Microstructure of friction surface developed on carbon fibre reinforced carbon-silicon carbide (C f /C-SiC)

We have used TEM to study the microstructure of friction surface of carbon fibre/carbon-silicon carbide composites brake discs after multi braking stop by using organic pads. A friction surface layer was developed consistently on the top of Si regions of the composites, but inconsistently on that of SiC and C. Inside the layer, amorphous silicon/silicon oxides appeared extensively with various non-metallic and metallic crystallites dispersed inside with sizes ranging from a few nanometers to several microns. A coherent interface between the friction layer and the composite surface was established under the braking conditions, whilst its sustainability varied notably in SiC and C regions. Microcracking near the friction surface appeared in SiC and C f /C regions largely due to the extensive ductile deformation of SiC and weak interfaces between C and C f . Material joining mechanisms were discussed to enlighten the friction transfer layer development on the surface of the composite discs.


Introduction
Carbon fibre reinforced carbon-silicon carbide (C f /C-SiC) composites have been continuously developed as friction materials for brakes of road transport vehicles since their initial commercial success over a decade ago [ 1 ] . However, existing knowledge is often unable to offer enough scientific rationale on the friction performance for a brake that includes a given carbon fibre ceramic as the disc and a chosen friction material as the pads. This situation could hinder the attempts to optimise the potential of ceramic composites as brake discs, as well as to develop suitable pads for such discs. For a while, much of the research attention has been given on the impact of carbon fibre architecture or microstructure of the composites on their mechanical and physical properties. For instance, several publications focused on the impact of carbon fibre architecture and ceramic constituents on the strength, toughness, thermal conductivity and specific heat capacity of the composites.
Whilst these material properties do have certain influence on the friction performance and wear resistance of the composites, the interactions between the composite and the friction materials of pads largely dictate the ultimatum friction performance of the brakes. Up to now, however, only limited study has been available in literature on the friction surface of ceramic matrix composites (CMCs) brake discs, including carbon fibre reinforced ceramic composites [2,3] .
Recently, we presented a fairly systematic observation of the development of friction surface on carbon fibre ceramic composite discs, when a pad made of organic friction material was used to stop the rotating discs [4] . It was noticed that, from very early on, friction surface was subjected to significant development during braking. Whilst the transferred materials, mainly contributed from the pad, deposited on the surface of the discs, their existing format and sustainability demonstrated a strong dependence on the constituents in the composite and surface structure of the discs. It was likewise noticed that fracture occurred extensively on the friction surface, which might have had a significant impact on the deposition of transferred materials, and the premature failure of friction transfer layer, if any. Though friction transfer layer has not generally been emphasized in understanding the friction performance of carbon fibre ceramic discs, some studies did notice the appearance of transfer layers [5] . Fracture on the friction surface, particularly in carbon fibre/carbon (C f /C) regions, had also had the attention of researchers [6] . However, detailed understanding is yet unavailable of the observed deposition of transferred materials.
In traditional brakes including grey cast iron as the discs and organic friction materials as the pads, friction transfer layers are developed on the friction surfaces, and it is these transfer layers that dictate the friction performance of the brakes [7] .
Knowledge on these friction transfer layers has been accumulated for more than three decades already [8] . It is recognised that the stability of the friction layer on the sliding surfaces depends on the cohesive bond strength of the friction layer itself, as well as that between the layer and the surface of disc. It is also generally agreed that stable friction surface is needed for stable friction, low wear rate, noise-and vibration-free performance across a wide temperature regime [7,8] , which directly influence the users' safety, comfort and durability. With the progress of characterisation techniques such as focused ion beam microscopy (FIB) and transmission electron microscopy (TEM), the microstructure of the transfer layers, and therefore the cohesive bond inside the friction film, have been studied with more details most recently. These studies indicated that the transfer layers were composed of nanocrystalline microstructure developed through strong interactions among mechanical mixing, oxidation and severe deformation of the matters on the friction interface [9,10,11,12] . It was also demonstrated that the existence of friction transfer layer was a consequence of dynamic equilibrium between initiation, growth and degradation of contact patches [11,13] . There is little knowledge on the bonding between the friction transfer layer and the discs and pads, whilst a discernable friction transfer layer and a fragmented layer were recently discerned on the surface of grey cast iron [14] . This situation might be due to a belief that bonding should not be a problem in traditional brakes, whilst it could become problematic for CMC bake discs.
In this study, we will expose the details of the friction surface of the ceramic composites by analysing the cross section with TEM. The results of the examination may imply if a robust and strong friction layer can be developed on the surface of carbon fibre ceramic composites brake discs, which might shed light on the observed difference in friction performance among brakes. The study may also benefit any attempts to re-design CMCs for friction application.

