Use of inorganic materials to enhance thermal stability and flammability behavior of a polyimide

Polymer Degradation and Stability 96 (2011) 23e32

Contents lists available at ScienceDirect

Polymer Degradation and Stability journal homepage: www.elsevier.com/locate/polydegstab

Use of inorganic materials to enhance thermal stability and flammability behavior of a polyimide Alexander B. Morgan a, *, Sirina Putthanarat a, b a b

Multiscale Composites and Polymers Division, University of Dayton Research Institute, Dayton, OH 45469, USA Air Force Research Laboratory, AFRL/RXBC, Wright-Patterson AFB, OH 45433-7750, USA

a r t i c l e i n f o

a b s t r a c t

Article history: Received 18 August 2010 Received in revised form 28 October 2010 Accepted 9 November 2010 Available online 4 December 2010

While a great variety of high temperature polyimide materials exist, these materials are being subjected to higher and higher use temperatures in oxidative and environmentally aggressive environments. There is a limit to the extent one can take a polyimide before it will oxidize and subsequently suffer property degradation, thermal decomposition, and structural failure. Therefore, we instead sought to use materials which do not oxidize (inorganic materials) to enhance the polyimide composition and perhaps move the properties of the organic polymer more into the realm of ceramics while maintaining polyimide composite weights and processing advantages. In this paper we present results of the combination of inorganic micron sized particles with and without carbon nanofibers to produce a variety of highly inorganic particle filled polyimides. These polyimides were tested for thermal stability and flammability in resin pellet form and as a protective coating for a carbonefiber composite structure. Our results demonstrate that the resin with inorganic particles exhibited significant reductions in flammability by themselves, but minimal flammability reduction when used as a thin coating to protect a carbonefiber composite. Further, the gains in thermal stability are limited by the thermal stability of the polyimide matrix, suggesting that more work is needed in measuring the limits of inorganic fillers to improve thermal stability. Still, the results are promising and may yield polyimide systems useful for providing resistance to damage from high heat flux exposures/fire risk scenarios. Ó 2010 Elsevier Ltd. All rights reserved.

Keywords: Polyimides Thermal decomposition Flammability Inorganic nanoparticles

1. Introduction One of the most common material replacements in aircraft structures today is the replacement of metal or ceramic with a fiber reinforced polymeric composite. This is primarily done to achieve weight savings for improved fuel efficiency/flying range for an aircraft, but sometimes is also put in place to avoid corrosion issues encountered with metals or to yield complex structures which cannot be easily made from ceramic. However, the insertion of polymeric materials into places where metals and ceramics had originally been used means that these polymers must survive extremes of temperature (heat, flame). Therefore there is an increasing desire to combine the most advantageous features of metals and ceramics into a polymeric material and create a hybrid of the two materials. In this paper we discuss recent developments in polyimide þ ceramic hybrids to provide enhanced thermal stability.

* Corresponding author. E-mail address: [email protected] (A.B. Morgan). 0141-3910/$ e see front matter Ó 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.polymdegradstab.2010.11.005

When considering which inorganic/ceramic filler to use in a polyimide composite, some background is needed to explain the choices that were made in this study as well as what we hoped to achieve. While polyimides can be used to replace metals and ceramics, they still suffer from thermal oxidative damage which causes the polyimide to slowly ablate away resulting in microcrack and structural failure of the composites [1e10]. Since polyimides will always oxidize when exposed to enough heat and oxygen, the best protective scheme for this material is to either help assist the thermally decomposing polymer form into glassy carbon char [11e13], or use materials which cannot oxidize further to form a ceramic shield in situ as the polymer decomposes [14e20]. Since polyimides are pre-disposed to form carbon chars upon exposure to high levels of heat and/or flame, we added finely dispersed inorganic particles in the polyimide matrix to improve thermal oxidative stability. The reason for this approach comes from the polymer nanocomposite literature, where the use of finely dispersed nanoscale inorganic particles (such as montmorillonite clays, aluminas, other metal oxides) delays the onset of thermal decomposition in a polymer and slows down the mass loss rate of the pyrolyzing polymer once thermal decomposition temperatures are reached

24

A.B. Morgan, S. Putthanarat / Polymer Degradation and Stability 96 (2011) 23e32

[21e24]. Nanomaterials such as vapor grown carbon nanofibers (CNF), multi-wall carbon nanotubes (MWNT), or exfoliated graphite flake (EXGF) were focused on as these nanomaterials have shown significant reductions in mass loss rate under polymer decomposition temperatures [25e27]. These nanomaterials may also provide thermal and electrical conductivity enhancements to the composite for multifunctional performance [28e32]. However these commonly studied nanomaterials at their typical loading levels (0.1e10 wt%) cannot provide all of the needed thermal protection by themselves and so the chemistry of ceramics must be considered as well. When it comes to thermal protection provided by ceramics the two chemistries most appropriate are silica (SiO2) and alumina (Al2O3), both of which are widely and commercially available. These materials are available in a wide range of forms, especially as very fine (small size) particles appropriate for mixing into a fiber reinforced composite. If the inorganic particles are too big then they will filter out on the carbon fiber fabric during composite manufacture or worse, create defect sites for microcracking and mechanical failure. Therefore particle sizes of 5 mm or smaller are needed and silica and alumina are both available in this particle size. Fumed silica was chosen due to known effects of reducing mass loss and helping to form char in some polymer systems under thermal decomposition temperatures [15,18,19,33,34]. Alumina represents an inert filler and bulk ceramic shield, and so was chosen as a relatively inexpensive filler to do most of the thermal protection and work in a potentially synergistic manner with the other nanofillers described above. If alumina is calcined (heat treated to remove hydroxyls) and finely ground, then it should have very little effect on polyimide processing/resin viscosity since its typical surface chemistry (AleOH) that would interact with polyimides will have been minimized. This point is important since nanomaterials used to produce polymer nanocomposites often increase resin viscosity to the point that the material is difficult to process, [35e37] and it is essential in this project to maintain ease of processing/composite quality while providing thermal protection. So for this project a calcined alumina used to produce larger high performance ceramic parts was chosen as a potential filler for the formation of polyimide þ inorganic hybrids. To produce a polyimide þ inorganic hybrid with enhanced thermal durability, we focused on the development of a polymer formulation with more ceramic than polymer content. We achieved this by focusing on high loadings (>50 wt%) of inorganic material so that as the polymer decomposed, there would be high potential to form a ceramic rich layer which would protect the underlying (undamaged) material. This filler approach to producing a polymer þ inorganic hybrid is admittedly simplistic and avoids other known polymer þ inorganic hybrids such as sol-gel and ceramer (pre-ceramic monomers) chemistry, but this is done for two key reasons. The first reason is that co-reactive inorganic materials are very likely to interfere with polyimide polymerization chemistry and may in turn make thermal stability for the resulting composite worse, not better. The second reason is that combining fillers into a resin system is much more cost effective and easier to implement/ commercialize than development of new polymer chemistry. Interestingly for commercial polyimide systems, most fillers are actually cheaper per kilogram than the polyimides and so by pursuing a filler approach, we may improve polyimide composite thermal damage resistance properties while lowering cost at the same time. With all of these criteria in mind, we combined inorganic fillers with an aerospace grade resin transfer molding (RTM) polyimide to produce several different nanocomposites which were studied for thermal stability improvement and lowered flammability.

