Self-healing polymers and composites

Inspired by the unique and efficient wound healing processes in biological systems, several approaches to develop synthetic polymers that can repair themselves with complete, or nearly complete, autonomy have recently been developed. This review aims to survey the rapidly expanding field of self-healing polymers by reviewing the major successful autonomic repairing mechanisms developed over the last decade. Additionally, we discuss several issues related to transferring these self-healing technologies from the laboratory to real applications, such as virgin polymer property changes as a result of the added healing functionality, healing in thin films v. bulk polymers, and healing in the presence of structural reinforcements.


Introduction
The past half century has witnessed a dramatic increase in the use of polymers and polymer matrix composites (PMCs), which are rapidly supplanting their metal, ceramic and wood counterparts. Perhaps one of the best examples of this polymer insurgency is the recent completion of the Boeing 787 airplane, widely publicised as the first major airliner to be constructed mostly of PMCs. 1 However, while the exponential growth of composites in the aerospace market looks promising as a microcosm for the continuing growth of the polymer industry as a whole, problems arise when exchanging different material types that exhibit inherently different failure behaviours. For example, failure of PMCs generally results from delamination, fibre-matrix debonding, fibre fracture and/or microcracking of the brittle polymer matrix. 2,3 This is in contrast to, for example, the metal parts that the PMCs are replacing, which are typically tougher and used without reinforcement. So in order for the potential of polymer composites (such as high strength to weight ratio, corrosion resistance, versatility in manufacturability, part count reduction, good vibration damping, etc.) to be fully realised, methods to prevent or arrest these common polymer failure modes must be developed.
Traditionally, failure in PMCs is approached with the mindset of managing damage and/or designing materials to better withstand thermal and mechanical loads. However, conventional approaches to managing failure in PMCs (such as damage detection, 4 damage prediction, 5,6 protective coatings and manual repair/ replacement 7,8 ) are limited in that they all require some type of manual intervention, which can be costly and time consuming. And while the numerous approaches to improve PMC's tolerance to thermal and mechanical stresses (either by optimising or developing new polymer matrices and structural reinforcements) have indeed proven fruitful, simple property optimisation cannot free any material from its fate of a finite lifetime.
Instead, given that all PMCs will inevitably fail, perhaps we should entertain the idea of shifting the current paradigm away from simply improving materials to be less susceptible to damage. In other words, a stronger material can only delay catastrophic failure and not prevent it, and that failure could potentially be of great consequence. Surely nature would agree with this mentality, as the protective skin and shell materials of many flora and fauna are significantly weaker than the synthetic composite materials used for structural applications. 9 Instead, nature incorporates her structural materials with sufficient mechanical integrity for its intended applications and handles external damage by implementing advanced, autonomic repairing mechanisms to quickly and efficiently clot and heal cell tissue. In this sense, the 'product lifetime' of natural, biological materials often greatly exceeds that of synthetic polymers and composites. So perhaps a shift towards nature's mentality in material design would also be well suited for man made applications.
To this end, scientists and engineers have taken their cue from nature and, starting from the early 1990s, began creating synthetic mimics of living biological systems. These self-healing polymers emulate biological systems to varying extents, with some being completely autonomic, and some requiring external stimuli to undergo the healing event. Here, we review recent work to develop, optimise and evaluate these numerous approaches to impart a self-healing function to polymers and composites. Additionally, we will discuss in detail various issues related to incorporating self-healing materials into real applications, such as virgin property reduction, healing in films/coatings v. bulk composites, and healing in the presence of structural fibre reinforcements.
The inherent multidisciplinary nature of the field of self-healing polymers, along with the diversity of the different self-healing methodologies, has left some ambiguity on how to best categorise self-repairing systems. For example, some have suggested an organisation based on the underlying chemistry involved in the healing process, 10 the degree of biomimicry, 11 the type of material being healed (e.g. thermoplastic polymers, thermoset polymers, composites, metals, etc.), 12,13 the physical phase of the healing additive (solid v. liquid), 14 the external stimuli required to initiate healing event, 15 and the intrinsic or extrinsic nature of the healing. 16 This ambiguity, which is a direct consequence of the multidisciplinary nature of the field, will likely not be resolved here, but given the spirit of autonomic healing, we organise the different self-healing techniques with the mechanism by which the healing occurs. This organisational approach allows the many fundamentally different self-healing techniques to fall into four different categories: (1) healing by crack-filling adhesion; (2) healing by diffusion; (3) healing by bond reformation; and (4) virgin property strengthening in response to stress (Table 1). Healing via crack filling adhesion is the original and probably most well studied approach to self-healing and will be discussed first.