Materials and friction surfaces after testing
The C f /C-SiC used in this study was supplied by Surface Transforms (UK). The composites were manufactured through chemical vapour infiltration (CVI) to generate carbon fibre/carbon (C f /C) preforms first, where almost all spaces inside carbon fibre bundles were fully filled in with pyrolytic carbon, and all bundles were wrapped with a layer of pyrolytic carbon. Such preforms were then infiltrated with silicon melt at a temperature above its melting point in an inert atmosphere, normally argon.
The as-received C f /C-SiC composite blocks were cut into discs in laboratory with a diameter of 50 mm and a thickness of 10 mm. All testing surfaces of the discs were finished by polishing with 3 m diamond paste after multi step grinding by applying a series of diamond grits with a size of 68, 26 to 15 m to remove a thickness of about 300-400 m per step, which should be more than enough to eliminate any mechanical damages left from grinding steps.
The friction material for brake pads was selected from products of Federal-Mogul Friction Products Ltd (UK). This material should belong to the organic type, as so called in industry, and its essential characteristics will be highlighted in section 3.2.
The material was mounted on a steel back plate in a ring shape with an outside diameter of 48 mm, inner diameter 40 mm, and thickness 12 mm. The surfaces of the pads were finished by grinding with a grinding wheel consisting of SiC abrasives with a grit size of 3 mm. Cambridge, UK). In most cases, the operating electron acceleration voltage was set to 5 kV; other voltages could be chosen whenever further information was needed.

Cross section of surface
Cross section TEM samples were made on the friction surface with FIB (FEI, FEI Nava 600 nanolab, Oregon, USA) using the following procedure: (a) First, a 1 m thick platinum coating was deposited on the interested region of the friction surface, so as to prevent Ga + implantation and sputter erosion of the top portion of the surface.
(b) A rectangle trench was dug adjacent to the platinum strip using a 20 nA 30kV Ga + ion beam.
(c) When the trench was large enough, the second trench adjacent to the other side of the platinum strip was dug under the same condition. Before the thickness of the wall reduced to about 2 µm, both sides were milled using a 7 nA 30 kV Ga + beam to clean and smooth the surfaces of the cross section.
(d) The membrane for TEM was then lift out and welded on Cu or Mo grits after cutting off from the bulk using the ion beam.
(e) A final thinning process was carried out using a 1 nA 5 kV Ga + beam at an

Friction performance
The measured coefficient of friction (CoF), averaged from recorded data of four pairs of disc and pad, is available from our previous paper [4] for every single braking applied from the 1 st to 49 th . The highest CoF of 0.33±0.01was achieved at the last braking stop, which was raised from a starting level of 0.27±0.01 at the first one.
Whilst the friction had not yet reached to a sustainable stage of friction under the current test regime, the amount of increment of the measured CoF slowed down significantly after 25 braking stops. We believe that the bedding, a process for building up a stable friction, had made a significant progress after 49 braking stops under the current test conditions.

Overview of pad lining materials
Exemplary microstructure of friction material chosen for brake pad of this study is shown in fig. 1. The following chemical elements were detected using SEM/EDX: Fe, Cu, Zn, O, Si, Mg, Al, S, Sn and Ca. The key constituents in the formulation included metallic materials such as copper and steel, abrasives like alumina and silicon carbide, as well as graphite and a small amount of minerals. All these constituents, including polymeric fibres were bonded together by phenolic resin. Like any other commercial organic pad materials, it is difficult to precisely define the composition as each supplier has its own specific formulation with certain modification.

Carbon fibre ceramic composites discs before and after braking testing
Optical other studies [15] . It is believed that during the melt infiltration, the contact regions between Si melt and pyrolytic carbon could be converted into SiC regions fairly quickly through a carbonization process of Si [15][16] . The degree of carbonisation should be controlled by the kinetics of C diffusion through SiC product and the supply of the two chemicals; it is therefore not unusual to see significant amount of residual Si and pyrolytic C existing in the microstructure of the composites. It was noticed that voids and cracks appeared on the surface, which was likely introduced by the surface finish operation, if not all left by manufacture process. Under high magnification, as shown in fig. 2(b), some small pockets of silicon scattered among SiC crystallites, apart from the large Si "lakes" surrounded by SiC. The contrast in pyrolytic carbon wrapping around the vertical carbon fibres was formed by the polarized lighting condition, indicating the degree of graphilization of the carbon [17] .
After up to 49 braking stops on the dynamometer, transferred materials were mostly deposited in SiC+Si regions, as shown in fig. 3. Detailed observation on the depositing process has been published in an early paper through a series of focused examinations on the friction surface after applying a certain number of braking stops [4] .