2. Experimental section 2.1. Materials Polyimide monomers (MVK-10) in methanol solution used in this report was a Resin Transfer Molding (RTM) grade material provided by Maverick Corporation (Blue Ash, OH). Carbon fiber used to make the composites was T650-35 3K 309NT, Fabric Style 998, purchased from Fabric Development, Quakertown, PA. Dry Ball Milled Low Soda Alumina (99.75% purity e lot# BL6662) was purchased from Baikowski Malakoff (http://www. baikowskimalakoff.com/) through Brenntag Specialties. Fumed silica was purchased from SigmaeAldrich. Vapor grown carbon nanofibers (PR-25-HHT) were provided by Applied Sciences Inc. (Cedarville, OH), and were heat treated at UDRI (UDRI Lot #181). Exfoliated graphitic flake was prepared by UDRI in the labs of Prof. Khalid Lafdi (UDRI Lot #219). Multi-wall carbon nanotubes (Nanocyl 7000) were obtained from Nanocyl S.A. (Belgium). All additives were dried in a vacuum oven at 100  C for 24 h before use unless indicated otherwise below. 2.2. Resin and additive mixing & filming The MVK-10 resin varnish was weighed into a container and the alumina added. This mixture was stirred manually with a spatula for 15 min. A 20 cm (8 in) 3-roll mill was pre-heated to 50  C. The resin mixture was then rolled on the mill for 4 passes thru a gap spacing of 3 mil. On the fourth pass the material was collected into a Teflon lined tray placed in a sealed bag and then frozen at 23  C. Films were made on an 18 inch wide roll film line. The fully dispersed resin was placed between the rolls and filmed onto release paper so that the film thickness was 9 mil. The rolls were heated to 48  C. The films were cut to the appropriate size, bagged and frozen at 23  C. 2.3. Polymer pellet processing MVK-10 was dried at 105  C under full vacuum until the solvent was removed (w4 h). The resulting material was ground into a powder using a mortar and pestle. This powder was imidized for 2 h at 232  C and allowed to cool. The sample was reground and placed in a pellet mold. The pellet mold was placed in a press and heated at 5  C/min to 232  C under minimal pressure. Pressure was applied at 232  C and the press was allowed to continue heating to 316  C and held for 3 h. The press was allowed to cool to room temperature and the mold removed. If the resin contained only alumina the pressure was minimal (kiss pressure), for other filled system with silica, nanofiber, etc. the pressure was 3.44 MPa (500 psi). Once the pellets were removed, they were wafer-saw cut into small pieces for thermomechanical analysis (TMA), as well as for other small scale thermal tests mentioned in this paper.

2.4. Composite layup and autoclave processing The composite panels were made one of two ways. The first way was made using dry fabric then inserting MVK-10 resin films between each ply with the nano-modified film being on the tool surface. The second way was made using MVK-10 prepreg with the nano-modified material also being on the tool side. In the second case excess resin was bleed off. The panels were bagged and cured in the autoclave using a typical MVK-10 autoclave cycle as recommended by the manufacturer.

A.B. Morgan, S. Putthanarat / Polymer Degradation and Stability 96 (2011) 23e32

2.5. Analytical techniques Thermogravimetric analysis (TGA) was collected in triplicate with a TA Instruments Q5000 IR instrument, under atmosphere of nitrogen at a heating rate of 20  C/min from room temperature to either 260  C or 316  C, followed by a 4 h isothermal hold. Thermomechanical analysis (TMA) was collected on pellet samples with a TA Instruments Q400 TMA, using a heating rate of 2  C/min from room temperature to 400  C. Density measurements were collected with a modified Mettler-Toledo Analytical Balance using a water immersion density apparatus. Flammability testing was conducted via microcombustion calorimetry (ASTM D7309-07, Method A) with a Govmark MCC-1 instrument or with the cone calorimeter (ASTM E1354-07) with an FTT Dual Cone Calorimeter. Cone calorimeter testing was done at 50 kW/m2 heat flux with 10  10 cm2 specimens; data from these specimens and their sample thicknesses are recorded in Table 3. For scanning electron microscopy (SEM), a Hitachi S-4800 HRSEM was used under an accelerating voltage of 15 kV and a working distance of 5.0 mm. All samples were cryo-fractured to give a fresh surface, upon which gold was sputtered on all samples placed on the SEM pedestal at 50 mTorr, 45 mA for 30 s using a Denton Vacuum Model Desk II Cold-Sputter/ Etch Unit to deposit a coating of approximately 100 Å on the sample surface. X-ray CT was performed on TGA pellet samples (6  6  1.5 mm) as well as small sections of unaged and aged composite panels (6  25  3 mm) using an X-Tek HMX160 CT system (Metris, Brighton, MI). A source voltage of 80 kV and a source current of 100 mA were used. The specimen was rotated over 360 with a step size of 0.2 . Averages of eight projection images (1000  1000 pixels) were collected at each position. X-ray images were reconstructed using CT Pro software version 2.0 (Metris, Brighton, MI). A stack of flat cross sections was obtained after the reconstruction. Image analysis was carried out using Fovea Pro version 4.0 (Reindeer Graphics, Asheville, NC). For optical microscopy, pieces of the cured panels were cut, potted, and polished for view on a Nikon Epiphot 200 Metallograph microscope equipped with a digital camera. Using computer software the coating thickness was measured. The interface between top coating and underlying composite for measuring coating thickness was determined visually. For thermal oxidative stability (TOS) testing, the specimens were cut into 2.5  2.5 cm2 and labeled. After vacuum drying at 115  C (240  F) for 24 h the specimens were cooled in a desiccator. The specimens were weighed using a four decimal place Mettler

