The sandwich composites in this study were designed as Figure 1, which were produced from the carbon fibre/epoxy resin prepreg containing 49 wf% resin with an areal density of 345 g/m² (AVIC Composite, China). The Nomex honeycomb of an areal density of 240/m² was selected for core material (NHC-1, AVIC Composite, China). The manufacturing process was vacuum bagging at curing temperature. The skin with three layers prepreg was placed on the top and bottom of honeycomb, then covered by vacuum bag with vacuum at 1 bar and placed into the oven for 2 hours curing at 120˚C.

As can be seen from Table 1, there are five sample groups in this study. For flame retardant sandwich composites C-5 and C-10, the expandable graphite particles (180 μm mesh size and expansion ratio of 100:1) were sprayed on the face of skin and adhered by resin of prepreg before the vacuum bagging process, then vacuumed with 1 bar as well, and moved into the oven for curing at the same temperature. Additionally, the F-5 and F-10 composites where the expandable graphite was filled in the honeycomb with even distribution.

Vertical Bunsen testing was conducted to investigate the flame resistance of sandwich composites. Testing experiments were set up according to Federal Aviation Regulation (FAR) 25.853 standard of 60 s Vertical Bunsen burner testing specifications. The flame of methane gas was adjusted to the reach height of 38 mm flame above the top level of the burner barrel, which was 19 m below the edge of the specimen, and the flame was positioned under the centre of the specimen. There were three specimens for each group with a dimension of 305 mm (Length) × 75 mm (width) × ⁓7.8 mm (thickness) individually hanged upright with the aid of a retort stand and clamp, and tested accordingly using burner (TESTech Instrument Technologies Co., Ltd, China). The burn length, drip flame time, flame time were studied and recorded.

Due to the sandwich structure and different materials on skin and core, the burning length of skin and core was both investigated, as such the cross-section of the sandwich composite was cut for measurement of the burn length of the sandwich structure.

The cone calorimeter tests were conducted on a cone calorimeter (Fire Testing Technology Ltd., UK) with sample dimension of 100 mm ×100 mm × nominal thickness and subjected to one-sided radiant heating according to the ISO5660-1 at the heat flow of 35 kW/m². The heat exposed surfaces of the test specimens were placed 25 mm away from the cone heater. Cone calorimeter parameters including the time to ignition (TTI), time to flame out (TTFO), peak heat release rate (PHRR), total heat release (THR), total smoke release (TSR) and maximum average heat rate emission (MAHRE) were measured.

The temperature on the fire retardant sandwich surface and the honeycomb were measured to estimate the thermal insulation performance. Two K-type thermocouples was attached on the surface of the sandwich specimen (T1) exposed to the flame and the opposite side surface (T3), respectively, and one was inserted into the honeycomb (T2) which had been marked in Figure 1. The surface of the sandwich was exposed to alcohol burner with a flame temperature of 500˚C for 1000 s, the time to thermal equilibrium temperature for each side and honeycomb was recorded.

To analyse the mechanical properties behaviour of the sandwich composite, the specimens were subjected to flexural test by three-point bending test through a 50KN UTM (MTS E42) following the ASTM C393. The specimens were prepared with 200 mm length, 75 mm width, and a span of 150 mm with the rate of loading was 2 mm/min.

The micromorphology images of the char residues after cone calorimeter test were observed on a scanning electron microscope (SEM, Zeiss®, Germany) at a 10 KV voltage after sputtered with 5 nm gold using vacuum film deposition system (EM SCD500, Leica®, Germany).

Statistical analysis was performed with GraphPad Prism (Version 7.00, GraphPad Software, San Diego, CA, USA) with unpaired t-tests. Confidence level was 95%.

Vertical Bunsen Burner Test is a traditional method to investigate the flammability of materials. The results in Table 2 showed that the flammability of sandwich was reduced significantly with the incorporation of expandable graphite (EG) on the surface or into the honeycomb of the sandwich. The Flame Time of the sandwich was observed to decrease from 54 ± 5 s to 45 ± 2 s with the addition of EG into honeycomb from 0 wt% to 10 wt%. According to Figure 2 and Table 2, a decrease of Burn Length was also observed for sandwich composites with EG filling, and Burn Length of honeycomb was significantly lower than Skin for F-5 and F-10 sandwich composite respectively. This was attributed to the EG in the honeycomb rather than on the surface, the Skin-laminate with rich resin was ignited and burning with the fire whilst the EG was able to expand and form the worm-like char filling the gap of honeycomb, becoming the thermal barrier to prevent flame spreading along the Core.