Cross section on the friction surface of carbon fibre/carbon regions
Previous SEM/EDX analysis of the friction surface of C f /C regions had demonstrated that most transferred materials existed in a crumbled format inside topographically lower places, whilst some carbon fibre surfaces showed clear abrasive wear tracks only [4] . It was also noticed that transferred materials could fill in around interfaces  fig. 4 (b). This layer should be the similar one as that observed on the friction surface of carbon fibre reinforced carbon composites (CCC) [ 18 , 19 ] . It was noticed that a small amount of debris occasionally penetrated into the surface of the carbon fibre, but with a depth inside ~200 nm, as shown in fig. 4(b).
On the C f /C interfaces where transferred materials existed like fillers (inset in fig. 4 (a)), the TEM images revealed that these transferred materials did penetrate into a position with a depth of ~2 m, as shown in fig. 4 Underneath the C f /C region, micro-cracks were initiated along the interface between fibre and matrix, whilst no crack was noticed inside either fibre or matrix.

Transfer layer
As shown in fig. 5(a), the transfer layer was developed with a thickness of a few microns, but varied with the contour of the surface of Si substrate to keep the friction surface on the same level. An exemplary microstructure of the transfer layer near the friction surface is presented in fig. 6(a). It was composed of metallic crystallites (called M particles or regions hereinafter), marked as M, and fine particulate zones (called F zones), marked as FN or FA, the types of F zones defined by characteristics existing in these zones.
In the M zones, most comprised iron-based alloys, but some consisted of copper/copper-based alloys; their sizes ranged from sub-micron to a few microns.
EDX detected significant amount of Si in M zones that contained Fe, though the Si/Fe ratios varied from site to site. A copper zone was marked in fig. 5(b) without Si detected, whilst the neighboured iron alloy contained substantial Si (marked as Fe+Si).
Within the debris, dislocations and sub-grain boundaries existed extensively, along with high angle grain boundaries. The sizes of the polygonized zones or refined grains were around a couple of hundreds nanometers.
In the F zones, two types can be distinguished. The first type comprised densely packed nanocrystallites (called FN zone). Two FN zones were marked in fig. 6(a). metallic particles was also noticed. An example is shown in fig. 5(b) where the iron alloy particle was directly bonded with a copper particle.

Cross section on friction surface of silicon carbide regions.
Exemplary cross section microstructure of the friction surface in SiC regions is shown in fig. 8(a). The transfer layer and the SiC were differentiated with EDX information and/or crystal defects in SiC that could be highlighted under chosen diffraction contrast imaging condition. In fig. 8(a), there were three regions embedded with a transfer layer, marked as TL; others were SiC only on the friction surface, marked as SiC. Clearly the transfer layers were patched, deposited in craters on the surface, with a thickness up to ~2 µm. Therefore, it was very likely that the shape and size of the patches of the transferred materials were defined by the craters themself.
More details on the transfer layer in SiC region is presented in fig. 8 fig. 8(a-b). A large crack was seen that went through the region underneath the surface by following a transgranular fracture path, as arrowed in fig. 8(a)).

Discussion
Experimental evidences have demonstrated that when a pad was applied on a rotating C f /C-SiC composite disc, a continuous transfer layer was successfully developed in contacted Si regions, but only a small amount of patches in SiC and C regions. Whilst Si accounted for < 10% of the total surface area for the composite studied here, it can be much higher than the current level for other commercial C f /C-SiC composites. To understand the sustainability of any transfer layers on the surface, it is necessary to appreciate the followings under the current braking context: (a) the cohesion inside the friction transfer layer; (b) the cohesion between transfer layer and surface of brake disc.