25

balance. An aluminum tray was lined with glass cloth and the specimens placed on top. Another piece of glass cloth was place on top of the specimen. This was then placed into a pre-heated oven. The specimens were pulled out and weighed approximately once a day always being allowed to cool in a desiccator. 3. Results and discussion 3.1. Polymer formulation/pellet data Our initial work focused on studying the effects of alumina with fumed silica and/or carbon nanomaterials (exfoliated graphite flake or carbon nanotubes/nanofibers) to enhance the thermal durability of the base polyimide. We focused on the synthesis of seven formulations, outlined in Table 1 below. In this table we also show the measured glass transition temperature (Tg) and coefficient of thermal expansion (CTE) data as well as thermogravimetric analysis (TGA) done with isothermal holds to roughly mimic industry standard 260  C (500  F) and 316  C (600  F) thermal aging temperatures used for polyimide composites. The polyimide used here (MVK-10) is a resin transfer molding grade polyimide rated for prolonged use up to 260  C. That being said, if our inorganic filler hypothesis works, we should be able to increase the use temperature of this 260  C polyimide to higher temperatures. Starting with the density measurements, the measured densities are in reasonable agreement with the calculated densities with the exception of the fumed silica containing samples. There are two possible reasons for this discrepancy. The first is the method used to measure the density of the fumed silica powder was incorrect and this measurement yielded a density higher than it will be when actually wetted out by polymer. The method used here was simply filling a 10 ml graduated cylinder with the silica and weighing the cylinder for increase. The second is that the fumed silica, being a very low density material, does actually lower the density of the product dramatically as it fills up free volume in the final polyimide thus preventing the polymer chains from coming close to each other as they would in pure form, which in turn would lower the density for the entire formulation. At this time we are not sure which is the dominant reason for the differences in measured vs. calculated density but we believe it is due to our measurement method of the dry fumed silica which yielded an incorrect measurement result. Regarding the logic behind our choices of filler loading amounts, the carbon based nanofillers (nanofibers, nanotubes, exfoliated graphite) were chosen at typical loading levels where they are used

Table 1 Polyimide þ Nanoparticle formulations. Formulation ID

PI Hybrid 0 PI Hybrid 1 PI Hybrid 2 PI Hybrid 3 PI Hybrid 4 PI Hybrid 7 PI Hybrid 5 PI Hybrid 6

Formulation details

MVK-10 MVK-10 þ 60 wt% Al2O3 MVK-10 þ 50 wt% Al2O3 þ 10 wt% Fumed silica MVK-10 þ 40 wt% Al2O3 þ 10 wt% Fumed Silica þ 5 wt% Carbon nanofiber MVK-10 þ 50 wt% Al2O3 þ 5 wt% Exfoliated graphite MVK-10 þ 40 wt% Al2O3 þ 10 wt% Fumed silica þ 5 wt% Exfoliated graphite MVK-10 þ 50 wt% Al2O3 þ 0.5 wt% Carbon nanotubes MVK-10 þ 5 wt% Carbon nanofiber

Measured density (g/cm3)

Calculated density (g/cm3)

TMA CTE (microns/meter deg C)

Tg (deg C)

TGA (Air) wt% lost @ 260  C-4 h isothermal

wt% lost @ 316  C-4 h isothermal

1.24 2.01 1.57

1.26 1.90 1.88

56.90 41.04 38.83

263 259 259

0.22 0.14 0.12

0.76 0.39 2.31

1.64

1.92

44.39

267

0.17

1.08

1.86

1.82

42.00

262

0.13

0.28

1.65

1.81

44.23

256

0.12

0.93

1.79

1.79

44.13

269

0.12

0.38

1.22

1.24

72.72

256

0.32

0.68

26

A.B. Morgan, S. Putthanarat / Polymer Degradation and Stability 96 (2011) 23e32

Fig. 1. SEM of pellets of MVK-10 þ 60 wt% Alumina (left), MVK-10 þ 50 wt % Alumina þ 10 wt% Fumed Silica (Right).

today to provide enhancements in mechanical/thermal properties without increasing resin viscosity excessively. Likewise, the amount of fumed silica added followed the same guiding principle (what amount could be added to improve properties but not excessively drive up viscosity). The alumina loadings however were driven by what maximum loading could be put into the system, however as other nanofillers were added in, sometimes the concentration of alumina had to be decreased to ensure the entire system could still be processed. These numbers are not necessarily optimal, but are instead loading levels that seemed reasonable from screening level experiments. In regards to the CTE and Tg measurements, one can argue that as more fillers are added to the material which increases its effective density, one makes the nanocomposite/polymer hybrid start to behave more like the ceramic or other filler and less like the polymer. CTE and Tg were measured by TMA through thickness as well as in-plane, with some differences noted for the samples compared to the polyimide control. Samples with fumed silica tended to show larger decreases in CTE while samples with carbon nanofibers tended to show increases in CTE. For example, the sample with just nanofibers and polyimide (no other inorganic fillers) showed a substantial increase in CTE. While we do not believe that the pellet manufacture process caused any orientation in the sample, we cannot rule out that it occurred and affected the observed CTE and Tg data. However the Tg results may not be significant since it is the Tg of the fiber reinforced composite that is

most important for this application, not of the base resin in a solid thick pellet form. However, minimal decreases in Tg and lowering of CTE are good results for the application described in the Introduction. If these hybrid materials/polymer nanocomposites have lower CTE then they should help bridge the gap between ceramic heat shields and fiber reinforced composites without causing a ceramic heat shield to fall off due to CTE mismatch in actual use. The results in Table 1 that are most interesting and promising are in the TGA data. At 260  C, all of the samples are better than the control with the exception of the polyimide þ 5 wt% carbon nanofiber sample (PI Hybrid 6). At 316  C, there are only three samples better than the control: polyimide þ alumina (PI Hybrid 1), polyimide þ alumina þ exfoliated graphite (PI Hybrid 4), polyimide þ alumina þ carbon nanotubes (PI Hybrid 5). The other samples now show higher amounts of mass loss compared to the control which suggests they may have poor thermal oxidative stability (TOS). The system that stands out as having the least amount of mass loss is the polyimide þ alumina sample, which suggests it may be a good candidate for scale-up studies. 3.2. Material characterization e particle dispersion in pellet Solid pellet samples described in Table 1 were examined by Scanning Electron Microscopy (SEM) to determine how the particles were dispersed in the polyimide. The SEM results showed that

Fig. 2. SEM of pellets of MVK-10 þ 40 wt% Alumina þ 10 wt% Fumed Silica þ 5 wt% Carbon Nanofiber (left), MVK-10 þ 50 wt% Alumina þ 5 wt% Exfoliated Graphite (Right).