Types of self-healing Crack filling healing
Most of the concerted effort to develop polymers and composites that exhibit autonomic repair has revolved around crack filling mechanisms, in which fluid substances fill damage volumes and heal with various different chemical and/or physical processes. These crack filling substances, which can include liquid monomers, liquid catalysts, thermoplastic polymers, organic or inorganic film formers, or solvent solutions, are often referred to as 'healing agents', and this nomenclature will be used herein. It is important to consider the advantages of crack filling healing, as opposed to the healing mechanisms presented later in this review that require intimate contact (in some cases, contact on the molecular level) of damage surfaces. Crack filling eliminates the need to close cracks, which otherwise may be difficult for stiff structural polymers and composites. Even if an easy route to crack closure was possible in real applications, this could hardly be considered truly autonomic healing since it would require both damage detection and some extent of manual intervention (e.g. applying pressure). And while it probably can be argued that the healing mechanisms requiring contacting damage surfaces are capable of repairing small sized cracks without the need for external closure, it is at least reasonable to presume that crack filling mechanisms are better suited for healing a wider range of damage volumes. Obviously, this is contingent on a sufficient supply of healing agent to entirely fill cracks, which can be accomplished by optimising the amount of healing agent present in the polymers.

Microencapsulated healing agents
One of the first successful techniques to impart a completely autonomic self-healing function to polymers was the incorporation of liquid healing agent filled microcapsules and catalyst (either as a solid particle or a separately encapsulated liquid) into the polymer matrix. In this approach, which is described schematically in Fig. 1, cracks propagate through the polymer matrix, rupturing several of the microcapsules. Capillary action then drives the liquid monomer out of the broken capsules and onto the crack surface where it comes into contact with the catalyst, which is either embedded in the crack plane (solid catalyst) or also released from ruptured microcapsules (liquid catalyst). Upon contact with the catalyst, the healing agent polymerises and adheres the two crack faces together. 17 The key to realising a successful microcapsule based self-healing system lies with carefully choosing a healing agent/catalyst combination with the requisite features to be compatible with the healing mechanism. For example, the healing agent and the catalyst must have a long shelf life and be stable to the composite processing conditions without undergoing decomposition, uncatalysed polymerisation, or leaching out of the microcapsule shell. Furthermore, once the microcapsules rupture, the healing agent must have a sufficiently low viscosity to flow out of the capsules and completely fill the crack volume in a reasonable timeframe, good wetting properties on the crack surface, and minimal loss of the healing agent from the crack plane through, for example, volatilisation or diffusion into the polymer matrix. And finally, the healing agent must have rapid catalyst dissolution (or for liquid catalysts, rapid mixing) and polymerisation kinetics, low shrinkage upon polymerisation, and the resulting polymer should have good mechanical and adhesive properties. Preliminary systems used styrene/polystyrene blends 18 and phenolic based resins 19 in their microcapsules with varying results, but the healing chemistry found to most completely fulfil this daunting set of healing agent requirements is the ring opening metathesis polymerisation (ROMP) [20][21][22] of dicyclopentadiene (DCPD) with the popular ruthenium based olefin metathesis catalyst bis(tricyclohexylphosphine) benzylidine ruthenium dichloride, colloquially referred to as 'Grubbs' catalyst'. [23][24][25][26][27] In this reaction, the highly strained olefin of DCPD coordinates to the ruthenium catalyst, followed by a cycloaddition with the ruthenium-carbene to form a metallocyclobutane intermediate, and finally a cycloreversion to open dicyclopentadiene's strained ring and add an ultimate unit to the growing polymer chain (Fig. 2). In addition to being a cheap, readily available byproduct of the petroleum industry, DCPD is particularly attractive as a healing agent because it contains a second cyclic olefin that can act as a cross-linking site, producing a polymer with good mechanical properties.
