat feedback from the flame. Peak values in the range between 15 and 20 kW/m2 were found for samples with surface 50 mm 300 mm, 5 mm thick. This supplementary flux is obviously affected by the sample size and the experimental conditions, but the method seems to be promising and the identification model could be refined for a systematic application. Acknowledgements Fig. 17. Evaluation of the additional flux due to auto-ignition for the 6 cases presented in Fig. 10.
Regarding the evaluation of the additional flux explaining the temperature jump due to the flame, Fig. 17 presents the flux result in all cases. A sudden peak occurs at the ignition instant (a significant concentration of pyrolysis gas is available and quickly burns), then a decrease is observed. We chose to stop this evaluation after a reasonably short time because the mass loss effect is not involved (as above-discussed) and also because constant equivalent thermal properties are assumed. This latter assumption could not be kept after significant degradation, as the delamination of the sample involves strong heterogeneities and modifications of the properties. As can be seen in Fig. 15, the duration of these experiments was quite short after ignition, and further comparison was not allowed anyway. There is some variability in the flux evaluation, which is logical when considering the temperature curves in Fig. 15. They are not perfectly super-imposed and the sudden rise in temperature due to the auto-ignition does not occur exactly at the same time. Consequently, the additional fluxes presented in Fig. 17 have the same global shape and the same order of magnitude but with a significant variability. Despite the well-controlled operating conditions under the cone, there are some variations in the sample composition degradation, the degradation process itself, the ignition conditions (local concentration and temperature) which may explain this observation. At this stage, the order of magnitude of an additional flux with a peak around 15e20 kW/m2 seems to be confirmed when the auto-ignition occurs. However, the abovediscussed warnings regarding the accuracy of the temperature measurement must be kept in mind. The method itself is promising and further tests and developments are encouraged, in particular refining the experiment, extending the model with the pyrolysis and mass loss effects, addressing the property modifications and perhaps the structure evolution of the sample. 5. Conclusion The surface temperature of a carbon composite sample was identified during its degradation under a cone calorimeter, based on infrared observation with an IR camera and a FTIR spectrometer simultaneously. A special subtraction process was applied on spectra and images in order to eliminate the parasitic signal due to
The research leading to these results has received funding from the European Union's Seventh Framework Programme (FP7/ 2007e2013) for the Fuel Cells and Hydrogen Joint Technology Initiative under grant agreement n 325329. This work also pertains to the French Government program “Investissements d’Avenir” (LABEX INTERACTIFS, reference ANR11-LABX-0017-01). References [1] ISO 12136:2011. Reaction to fire tests e measurements of material properties using a fire propagation apparatus. 2011. [2] ISO 5660. Reaction to fire tests - heat release, smoke production and mass loss rate - Part 1: heat release rate (cone calorimeter method). 2015. [3] Stoliarov SI, Crawley S, Lyon RE, Linteris GT. Predictions of the burning rates of non-charring polymers. Combust Flame 2009;156:1068e83. [4] Kacem A, Mense M, Pizzo Y, Boyer G, Suard S, Boulet P, Parent G, Porterie B. A fully coupled fluid/solid model for open air combustion of horizontallyoriented PMMA samples. Combust Flame 2016;170:135e47. [5] Hidalgo JP, Pironi P, Hadden RM, Welch S. A Framework for evaluating the thermal behaviour of carbon fibre composite materials. In: Proc.2nd european symp. Fire safety science; 2015. [6] Li J, Gong J, Stoliarov SI. Gasification experiments for pyrolysis model parametrization and validation. Int J Heat Mass Transf 2014;77:738e44. [7] Li J, Gong J, Stoliarov SI. Development of pyrolysis models for charring polymers. Polym Degrad Stab 2015;115:138e52. [8] De Vries J, Ren N, Chaos M. Temperature measurements on solid surfaces in rack-storage fires using IR thermography. In: Proceedings of SPIE - the international society for optical engineering, 9485, 94850H; 2015. [9] Melendez J, Foronda A, Aranda JM, Lopez F, Lopez del Cerro FJ. Infrared thermography of solid surfaces in a fire. Meas Sci Technol 2010;21. 105504 (10pp). [10] Boulet P, Parent G, Collin A, Acem Z, Porterie B, Clerc JP, et al. Spectral emission of flames from laboratory-scale vegetation fires. Int J Wildland Fire 2009;18(7):875e84. [11] Boulet P, Parent G, Acem Z, Kaiss A, Billaud Y, Porterie P, et al. Experimental investigation of radiation emitted by optically thin to optically thick wildland flames. J Combust 2011;2011. Article ID 137437. € rsth M, Bal N, et al. Radiation emission [12] Boulet P, Parent G, Acem Z, Collin A, Fo from a heating coil or a halogen lamp on a semitransparent sample. Int J Therm Sci 2014;77:223e32. [13] Boulet P, Parent G, Acem Z, Rogaume T, Fateh T, Zaida J, et al. Characterization of the radiative exchanges when using a cone calorimeter for the study of the plywood pyrolysis. Fire Saf J 2012;51:53e60. [14] Billaud Y, Boulet P, Pizzo Y, Parent G, Acem Z, Kaiss A, et al. Determination of woody fuel flame properties by means of emission spectroscopy using a genetic algorithm. Comb Sci Technol 2012;185(4):579e99. [15] Pouplin J, Collin A, Acem Z, Parent G, Boulet P, Vena P, et al. Study of a V-shape flame based on IR spectroscopy and IR imaging. J Phys Conf Ser 2016;676: 012018. [16] Batiot B, Quach THY, Acem Z, Brissinger D, Collin A, Richard F, et al. Correlation between degradation of an epoxy-carbon fiber material and surface radiation
438
Z. Acem et al. / International Journal of Thermal Sciences 112 (2017) 427e438
during the thermal exposure. 2016. Submitted to Fire Tech. [17] Boulet P, Brissinger D, Collin A, Acem Z, Parent G. On the influence of the sample absorptivity when studying the thermal degradation of materials. Materials 2015;8:5398e413. [18] Modest MF. Radiative heat transfer. third ed. Academic Press; 2013. [19] Rhodes BT, Quintiere J. Burning rate and flame heat flux for PMMA in a cone calorimeter. Fire Saf J 1996;16(3):221e40. [20] Beaulieu PA, Dembsey NA. Flammability characteristics at applied heat flux levels up to 200 kW/m2. Fire Mater 2008;32:61e86. [21] Li K, Pau DSW, Wang J, Ji J. Modelling pyrolysis of charing materials: determining flame heat flux using bench-scale experiments of medium density
fiberboard. Chem Eng Sci 2015;123:39e48. [22] Kacem A, Mense M, Pizzo Y, Boyer G, Suard S, Boulet P, et al. A fully coupled fluid/solid model for open air combustion of horizontally-oriented PMMA samples. Combust Flame 2016;170:135e47. [23] Kennedy J, Eberhart RC. Particle swarm optimization. In: Proceedings of IEEE international conference on neural networks, Piscataway, NJ; 1995. p. 1942e8. [24] Shi Y, Eberhart RC. A modified particle swarm optimizer. In: Proceedings of the IEEE congress on evolutionary computation (CEC 1998), Piscataway, NJ; 1998. p. 69e73.