A.B. Morgan, S. Putthanarat / Polymer Degradation and Stability 96 (2011) 23e32

27

Fig. 3. SEM of pellets of MVK-10 þ 50 wt % Alumina þ 0.5 wt% Carbon Nanotubes (left), MVK-10 þ 5 wt% Carbon Nanofibers (Right).

the alumina was well dispersed in the polymer matrix in all systems, and the other components (fumed silica, carbon nanofibers, carbon nanotubes) appeared to be well dispersed also (Figs. 1e3). The only nanoparticle that was shown to still be in large agglomerates was the exfoliated graphite flake (Fig. 2, right). It is notable that the alumina de-aggregates so well into finely dispersed 0.5 mm sized particles throughout all of the samples. As expected, the significant difference in size between carbon nanofibers and carbon nanotubes was noted in these systems. The nanofibers can be clearly seen at low and high magnification in the SEM (Fig. 2 left, Fig. 3 right) but the nanotubes on the other hand can just barely be seen at high magnification by SEM (Fig. 3 left). Clearly the nanotubes are much shorter in length and smaller in diameter than the carbon nanofibers. X-ray CT uses a series of 2D radiographic images of a specimen to generate a 3D tomographic image. By scrolling through the individual xy, yz, and xz cross sections in the x, y, and z directions of the sample, the density at all locations within the specimen could be observed. In general, areas of low X-ray attenuation (low density) appear darker than areas of high X-ray attenuation (high density). Fig. 4 shows slices of pellet samples that were isothermally aged at 260  C for 4 h. Contrast between inorganic fillers and matrix is visible in the formulations that have one or two types of nanoparticle (Fig. 4b and c). The inorganic particles appear as light regions on a grey background as they attenuate the X-rays more than the surrounding matrix. For formulations containing three

types of particles, their morphologies seem to be co-continuous (Fig. 4d). Similar observations were found for pellet samples that were isothermally aged at 316  C. There is no evidence of thermal degradation, such as cracking, in these pellets. It should be noted that pellet thickness varied from 1.6 to 2.3 mm. When comparing the results from X-ray CT to the SEM images, clearly one can see that there may be cases where the alumina is not fully dispersed and large agglomerates of alumina may still exist (Fig. 4b for example). Mixtures of multiple nanofillers with alumina seem to break up these agglomerates greatly, suggesting that perhaps during resin þ alumina þ nanofiller mixing that the nanofillers help mix/mill the alumina particles into smaller pieces, but more work would be needed to confirm this. SEM is really just a qualitative snapshot of particle dispersion and not quantitative. Xray CT provides much more useful data and suggests that while most of the alumina is dispersed, the agglomerates of alumina can serve as weak points/nucleation sites for thermal oxidative damage and so more care will need to be taken to ensure all of the agglomerates are fully mixed. 3.3. Polyimide þ protective film/coating panel data The polyimide þ nanoparticle systems described above were then filmed and laid up with other MVK-10 polyimide films and carbon fiber fabric to make 8 ply panels for testing and evaluation. The polyimide þ nanoparticle systems were only put on the tool

Fig. 4. Slices of x-ray images of pellets that were aged at 260  C for 4 h: (a) MVK-10, (b) MVK-10 þ 60 wt% Alumina (left), MVK-10 þ 5 wt% Carbon nanofiber (right), (c) MVK10 þ 50 wt% Alumina þ 10 wt% Fumed Silica (left), MVK-10 þ 50 wt% Alumina þ 5 wt% Exfoliated Graphite (middle), MVK-10 þ 50 wt% Alumina þ 0.5 wt% Carbon nanotubes (right), and (d) MVK-10 þ 40 wt% Alumina þ 10 wt% Fumed Silica þ 5 wt% CNF (left), MVK-10 þ 40 wt% Alumina þ 10 wt% Fumed Silica þ 5 wt% Exfoliated Graphite (right).

28

A.B. Morgan, S. Putthanarat / Polymer Degradation and Stability 96 (2011) 23e32

Fig. 5. Slices of X-ray images of the polyimide þ protective film of different formulations after aging: (a) MVK-10, (b) MVK-10 þ 60 wt% alumina (Note: The alumina particles are seen as light areas), (c) MVK-10 þ 50 wt% Alumina þ 10 wt% Fumed Silica (d) MVK-10 þ 40 wt% Alumina þ 10 wt% Fumed Silica þ 5 wt% Carbon Nanofiber, (e) MVK-10 þ 50 wt% Alumina þ 5 wt% Exfoliated Graphite, (f) MVK-10 þ 40 wt% Alumina þ 10 wt% Fumed Silica þ 5 wt% Exfoliated Graphite, and (g) MVK-10 þ 5 wt% Carbon Nanofiber. Note: The panels shown are panels that were isothermally aged at 316  C for 500 h. Thickness of panels shown is about 2.7e2.9 mm.

side of the composite such that they would co-polymerize with the rest of the composite during composite manufacture and provide protection just to the surface of the polymer. Microscopy analysis and X-ray CT of the panels was again conducted to determine where the inorganic particles went in the system. Microscopy and X-ray CT analysis show similar results. All of the panels with the exception of the panel containing alumina only had very thick top coatings which did not consolidate well into the panel. These layers of resin þ nanofillers with no fiber reinforcement were seen for all the panels except for the MVK-10 þ alumina panel and the MVK-10 control (Fig. 5). These top surface layers with no fiber reinforcement will be likely to crack before oxidation testing and during oxidation testing as well. Those cracks will serve as routes for oxidation damage deeper into the sample [1e10]. X-ray CT results also show that when alumina was used as the only filler, it migrated into the composite panels over half of the thickness as seen in Fig. 5b. When other nanofillers, which greatly increase the viscosity of the resin, were added e there was no migration of the alumina and/or other nanofillers into the panel. The alumina and/or nanofillers were only found in the protective layer. Further analysis in MVK-10 þ alumina only panel using image analysis shows that the alumina concentration was highest at the tool side and significantly decreases in the panel as seen in Fig. 6. So with these observations we need to understand why all of the coating systems with the exception of the MVK-10 þ alumina only system formed resin þ nanofiller-rich top layers. The likely answer as to why the resin þ nanofiller-rich thick films were formed is related to the viscosity increases caused by nanomaterials in a resin matrix. Many nanomaterials cause viscosity increases in polymers during processing as the nanomaterials create network structures or have surface chemistry interactions between nanoparticle and polymer which impede polymer chain mobility and flow [35,37]. The alumina does not appear to induce this problem as during processing of the polyimide resin with alumina the material processed no differently than the base polyimide. Indeed if the alumina causes no viscosity changes during processing, that would explain why the top coating consolidated well into the panel during autoclave processing. Those resin films which had high resin viscosity (such as those with carbon nanofibers or fumed silica) would not consolidate well and instead would form the thick