White et al. first reported using DCPD/Grubbs' catalyst in a self-healing system, 28 in which DCPD was encapsulated in a poly(urea-formaldehyde) shell and embedded with the catalyst in an epoxy matrix. It was shown that, after failure, the self-healing polymer could recover up to 75% of its virgin fracture toughness. And by optimising various parameters (catalyst and microcapsule size and loading), it was found that up to 90% toughness recovery could be achieved. 29 In addition to healing large cracks, fatigue lifetime of these systems could be improved to over 30 times longer than that of a polymer without a self-healing functionality, and under certain conditions (low applied stress and short rest periods), fatigue crack growth was indefinitely retarded. [30][31][32] Most utilisations of the DCPD/Grubbs' catalyst based healing agent system have been used to heal structurally dissimilar polymer matrices [predominately epoxies, but other polymers have also been healed, such as PMMA bone cement, 33,34 epoxy vinyl esters 35 and poly(styreneb-butadiene-b-styrene) 36 ]. Given that most living organisms ultimately repair their damaged cell tissue with structurally similar (often identical) tissue material, these ROMP based healing systems are not entirely biomimetic. So intuitively, a healing chemistry that produces a polymerised healing agent structurally similar or identical to the polymer matrix it heals would be of some benefit, not only as a means to better mimic nature, but also to increase the interfacial compatibility between the adhesive (healing agent) and substrate (crack surface). Therefore, two different types of matching polymer matrices/healing agents, based on epoxies and poly(dimethylsiloxane) (PDMS) polymers, were developed and are discussed below.
One of the first successful attempts to match a healing agent to its epoxy matrix was done by Rong et al., in which a diglycidyl ether bisphenol-A (DGEBA) based epoxy resin was encapsulated in a urea-formaldehyde microcapsule and, along with a commercially available capsulated imidazole hardener, embedded in an epoxy matrix made from the same DGEBA epoxy resin as in the microcapsules. 37 Healing with this system showed promising results, with just over 100% recovery in fracture toughness. Further improvements were achieved by replacing the capsulated imidazole with an epoxy soluble imidazole, CuBr 2 (2-MeIm) 4 , dissolved into the epoxy matrix during fabrication. 38 Using this soluble hardener, significantly lower loadings of epoxy/ imidazole healing agents were able to yield better toughness recovery after failure (111%) than with the discrete, capsulated imidazole. Extensive work has been conducted to test the versatility of this system towards glass fibre reinforced composites subject to long storage times (up to 18 months) 39 and different types of damage (mode I fracture v. impact). [40][41][42] All conditions showed good healing capabilities, with healing recoveries ranging from 60% (mode I fracture of composites aged for over 2 months) to over 100% (low energy impact damage of composites with minimal post-fabrication aging time). However, one disadvantage to all epoxy based self-healing systems using an imidazole hardener was the need for external heat to polymerise the healing agent, generally requiring a curing cycle after the damage event at temperatures from 120 to 140uC.