With the same mass fraction of EG coated on the face sheet of sandwich composites, C-5 and C-10 presented lower Flame Time and Burn Length, when compared with EG filled sandwich (F-5 and F-10). The Flame Time was decreased to 30 ± 2 s and 6 ± 2 s for 5 wt% and 10 wt%, respectively. The first

reason could be the EG on the face sheet of sandwich (Skin-Laminate) exposed to fire was fully expanded immediately, the formed thermal barrier on the composite surface could retard the flame spreading on the composite surface [19] .

The effectiveness of fire-proofing approach filling the Expandable Graphite (EG) into the Honeycomb of sandwich or coat on the face sheet of the sandwich composite was evaluated. The HRR-time curve for sandwich composites were shown in Figure 3, and the detail data were recorded in Table 3. It was suggested that the fire reaction behaviour of the sandwich was significantly improved via filling the EG into a honeycomb of composite or coating EG on the face sheet.

Considering the first 100 s of thermal exposure, the reaction behaviour of composite with EG filled (F-5 and F-10) were similar to that of Control. The HRR of Control, F-5 and F-10 composites increased rapidly following self-sustained ignition, reaching the peak value of 322 ± 12, 293 ± 11, 260 ± 9 KW/m²,

respectively. After 100 s, another HRR peaks were also observed at 197 s, 290 s and 334 s, with the peak values of 120 ± 6, 78 ± 5 and 61 ± 8 KW/m², respectively. The behaviour of the second HRR peak was due to the burning of the back face sheet of sandwich composite. With the filling of EG from 0 wt% to 10 wt%, more char formed in the honeycomb to be the thermal barrier retard the fire and heat burning through the honeycomb and transfer into the back face sheet, as such 50% decrease of the second peak of HRR could be observed by 10 wt% EG filled (F-10) composite.

On the other hand, the HRR value of C-5 and C-10 composites with EG coated was found to increase slowly following ignition which was out in 100 s, resulting in lower PHRR value when compared to other composite groups (Control, F-5 and F-10). It was also attributed to the dense char layer formed from the expanded EG (Shown in Figure 4) provided a thermal-physical barrier

that prevented heat conduction and oxygen diffusion to the receding pyrolysis zone.

According to the data in Table 3, the Total Heat Release (THR) value for sandwich composite was reduced from 27 ± 5 to 17 ± 2 MJ/m² with 10 wt% EG filled, whilst that reduced to 7 ± 1 MJ/m² with 10 wt% EG coated. In addition, a reduction of Maximum Average Heat Rate Emission profiles (MAHRE) with an addition of EG was observed for sandwich composite. MAHRE value of F-Composite was reduced from 103 ± 5 KW/m² for Control to 77 ± 6 and 66 ± 7 KW/m² for F-5 and F-10 composite, respectively; while 45 ± 4 and 33 ± 5 KW/m² for C-5 and C-10 was observed.

On the other hand, the Total Smoke Release (TSR) for both fireproof approaches showed interesting behaviours. The TSR value of F-Composite was found to decrease from 620 ± 10 (control) to 354 ± 8 m²/m² (F-10), whereas that of the C-Composite with lower PHRR value showed 1000 ± 12 and 838 ± 8 m²/m² for C-5 and C-10, respectively. This could be attributed to the expanded EG on the face sheet, where the pyrolysis of resin under the fire was retarded with self-extinguishing. Due to the partial combustion without enough heat and oxygen, more unburnt organic particles and gas could release into the air and increase the TSR value.