resin þ nanofiller layers observed. Those thick layers would then be more likely to crack open due to lack of fiber reinforcement and possible coefficient of thermal expansion mismatch between the composite and the top resin þ nanofiller layer. We believe that this is the likely reason for the observed behavior, but more study is needed to confirm this and then try to correct the problem. When studying the traditional measure of thermal oxidative stability (TOS) for polyimide composites, which is aging the specimens in ovens exposed to air for periods of time, we see two types of behavior. At 260  C (500  F), which is the rated use temperature for the MVK-10, we observe that some of the coated systems do slow down mass loss rates over time (Fig. 7, left). However at 316  C (100  F above the use temperature for MVK-10), the mass loss rates are worse for all of the coated samples when compared to the uncoated MVK-10 control sample. The reason for the accelerated mass loss at elevated temperature isn’t fully clear, but we can clearly hypothesize that there are two factors leading to the accelerated mass loss. The first is that the thermal stability of the base polyimide (MVK-10) itself is unchanged because we have not changed any of the polymer bond chemistry through the use of

Fig. 6. Plot shows alumina concentration vs. distance from tool side for polyimide þ protective film of MVK-10 þ 60 wt% Alumina panel. Note that the panel thickness is 2.8 mm.

A.B. Morgan, S. Putthanarat / Polymer Degradation and Stability 96 (2011) 23e32

29

Fig. 7. Weight loss vs. aging time at 260  C (left) and 316  C (right) for polyimide þ protective film panels. F0 ¼ MVK-10 control. F1 ¼ MVK-10 þ 60 wt% Alumina. F2 ¼ MVK10 þ 50 wt% Alumina þ 10 wt% Fumed Silica. F3 ¼ MVK-10 þ 40 wt% Alumina þ 10 wt% Fumed Silica þ 5 wt% Carbon Nanofiber. F4 ¼ MVK-10 þ 50 wt% Alumina þ 5 wt% Exfoliated Graphite F5 ¼ MVK-10 þ 50 wt% Alumina þ 0.5 wt% CNT. F7 ¼ MVK-10 þ 40 wt% Alumina þ 10 wt% Fumed Silica þ 5 wt% Exfoliated Graphite.

protective coatings. Therefore, the inorganic coatings can only slow down heat penetrating into the sample; they cannot prevent it nor can they cause the bonds in the polymers to become more thermally stable. Therefore once the polyimide is taken above its use temperature, it will decompose no matter what. The second reason is that since these systems are high in inorganic content, once the polymer is oxidized away, the inorganic particles create large cracks and conduits into the composite for additional decomposition to occur, which in turn accelerates the rate of mass loss of the decomposing polyimide. More study is needed to verify this hypothesis, but based upon literature data for pure polyimides (especially cracks accelerating decomposition) and our experience with these materials to date, the hypothesis seems quite plausible as an explanation for the higher mass loss at 316  C. 3.4. Flammability data Flammability of the resin þ inorganic filler samples (no carbon fiber fabric) was studied by microcombustion calorimetry (MCC). MCC is an oxygen consumption calorimetry test which measures the heat release (inherent flammability) of a material at a small scale [38,39]. The technique has been found to be very useful in the rapid screening of materials for flammability behavior so that best performers can be selected for scale-up in full-scale fire tests focusing on the selection of new materials for fire safe passenger aircraft interiors [40e46]. In this project, materials with low flammability would suggest that they would be of as heat/fire shields for applications where composite components would be

exposed to high heat fluxes. The reasoning for screening materials with low flammability was that materials with low flammability will have high char yields, and high amounts of carbon char þ inorganic fillers are even more likely to be durable in a high temperature application where direct exposure to high temperature flames or heat sources (radiant or convective) is possible. The data from MCC testing is shown in Table 2 and from the results it is clear that the addition of the alumina particles in combination with carbon nanomaterials and/or fumed silica results in very large reductions in heat release rate. For example, the combination of 40 wt% Alumina with 10 wt% fumed silica and 5 wt% carbon nanofiber results in a 95% reduction in total heat release for this material compared to the base polyimide. This reduction is quite impressive and suggests that composites made with this material will likely perform well in vigorous fire conditions or will not meaningfully contribute to fire spread. Example heat release rate curves are shown in Fig. 8 below. In this figure, heat release rate curves for one sample tested in triplicate are shown, and for the most part, the heat release behavior is very reproducible (Fig. 8, left) but from time to time there are some erratic heat release events which cause the HRR curves to not line up (Fig. 8, right). The likely reason for this is sample heterogeneity at the milligram scale that these samples are tested at in the PCFC. Equally impressive reductions in heat release are noted for combinations of alumina, fumed silica, and graphite, with lesser synergistic reductions in flammability noted for just alumina and fumed silica alone. While detailed mechanistic studies on how these combinations yielded flammability reduction have not been conducted at this time, based

Table 2 Heat release data for polyimide þ inorganic filled systems. Formulation details MVK-10 MVK-10 þ 60 wt% Al2O3 MVK-10 þ 50 wt% Al2O3 þ 10 wt% Fumed silica MVK-10 þ 40 wt% Al2O3 þ 10 wt% Fumed silica þ 5 wt% Carbon nanofiber MVK-10 þ 50 wt% Al2O3 þ 5 wt% Exfoliated graphite MVK-10 þ 40 wt% Al2O3 þ 10 wt% Fumed silica þ 5 wt% Exfoliated graphite MVK-10 þ 50 wt% Al2O3 þ 0.5 wt% Carbon nanotubes MVK-10 þ 5 wt% Carbon nanofiber

MCC data Char yield (wt%)

HRR peak Value (W/g)

Total HR (kJ/g)

% Total HR reduction

45.65 76.6 78.47 87.21

162 68 50 6

13.3 5 4.2 0.6

n/a 62.5 46.2 95.5

75.41 85.53

59 13

4.8 1.5

63.8 88.8

72.11 48.8

78 130

5.9 10.9

55.5 11.8

30

A.B. Morgan, S. Putthanarat / Polymer Degradation and Stability 96 (2011) 23e32

Fig. 8. Heat release rate curves for polyimide control (MVK-10, left) and polyimide nanocomposite (MVK-10 þ 40 wt% Alumina þ 10 wt% Fumed Silica þ 5 wt% Carbon Nanofibers, right).