In order to develop a more autonomic epoxy based self-healing system (i.e. without the need for external heat), more aggressive hardeners were used. These hardeners included a mercaptan, encapsulated separate from the epoxy and used as a two-capsule system, 43 and the cationic initiator BF 3 .OEt 2 , which was either encapsulated and used as a two-capsule system 44 or dispersed throughout the polymer matrix via diffusion from BF 3 .OEt 2 absorbed short sisal fibres. 45 Both hardener types cured the epoxy healing agent rapidly at ambient or subambient temperatures and resulted in good healing -80% recovery of impact strength and 104?5% recovery of fracture toughness -at low microcapsule loadings (y5 wt-%). Keller et al. investigated healing in a PDMS elastomer with a healing agent that, when polymerised, is identical to the polymer matrix. 46,47 In this system, a two-capsule healing agent was used: the first capsule contained a vinyl functionalised PDMS resin and a platinum catalyst, and in the second, capsule was a liquid initiator containing a hydrosiloxane copolymer diluted with 20% solvent to reduce its viscosity. Both components were encapsulated in urea-formaldehyde shells. Upon rupture of the microcapsules and release of the two shell materials into the damage area, the platinum catalyst adds the Si-H bonds of the hydrosiloxane copolymer across the vinyl groups of the PDMS resin to cure and heal damage. This system was shown to heal tear damage by recovering 70-100% of tear strength and significantly retard fatigue crack growth. Additionally, it was recently proposed, as a modification to the logistics of this healing chemistry, to use a heterogeneous platinum catalyst supported on a glass fibre to initiate the hydrosilylation of the siloxane polymers. 48 While glass fibres would likely not be used in an elastomer matrix, this concept could be applied to a structural composite, allowing both polysiloxane polymers to be incorporated into one microcapsule, thus eliminating any issues related to mixing of two microcapsule components. These fibres have been shown to quickly and efficiently catalyse hydrosilylation, but have not yet been incorporated into a self-healing system.
Other PDMS self-healing polymers using different healing chemistries have been developed as well. For example, a tin catalysed condensation polymerisation of polysiloxane based healing agents was able to heal puncture damage sufficiently enough to withstand 101?3 kPa of pressure without leakage. 49 'Healing' has been defined thus far as the postdamage recovery of structural integrity through some type of polymerisation. However, not all polymers are intended for structural applications. Paints and primers, for example, are used primarily for their ability to coat and protect other materials, such as metals that commonly rust and corrode if exposed to ambient temperature and humidity. Thus, in these applications, 'healing' can also be defined as recovery of corrosion inhibition. 50,51 While corrosion inhibition can indeed be recovered by applying conventional, polymerisation based healing to polymer coatings, many have also investigated the deposition of corrosion and rust inhibiting film formers into areas of damage. Similar to the previously discussed self-healing systems, microcapsules have been shown to successfully deliver these film forming compounds (which have included organic molecules, organic salts 52 and inorganic salts 53 ) to the damaged regions of coatings and subsequently inhibit metal corrosion and rusting. Another intuitive delivery approach was recently developed by Shchukin and coworkers where the film forming compounds were incorporated into thin layers of polyelectrolytes, which were either directly coated on the metal surface or layered on SiO 2 nanoparticles embedded in a coating. [54][55][56] Once the coating in these systems is damaged and the metal surface is revealed, the onset of corrosion mechanisms changes the local pH in the defect region, thereby distorting the polyelectrolyte layers' adhesion and depositing the film former on the metal surface.
Healing agent/catalyst optimisation Much effort has been dedicated to optimising existing microcapsule based self-healing that uses ROMP healing agents. A large amount of this effort has been focused on improving the kinetics of healing, which is a crucial factor when deciding the appropriate applications for self-healing polymers that may be subject to constant or frequent stress. Perhaps the most straightforward approach to this is to improve the polymerisation rate of the healing agent, but a healing agent's bulk polymerisation rate does not necessarily correlate to the rate of healing, since the complex healing mechanism is a function of numerous other kinetic parameters (e.g. bulk polymerisation rates, catalyst dissolution kinetics, mobility of the healing agent on the crack surface, and mobility of dissolved catalyst throughout liquid healing agent). For example, DCPD/Grubbs' catalyst achieves a maximum room temperature degree of cure in 7-8 h, 57 but self-healing polymers using DCPD and Grubbs' catalyst as a healing agent system reaches a maximum healing in the range of 12-15 h. 58 Two ROMP based monomers that have received significant attention as more rapid healing agents are ethylidene norbornene (ENB) and the exo-isomer of DCPD. DCPD is most easily obtained as the commercially available endo-isomer, and all ROMP based selfhealing systems discussed thus far have also used this isomer. However, the exo-isomer, which can be prepared from endo-DCPD in a two-step isomerisation process, 59 is known to undergo ROMP nearly 20 times faster than its endo-counterpart. 60 So unsurprisingly, when exo-DCPD was incorporated into a self-healing polymer steady-state healing was reached after only about 30 min, nearly 20 times faster than the time required to fully heal with the endo-isomer. 61 However, increasing the bulk polymerisation rate using exo-DCPD as a healing agent came at the expense of significantly decreasing the quality of healing. This was found to be a result of a mismatch between the bulk polymerisation and catalyst dissolution kinetics, causing polymer to be formed only intermittently on the self-healing crack surface, localised around the catalyst particles. An analogous effect was observed with epoxy based healing agents, in which rapidly curing epoxy/hardener combinations also formed non-continuous films of polymer in the crack volume, resulting from a mismatch of polymerisation and epoxy/hardener mixing kinetics, 43,44 if the epoxy and hardener filled microcapsules were poorly distributed throughout the polymer matrix. 62 Some approaches to reduce the severity of these kinetic mismatches, and consequently increase both the quality and speed of healing, will be discussed later in this section.