Inferred temperature and pressure for cohesion development inside the friction transfer layer
Outcomes from TEM/EDX analysis have demonstrated that the deposited transfer layer includes metallic debris (M) and fine-grained zones (F). The schematic structure of the transfer layer is shown in fig. 9. It is reasonably supported by he TEM observation (e.g. fig 6(a)) that the structure is characterised by an intercalation of the M and F zones. The cohesion of the structure should hence be established by the bonding among these distinctive zones, as well as that inside the zones under certain temperature and pressure conditions during braking, which are generally difficult to know. Here we are trying to infer these conditions from the exposed microstructure features in metallic debris in the transfer layer.
Inside the M zones, Fe and Si (Fe+Si) were widely detected together, but Fe or Cu alone, or Cu+Si was less frequently. These zones are comprised of sub-grains, as shown in figs. 5-6, and we accordingly believe that a dynamic recovery and/or recrystallization must have occurred in the metal debris under certain temperatures and pressures [20] , which should be defined by the braking operation.
As dynamic recovery/recrystallization happens, the mean subgrain size (d) is found to be strongly dependent on the deformation stress (σ) and only weakly on temperature.
It is therefore possible to estimate the applied pressure during braking by applying following relationship [21] : where K is a constant (for metals, K ≈ 10), µ the materials shear modulus, and b the Burgers vector. By using an estimated d of ~ 250 nm from fig. 6(a), the deformation stress could be ~ 770 MPa, given µ and b a value of 77 GPa and 0.248 nm respectively for steel. The estimation implies that the transferred material can subject to a pressure 2-3 orders larger than the normal pressure applied on the brake.
It is recognized that dynamic recrystallisation should start at a temperature > 0.8T m (T m the melting point of a crystal) [21][22] . T m for mild steel is ~1370 C, and the eutectoid temperature for Fe-Si binary system is ~ 1194 C at a Si/Fe ratio of ~0.5 [22] .
We accordingly infer that a minimum temperature of ~ 1000 C might have been achieved on the friction surface during braking.
Note that the above inference can be applied to understand the development of the cohesion inside the transfer layer, whilst the real conditions, particularly the pressure, can hardly be validated by experimentally testing.
In the FN regions, one of the two subtypes of F zones, nanocrystal Fe 3 O 4 and other minerals from the pad are likely developed through a tribochemical process. Under the inferred temperature and exceedingly high pressure during braking, it is not unreasonable to assume that these nano-sized particles can be packed together through a sintering process. In the FA regions, another subtype of F zones, the cohesion is clearly defined by the amorphous Si itself, as shown in fig. 6 to 7. We have yet no evidence to show how these FA regions were developed, but it is well-known that amorphization in Si can be realized even at ambient temperature when extremely high pressure, e.g. this underneath sharp indenter, is applied [23] . It is reasonable to assume the required pressure for the amorphization of Si would decrease as the temperature rises from room temperature. Therefore, the inferred high pressure and temperature inside the transfer layer might have been high enough to have the amorphous Si developed. At the same time, this amorphous Si could provide enough flowing capability to allow those fine crystallites dispersed inside to form the typical FA regions, as observed in fig. 5 to 6.
The cohesion between the zone is likely developed through a welding process under the inferred high temperature and pressure, a mechanism that is widely believed by researchers who studied the structure of friction transfer layers on the surface of cast iron discs or organic pads [14] . We also anticipate that the existence of silicic constituents, such as Fe+Si, Cu+Si, can be beneficial to bond metallic debris as Si can normally reduce the surface tension [ 24 ] . However, more experimental details are needed in order to understand the roles of Si on the development of the transfer layer from this perspective. It is worth noting that the detected Si was likely from the composites, as no silicon, apart from SiC, was found in the pad formulation.