Table 3 Cone calorimeter data for MVK-10 and MVK-10 þ alumina (top ply only) panels. Sample description

Sample thickness (mm)

Time to ignition (s)

Peak HRR (kW/m2)

Time to peak HRR (s)

Average HRR (kW/m2)

Starting mass (g)

Total mass loss (g)

Weight % lost (%)

Total heat release (MJ/m2)

Total smoke release (m2/m2)

Avg. effective heat of comb. (MJ/kg)

MVK-10þ T650 Carbon fiber Control sample

2.7 2.7 2.7

93 94 89

248 263 267

119 123 116

127 125 128

42.0 42.1 41.8

6.5 6.6 6.5

15.5 15.7 15.5

13.5 13.8 13.9

492 495 503

20.81 20.89 21.22

Average data

2.7

92

259

119

127

42.0

6.5

15.6

13.7

497

20.97

MVK-10þ 60 wt% Al2O3 þ T650 Carbon fiber

2.8 2.8 2.8

94 94 98

263 252 274

125 127 118

139 126 134

44.7 44.7 44.8

6.6 6.7 6.4

14.8 15.0 14.3

14.2 14.3 13.7

509 502 486

21.50 21.40 21.25

Average data

2.8

95

263

123

133

44.7

6.6

14.7

14.1

499

21.38

Fig. 9. Cone calorimeter data e MVK-10 control (left) and MVK-10 þ 60 wt% Alumina top ply only (right).

A.B. Morgan, S. Putthanarat / Polymer Degradation and Stability 96 (2011) 23e32

Thermal Conductivity (W/mK)

1.60

F0

1.40

F1

1.20

F2

1.00

F3

0.80

F4

0.60

F5

0.40

F6

0.20

F7

0.00 100

120

140

160

180

200

220

240

Temperature ( °C) Fig. 10. Thermal conductivity for pellets of polyimide (MVK-10) þ inorganic systems vs. temperature. (F0) ¼ MVK-10 control. (F1) ¼ MVK-10 þ 60 wt% Alumina. (F2) ¼ MVK-10 þ 50 wt% Alumina þ 10 wt% Fumed Silica. (F3) ¼ MVK-10 þ 40 wt% Alumina þ 10 wt% Fumed Silica þ 5 wt% Carbon Nanofiber. (F4) ¼ MVK-10 þ 50 wt% Alumina þ 5 wt% Exfoliated Graphite (F5) ¼ MVK-10 þ 50 wt% Alumina þ 0.5 wt% Carbon Nanotube. (F6) ¼ MVK-10 þ 5 wt% Carbon Nanofiber (F7) ¼ MVK-10 þ 40 wt% Alumina þ 10 wt% Fumed Silica þ 5 wt% Exfoliated Graphite.

upon known polymer nanocomposite flammability theory, we can assume that the combination of the network structure formed by the nanofibers [25e27] plus the inorganic char residue from the alumina and silica which further protects the polymer, [15,17,18,24] are responsible for the flammability reduction observed in this study. It should be noted though that the flammability data collected with the MCC was for solid polymeric materials and so when these formulations are only utilized as a top protective barrier for a fiber reinforced composite as described above, the flammability reduction may not be as large. Indeed that appears to be the case with preliminary cone calorimeter (ASTM E1354, a bench scale flammability test) on composite panels where MVK10 þ 60 wt% Alumina was applied to the top part of the panel. From the data in Table 3 and Fig. 9, the effect of alumina alone as a surface ply/migrated structure (see Discussion above on MVK10 þ alumina) is to delay ignition e otherwise flammability is changed only minimally. The rest of the heat release rate curves are not shown since their effect is the same regardless of the coating used; the HRR curve shape is nearly identical with only slight changes in time to ignition noted. It should be noted that the heat release rate (HRR) curves in Fig. 9 show good reproducibility for the samples tested in triplicate, and this high level of reproducibility was seen in all of the samples. With the large discrepancy in flammability data between solid resin pellets and coatings on composites, some additional analysis and explanation is needed. The solid resin pellets were studied for thermal conductivity (Fig. 10) and it was found that all of the systems where inorganic fillers were added had higher thermal conductivity than the base polymer. Even the addition of carbon nanofiber with no additional inorganic fillers increased the thermal conductivity of the system. Therefore a single thin coating across the composite surface is not a good enough insulator to prevent the underlying material from thermally decomposing and providing flammable gas for ignition. Only if the entire composite is composed of the polyimide þ inorganic filler would one see dramatic reductions in flammability in the cone calorimeter. 4. Conclusions Based upon the results collected to date, the use of high loadings of alumina presents some interesting materials science phenomena and effects. At short periods of time, the use of inorganic particles