ENB, a monomer also active towards the ROMP chemistry, is particularly attractive as a healing agent because it is a cheap, commercially available chemical with an extremely rapid bulk polymerisation rate; the time to reach maximum room temperature cure for ENB is nearly an order of magnitude faster than endo-DCPD with an order of magnitude lower loading of catalyst. [63][64][65] Liu et al. first investigated the feasibility of ENB and ENB/DCPD blends as healing agents using a quick and convenient rheokinetic technique developed to simulate a self-healing polymer. 66,67 In this technique, a rheometer's bottom plate was coated with a layer of epoxy polymer containing embedded Grubbs' catalyst (which was polished to reveal the catalyst on the epoxy surface), and the healing agent was injected between the top and modified bottom parallel plates (Fig. 3). This approach, with dissolution and diffusion of the catalyst in the healing agent occurring concurrently with polymerisation, encompasses all of the different kinetic phenomena occurring during the self-healing mechanism. ENB was shown to reach maximum cure in this simulative selfhealing environment in y25 min, and blends of ENB and endo-DCPD were able to effectively merge the rapid healing kinetics of ENB with the good mechanical properties of polyDCPD.
To complement these kinetic optimisations of ROMP based self-healing polymers, a concerted effort is currently under way in our labs to improve the mechanical properties of polymerised healing agents. Sheng et al. developed several custom, ROMP active cross-linking agents that, when blended with other selfhealing ROMP monomers (endo-DCPD, exo-DCPD or ENB), significantly increases storage modulus (12% increase at 20?3 wt-% cross-linker loading), glass transition temperature (y30uC increase at 16?7 wt-% crosslinker loading) and cross-link density (62% decrease in molecular weight between cross-linking sites at 19?6% cross-linker loading) of the resulting polymers. 68,69 Also, Jeong and Kessler have reported one of the first approaches towards improving mechanical properties of healing agents using nanofillers. 70 In this work, multiwalled carbon nanotubes (MWNT) were chemically modified with a ROMP active norbornenyl group, and addition of these functionalised MWNT to DCPD showed massive improvements in tensile toughness (.900%), relative to neat polyDCPD, at nanotube loadings (0?4 wt-%) low enough to not significantly increase the viscosity of the liquid healing agent. More specifically, the strain to failure in these systems increased from 5?75% for neat polyDCPD to 51?8% for the composite, resulting in a corresponding toughness increase of 2?44-25?0 MPa. The functionalised MWNTs also showed excellent long term dispersion in liquid DCPD and ENB, 71 and moderately improved modulus, glass transition temperature and cross-link density of both polymerised healing agents. One downside to microcapsule based self-healing is the non-negligible detrimental effects the embedded microcapsules and catalyst impart to the virgin mechanical properties of the composite. Details related to this virgin property reduction will be discussed later, but in order to partially offset this effect, it is of interest to optimise self-healing systems to require lower loadings of healing additives to achieve high degrees of healing. Rule et al. showed that self-healing polymers with smaller crack volumes requires lower amounts of microcapsules with smaller capsule diameters to achieve full healing, 72 so a method to reduce the size of cracks could presumably decrease the amount of necessary healing components. One technique that could fulfil this requirement is the embedment of shape memory alloys or polymers, which has been shown in the past to adequately facilitate crack closure. [73][74][75][76][77][78] Kirkby et al. recently incorporated shape memory alloy wires into self-healing polymers, which showed that in situ crack closure, resulting from activation of the shape memory wires, did indeed reduce the necessary loadings of healing components to achieve maximum healing efficiencies, and the heat generated from the wires also assisted in curing the healing agent faster and to a higher degree of cure. 79,80 In spite of all the great improvements to self-healing polymers described above, perhaps the largest