The cohesion between transfer layer and the constituents in the disc.
Two possible joining mechanisms can be borrowed from material joining sector to understand possible bondings between transfer layer and the surface of disc: fusion and brazing joining.
Fusion joining involves melts of materials. On the disc surface, SiC and C have no melting but subliming when temperature is high enough. However, when oxygen is involved, SiC can be oxidized into SiO or SiO 2 ; these oxides may fuse with other oxides or compounds that exist in the transferred materials or oxidising variants of some constituents, such as FeO x , CuO x , Al 2 O 3 , MgO, as long as the temperature is high enough. Fusion between silicon oxides and other oxides has been claimed to generate silicates by previous tribological studies [25] . However, it is noted that melt cannot appear in a ternary system like FeO-Fe 2 O 3 -SiO 2 until the temperature approaches to just under 1300 C, as shown in fig. 10(a). We have no evidence to support that such a level of temperature achieved on the friction surface, nor experimental details to demonstrate such silicates existing on the interface between SiC and the transfer layer.
However, melt could appear under 1100 C for a Cu-O-SiO 2 ternary system, as isotherm cross section phase diagram shows in fig. 10(b) [26] . These facts indicate that if any fusion occurs on SiC surface via its oxidized variant, a temperature of more than 1000 ºC is necessary. Therefore, when a friction material contains both steel and copper, cuprous/cupric melts can be formed at relatively lower temperatures than ferrous ones, if any transfer layer is bonded on the surface of SiC through fusion.
Comparing to SiC, Si on the disc surface has relatively lower eutectic temperatures with metallic constituents in the debris, e.g. 802 ºC for Si-Cu, though it is still high for Si-Fe (1212 ºC) [27] . This implies that Si and Cu could be jointed together through melting fusion at temperatures significantly lower than 1000 ºC. Evidence for such fusion bonding is shown in fig. 5. It should be pointed out that the existence of oxygen in some places, apart from Si and Cu, raises the fusion temperature above the eutectic temperature of Si-Cu, but fusion may still become possible as the temperature during braking could be well above the eutectic temperature unless the oxygen content is too high. Note, we did not see any direct bonding between steel debris and silicon on the interface in this study.
If there is no fusion bonding, ceramic constituents and metals can still be joined together on the basis of adhesion bonding, because a significant amount of metallic materials exist in the transferred materials. In material joining sector, this is also called brazing joining. To predict the adhesion joining, the following relationship is well established between melt and solid surface (here ceramic constituents on the surface of disc) to predict the possibility of adhesion bonding: Here  SL ,  SV ,  LV denote the liquid-vapour, solid-liquid and solid-vapour surface energy respectively, and  the contact angle between liquid and solid, i.e. the wetting angle. When  SL >  SV , the  is greater than 90 o , no spreading occurs, i.e. the liquid drop tends to spheroidize. In contrast, when  SL <  SV , the  is smaller than 90 o , the solid surface will be covered by a thin film of liquid, spreading along the surface of solid.
Studies have shown that for pure copper, the wetting angles are well above 90º on the surface of carbon and silicon carbide at temperature up to 1000-1100 ºC [28] . This fact implies that pure copper exiting in the debris can no longer be bonded with C and SiC on the composites surface, even if a high temperature reaches up to 1000 ºC on the friction surface of the current study. In the joining sector, however, much effort has been drawn to improve the wetting ability of copper by using active elements. Most of these are IVB elements such as Ti, Zr, Hf, and others such as Ni, Be, Cr, V, In and Co (a range of references on development of active filler metal are available from a review paper on advances in brazing of ceramics [29] ). These active elements can reduce the wetting angle to less than 90º. In our study, EDX analysis either in SEM or TEM did not detect any of such active elements. Nevertheless, Cu+Si was detected on the interface over SiC when cross section TEM/EDX was applied on a friction surface after car testing. Literature shows that the wetting angles of Cu+Si are around 30-40 at a temperature >1100 C [24] . Therefore, even without the active elements, the studied transfer layer can be bonded on the surface of SiC, only when the temperature on the friction surface is high enough.
The wetting angle between Cu and Si is significantly smaller than 90º. For instance, it was shown that, at 1100 ºC, the wetting angle can quickly drop to 6º from initial 45º [30] . However, as the aforementioned temperature of fusion between Si and Cu is significant lower than 1100 ºC, we therefore believe fusion bonding likely superseded the adhesion bonding in Si regions.
Based on the above analysis, we could conclude that although the temperature on the friction surface might have reached around 1000 ºC, and a pressure up to a few hundreds MPa, it might not have been high enough to facilitate any possible bondings between the transfer layer and SiC+Si on the surface of the disc, apart from the fusion bonding between Cu and Si which has a significantly lower melting temperature than 1000 ºC. Note, no bonding was seen between Fe and Si in our study, and it should be another evidence to support that the temperature on the friction surface is lower than <1200 ºC at least.

Mechanical damage in the near surface of the composites
Cross section microstructure analysis has indicated that mechanical damage occurred in carbon and SiC, whilst it was rarely seen in silicon regions. In carbon regions, interfaces between carbon fibre and pyrolytic carbon regions are purposely weakened to provide the composite with enough damage tolerance, i.e. crack propagation resistance, through the bifurcation of propagating cracks [31] . The weakness of the interface did provide preferred sites for harder debris to penetrate in, as seen in section 3, and simultaneously for mechanical damage through microcracking. These microcracks could liaison together to generate large cracks on the surface, which leads to a dislodgment of carbon materials from the friction surface.
Detailed analysis of the friction surface has further reinforced the previous conclusion that surface damage did happen inside SiC regions through cracking at a small scale.
This damage was likely due to the interaction of ductile deformation slips under a smearing condition, as shown in fig. 8. Across each domain or grain, slipping was clearly developed, and these slips from difference adjacent domains interacted each other, which may lead to microcracking.

Summary
The cohesive structure was exposed by cross section TEM both inside the friction