31

and carbon nanofillers improves thermal stability of the base polymer, including large reductions in flammability. Over long periods of time though, the high levels of inorganic filler do not seem to delay mass loss of the polyimide, and in some cases may accelerate it. It should be again mentioned though that the base polymer used in these systems is not typically considered to be thermally stable (usable) above the manufacturer recommended use temperature of 260  C (500  F). In light of this information, the higher mass loss rates measured 56  C (100  F) above the use temperature of the MVK-10 polyimide are not that surprising, nor are they a negative result. Instead, they help confirm the importance of resin matrix choice when designing for long-term thermal oxidative stability tests. For fast heating rate thermal damage scenarios (fire damage), this use temperature may not be relevant as the heat release of the base polymer is more important whereas for slow heating rate damage scenarios (thermal soak/long term heat exposure at normal use temperatures), the onset of thermal decomposition is more important and so for that type of testing, one does need to select a base resin with higher “use” temperature. To comment further on the observed mass loss for the base polyimide, the analysis is complicated by the fact that the thermal oxidative stability testing which led us to this conclusion was based upon solid resin pellet data e so the effects we observe may become quite different when the materials are present in a fiberreinforced specimen. The results are complex and still not fully understood at this time, and therefore the outcome of our measurements will require more study before it is definitive and even then, sample geometry will dominate mass loss performance and so the results obtained are only relevant to the type of sample tested. More specifically, mass loss rates from solid resin plaques cannot correlate exactly to mass loss rates for fiber reinforced composites. In regards to fire performance, it was noted that these highly filled systems have higher thermal conductivity than the base polyimide. Therefore the highly filled coatings cannot be used as a thin fire protection barrier on a fiber reinforced composite nor do they slow down mass loss in a coated fiber reinforced composite panel. For the composite to have fire resistance when using the materials shown in this paper, the inorganic filled system would need to be the matrix resin/system for the entire fiber reinforced composite rather than just a top coating composed of MVK-10 þ inorganic þ nanoparticles. Certainly there are thermal protection systems today which can be used to provide fire protection for polyimides, but those systems would be applied after polyimide part manufacture and the strength of our approach outlined in this paper is that the potential fire protection system is the polyimide itself; no additional barriers would be needed if the entire part was made out of the MVK-10 þ inorganic þ nanoparticle system. The alumina filler presents some interesting materials science processing phenomena which requires more study. The migration of the alumina through the carbon fibers in the composite is likely part of the reason why the MVK-10 þ alumina system has a wellconsolidated “coating” on the surface of the polyimide. At the same time, one wonders if thermal protection would be even better if the alumina had stayed on the surface of the part and not migrated into the polyimide. The alumina used in this paper presents a very promising approach to improving the thermal stability of the polyimide, but understanding the how and why of its migration behavior is essential to ensuring control over thermal protection in these composites as well as how to design hybrid materials with even better thermal protection behavior. The multi-component nanocomposite (alumina, fumed silica, carbon nanofiber) was shown to have very low flammability such that it may be useful as is for fire

32

A.B. Morgan, S. Putthanarat / Polymer Degradation and Stability 96 (2011) 23e32

protection, but given the observation that it forms a thick cracked coating on top of the panel, it cannot be used as is. Likely, some additional fiber reinforcement of that structure would be needed to allow it to be used for fire or thermal protection. While there is still more work to be done, it is clear that using inorganic fillers can gain a polyimide some thermal improvements and this approach may be a cost-effective way of improving these materials for higher use temperatures.

Acknowledgements This publication is approved for public release, AFRL release number 88ABW-2010-4325 (August 12, 2010). The authors would like to thank Maverick Corporation for the donation of MVK-10 polyimide to this program. Additional support for the work in this report was provided by the Ohio Department of Development Research Commercialization Program “Protective Integrated Coatings for Extreme Environments” ODOD TECH 09-007. The authors also wish to thank Dr. Weidong Liu for the SEM images. Direct funding for the work in this report was provided by Air Force Research Laboratory “Hybrid Demonstration Materials Development” task on Task Order #4 of Contract FA8650-05-D-5052. This paper is authorized for release. Finally the authors wish to thank Dr. Vernon Bechel and Mrs. Marilyn Unroe for their technical advice on the program.

References [1] Schoeppner GA, Tandon GP, Pochiraju KV. Predicting thermo-oxidative degradation and performance of high temperature polymer matrix composites. In: Kwon Y, Allen D, Talreja R, editors. Multiscale modeling and simulation of composite materials and structures. Springer Verlag, ISBN 978-0-38736318-9; 2008. p. 359e462. [2] Pochiraju KV, Tandon GP. Modeling thermo-oxidation layer growth in high temperature resins. J Eng Mater Technol 2006;128(1):107e16 [special issue on time-dependent behaviors of polymer matrix composites and polymers]. [3] Tandon GP, Pochiraju KV, Schoeppner GA. Modeling of oxidative development in PMR-15 resin. Polym Degrad Stab 2006;91(8):1861e9. [4] Schoeppner GA, Tandon GP, Ripberger ER. Anisotropic oxidation and weight loss in PMR-15 composites. Compos Part A: Appl Sci Manuf 2007;38:890e904. [5] Tandon GP, Pochiraju KV, Schoeppner GA. Thermo-oxidative behavior of high temperature PMR-15 resin and composites. Mater Sci Eng e A 2008;498:150e61. [6] Putthanarat S, Tandon GP, Schoeppner GA. Influence of aging temperature, time, and environment on thermo-oxidative behavior of PMR-15: Nanomechanical characterization. J Mater Sci 2008;43:6714e23. [7] Pochiraju KV, Tandon GP, Schoeppner GA. Evolution of stress and deformations in high-temperature polymer matrix composites during thermo-oxidative aging. Mech Time-Dependent Mater 2008;12:45e68. [8] Tandon GP, Ragland WR, Schoeppner GA. Using optical microscopy to monitor anisotropic oxidation growth in high-temperature polymer matrix composites. J Compos Mater 2009;43:583e603. [9] Pochiraju KV, Tandon GP. Interaction of oxidation and damage in high temperature polymeric matrix composites. Compos Part A; 2009. doi:10. 1016/j.compositesa.2009.03.010. [10] Lu YC, Tandon GP, Schoeppner GA. Elastic and viscoelastic characterization of thermally-oxidized polymer resin using nanoindentation. Mech TimeDependent Mater; 2009. doi:10.1007/s11043-009-9088-7. [11] Morgan AB, Tour JM. Synthesis and testing of nonhalogenated alkyne-containing flame retarding polymer additives. Macromolecules 1998;31:2857e65. [12] Lefebvre J, Mamleev V, Le Bras M, Bourbigot S. Kinetic analysis of pyrolysis of cross-linked polymers. Polym Degrad Stab 2005;88:85e91. [13] Farkas E, Meszena ZG, Toldy A. Modelling of transport processes in a developing char. Polym Degrad Stab 2008;93:1205e13. [14] Eguchi K, Zank GA. Silicon oxycarbide glasses derived from polymer precursors. J Sol-Gel Sci Technol 1998;13:945e9. [15] Kashiwagi T, Gilman JW, Butler KM, Harris RH, shields JR, Asano A. Flame retardant mechanism of silica gel/silica. Fire Mater 2000;24:277e89. [16] Hermansson A, Hjertberg T, Sultan B-A. Linking the flame-retardant mechanisms of an ethylene-acrylate copolymer, chalk and silicone elastomer system with its intumescent behavior. Fire Mater 2005;29:407e23. [17] Marosi G, Marton A, Anna P, Bertalan G, Marosfoi B, Szep A. Ceramic precursor in flame retardant systems. Polym Degrad Stab 2002;77:259e65.