The flame retardant mechanism of expandable graphite could be confirmed via observation of expandable graphite (EG) before and after Cone Calorimetry test. The micrographs of EG were shown in Figure 5. Normally, the EG exfoliation initialled when heated over 200˚C where the most of intercalating substances degraded and form gases escaped from the interlayer of EG resulting in expanding the graphite with a porous structure. As shown in Figure 5, the virgin EG received from the vendor was in the form of platelet with a diameter range of 210 - 505 μm. After undergoing the thermal conditions caused by Cone Calorimetry Test, the EG expanded along the z-axis of EG and transformed to be vermicular structure, as such the similar diameter range to the virgin EG was observed for expanded graphite. Therefore, it was indicated that the EG filled in the honeycomb or coated on the face sheet could expand to form worm-like porous char and cross combine to be the dense char layer, which can prevent the fire and heat diffusion and reduce the heat release.

In this paper, temperature-time profiles were collected at locations across the thickness of sandwich composites during the thermal insulation test. The temperature-time data at the front surface sheet of sandwich exposed to flame (T1), the honeycomb centre of sandwich (T2) and the back face sheet of the sandwich (T3) for all sandwich composites are shown in Figure 6. The value of T1, T2 and T3 at steady-status, and the time to reach the steady-status were shown in Figure 7.

For T1 temperature of all composite, no significant difference was observed, while the T1 of steady-state (T1_SS) was recorded at ⁓500˚C. However, referred to Figure 7, the C-5 and C-10 composite with EG coating cost ⁓400 s to reach the T1_SS, is much longer than that for Control, F-5 and F-10 composite. It was also attributed to the EG expanded on the surface of the face sheet exposed to fire; as such the heat conduction was retarded by the char layer, resulting in a delay of reaching steady thermal status.

For F-Composites with EG filling, the steady-status temperature of T2 (T2_SS) was observed to decrease from ⁓500˚C (Control) to ⁓406˚C (F-5) and ⁓368˚C (F-10), due to the expanded EG in the honeycomb forming the thermal barrier region in the core of sandwich composite. Moreover, the T3_SS of F-Composite

was shown to decrease from ⁓285˚C (Control) to ⁓270˚C (F-5) and ⁓243˚C (F-10) with the addition of EG. Different from F-Composites, the fire-proof approach of C-Composites was using the EG on the face sheet to form dense char layer exposed to fire. Compared to Control, the T2_SS of C-Composite was decreased from ⁓500˚C to ⁓374˚C (C-5) and ⁓350˚C (C-10), respectively. In addition, the T3_SS of C-Composite was decreased from ⁓285˚C to ⁓239˚C (C-5), and ⁓235˚C (C-10). With the comparison between T1 and T3, the ∆T of all composite could be calculated to indicate the thermal insulation performance of sandwich with EG. According to Figure 6 and Figure 7, with an increase of EG from 0 to 10 wt%, ∆T for F-Composite was increased from ⁓218˚C (Control) to ⁓227˚C (F-5) and ⁓254˚C (F-10), whilst that for C-Composite was increased to ⁓255˚C (C-5) and ⁓269˚C (C-10).

Therefore, the EG filled into the core-honeycomb and coated on the face sheet of the sandwich were available approaches to improve the flame retardancy and thermal insulation of sandwich composite. For the same mass fraction of EG addition, the composite with EG coating presented shorter burning length, lower PHRR value and MAHRE value, higher ∆T value, when compared to composite filled with EG. However, the more smoke released with EG coating should be

considered during the application, so that the coordination with other flame retardant agent for EG coating could be a good approach to improve the fire retardant performance.

The effect of EG on the mechanical composite has been illustrated in literature, whilst the most of them were focused on the laminate composite with the EG mixed in the resin. As such, the effect of EG on the sandwich composite should be analysed and the mechanical performance of novel structure sandwich where the EG was filled in the honeycomb core or coated on the laminate skin could be compared. The Load-displacement curves of the specimens during the three-point bending test was shown in Figure 8. The sandwich structure with EG filled and coated had presented similar failure behaviour when compared to Control. As such, no significant difference was observed amongst the maximum load and the core shear failure of specimens with EG and Control. However, the sandwich with EG filled and coated was easier delaminated and failed with

core-to-face disband in the bending test. According to Figure 9, there is no significant difference figured out from the Core Shear Ultimate Strength ( F s ult ) and Face Stress around the sandwich structure with EG (F-5, F-10, C-5, C-10) and Control. This was confirmed that the sandwich structure was not influenced by the EG filled in the honeycomb or coated on the face sheet.