[18] Kashiwagi T, Gilman JW. Silicon based flame retardants. In: Grand AF, editor. Flame retardancy of polymeric materials. NY, NY: Wilkie CA Marcel Dekker, Inc; 2000. p. 353e89. [19] Gilman JW, Ritchie SJ, Kashiwagi T, Lomakin SM. Fire-retardant additives for polymeric materials e I. char formation from silica gel-potassium carbonate. Fire Mater 1997;21:23e32. [20] Bourbigot S, Turf T, Bellayer S, Duquesne S. Polyhedral oligomeric silsesquioxane as flame retardant for thermoplastic polyurethane. Polym Degrad Stab 2009;94:1230e7. [21] Morgan AB. Flame retarded polymer layered silicate nanocomposites: a review of commercial and open literature systems. Polym Adv Technol 2006;17:206e17. [22] Gilman JW, Jackson CL, Morgan AB, Harris R, Manias E, Giannelis EP, et al. Flammability properties of polymer-layered silicate nanocomposites. polypropylene and polystyrene nanocomposites. Chem Mater 2000;12:1866e73. [23] Morgan AB, Harris RH, Kashiwagi T, Chyall LJ, Gilman JW. Flammability of polystyrene layered silicate (clay) nanocomposites: carbonaceous char formation. Fire Mater 2002;26:247e53. [24] Gilman JW, Harris RH, Shields JR, Kashiwagi T, Morgan AB. A study of the flammability reduction mechanism of polystyrene-layered silicate nanocomposite: layered silicate reinforced carbonaceous char. Polym Adv Technol 2006;17:263e71. [25] Cipiriano BH, Kashiwagi T, Raghavan SR, Yang Y, Grulke EA, Yamamoto K, et al. Effects of aspect ratio of MWNT on the flammability properties of polymer nanocomposites. Polymer 2007;48:6086e96. [26] Kashiwagi T, Du F, Winey K, Groth KM, Shields JR, Bellayer SP, et al. Flammability properties of polymer nanocomposites with single-walled carbon nanotubes: effects of nanotube dispersion and concentration. Polymer 2005;46:471e81. [27] Kashiwagi T, Du F, Douglas JF, Winey KI, Harris RH, Shields JR. Nanoparticle networks reduce the flammability of polymer nanocomposites. Nat Mater 2005;4:928e33. [28] Moniruzzaman M, Winey KI. Polymer nanocomposites containing carbon nanotubes. Macromolecules 2006;39:5194e205. [29] Haggenmueller R, Guthy C, Lukes JR, Fischer JE, Winey KI. Single wall carbon nanotube/polyethylene nanocomposites: thermal and electrical conductivity. Macromolecules 2007;40:2417e21. [30] Mu M, Walker AM, Torkelson JM, Winey KI. Cellular structures of carbon nanotubes in a polymer matrix improve properties relative to composites with dispersed nanotubes. Polymer 2008;49:1332e7. [31] Bredeau S, Peeterbroeck S, Bonduel D, Alexandre M, Dubois P. From carbon nanotube coatings to high-performance polymer nanocomposites. Polym Int 2008;57:547e53. [32] Bauhofer W, Kovacs JZ. A review and analysis of electrical percolation in carbon nanotube polymer composites". Compos Sci Technol 2009;69: 1486e98. [33] Kashiwagi T, Shields JR, Harris RH, Davis RD. Flame-retardant mechanism of silica: effects of resin molecular weight. J Appl Polym Sci 2003;87:1541e53. [34] Kashiwagi T, Morgan AB, Antonucci JM, VanLandingham MR, Harris RH, Awad WH, et al. Properties of a silica-poly (methyl methacrylate) nanocomposite. J Appl Polym Sci 2003;89:2072e8. [35] Beigbender A, Linares M, Devalckenaere M, Degee P, Claes M, Deljonne D, et al. CH-p interactions as the driving force for silicone-based nanocomposites with exceptional properties. Adv Mater 2008;20:1003e7. [36] Isayev AI, Kumar R, Lewis TM. Ultrasound assisted twin screw extrusion of polymer-nanocomposites containing carbon nanotubes. Polymer 2009;50:250e60. [37] Koerner H, Misra D, Tan A, Drummy L, Mirau P, Vaia RA. Montmorillonitethermoset nanocomposites via cryo-compounding. Polymer 2006;47:3435. [38] Lyon RE, Walters RN. Pyrolysis combustion flow calorimetry. J Anal Appl Pyrolysis 2004;71:27e46. [39] Walters RN, Lyon RE. Molar group contributions to polymer flammability. J Appl Polym Sci 2002;87:548e63. [40] Hergenrother PM, Thompson CM, Smith JG, Connell JW, Hinkley JA, Lyon RE, et al. Flame retardant aircraft epoxy resins containing phosphorus. Polymer 2005;46:5012e24. [41] Jurs JL, Tour JM. Novel flame retardant polyarylethers: synthesis and testing. Polymer 2003;44:3709e14. [42] Zhang H, Farris RJ, Westmoreland PR. Low flammability and thermal decomposition behavior of poly(3,3’-dihydroxybiphenylisophthalamide) and its derivatives. Polymer 2003;36:3944e54. [43] Schartel B, Pawlowski KH, Lyon RE. Pyrolysis combustion flow calorimeter: a tool to assess flame retarded PC/ABS materials? Thermochim Acta 2007;462: 1e14. [44] Morgan AB, Galaska M. Microcombustion calorimetry as a tool for screening flame retardancy in epoxy. Polym Adv Technol 2008;19:530e46. [45] Cogen JM, Lin TS, Lyon RE. Correlations between pyrolysis combustion flow calorimetry and conventional flammability tests with halogen-free flame retardant polyolefin compounds. Fire Mater 2009;33:33e50. [46] Lyon RE, Takemori MT, Safronava N, Stoliarov SI, Walters RN. A molecular basis for polymer flammability. Polymer 2009;50:2608e17.