drawback of using ROMP based self-healing is economics, the Grubbs' catalyst uses ruthenium -a precious metal that will likely never be inexpensive. One way this problem can be addressed is using the monomers described above with faster healing kinetics (such as ENB and exo-DCPD), which require lower loadings of catalyst to achieve high degrees of cure in a reasonable time period. Additionally, increasing the rate of catalyst dissolution in healing agent, either by treating the catalyst particles to have a larger surface area 81 or selecting catalysts and healing agents with inherently matching chemical compatibilities, 82,83 is known to reduce the amount of catalyst required in a self-healing polymer. Also, Rule et al. showed that by encasing Grubbs' catalyst in a protective wax shell, a 10-fold decrease in catalyst loading can achieve similar healing to an unprotected catalyst. 84 This resulted from both the especially small catalyst particle size in the wax increasing the surface area of catalyst available for the healing agent to dissolve and the wax casing protecting the catalyst from decomposition at the hands of incompatible polymer matrix resins. The development of this wax encasing technique also allowed other ROMP catalysts, with significantly lower air stability and lower functional group tolerance (but also lower price) compared to Grubbs' catalyst, to be incorporated into self-healing polymers. 85

Microcapsule optimisation
One of the crucial factors related to engineering selfhealing polymers is the development of adequate microcapsules. The encapsulation technique should, ideally, be simple and user friendly, and the liquid healing agent should be chemically inert to the microencapsulation conditions. The healing agent should not leach out of the capsules over time, nor should any other components destructively permeate into the microcapsule. The capsules should be robust enough to withstand handling and fabrication of self-healing composites, but fragile enough to break and subsequently release core material once the self-healing polymer is fractured. And finally, these capsules should also have a compatible outer shell wall to promote good adhesion to the surrounding polymer matrix.
The encapsulation of DCPD in a poly(urea-formaldehyde) shell was among the first reports of microcapsules for self-healing materials 86 (Fig. 4), and many variations of this seminal fabrication technique, as well as mechanical testing of the resulting microcapsules, have since been reported. 87,88 One significant improvement to the DCPD filled UF capsules was the development of a procedure to produce nano sized microcapsules 89 (Fig. 4), which is important, for example, when fabricating thin self-healing coatings. In other reports, epoxy matrix/microcapsule adhesion was enhanced by grafting either silane coupling agents 90 or epoxy functional groups 91 to the surface of the ureaformaldehyde capsules. One especially interesting variant of these systems was a binary capsule consisting of a large, DCPD filled microcapsule (y140 mm) with smaller capsules (y1?4 mm) containing a second, different liquid at the periphery of the larger capsule 92 (Fig. 4). In this work, the common plasticiser dibutylphthalate was encapsulated in the smaller, peripheral capsules, and preliminary results have shown some potential for encapsulating a liquid initiator in these peripheral capsules, thus potentially eliminating the need for a two-part self-healing system. 93 Additional improvements to encapsulating ROMP based monomers involve the use of stronger shell wall materials. In one example, a technique was developed to encapsulate DCPD in melamine-formaldehyde polymer shells, 94 which was shown to produce capsules with stronger shell materials that rupture at larger deformations than the urea-formaldehyde systems. 95 Recently, Liu et al. developed an approach to encapsulate a blend of ENB and ROMP cross-linkers in a ternary melamine-urea-formaldehyde shell wall 96 (Fig. 4). This fabrication was a very user friendly technique (unlike other microencapsulation procedures, external control of pH was not necessary) that produced a narrow size distribution of capsules without any of the non-capsule polymer debris that plague many of the abovementioned fabrication techniques. These microcapsules also had a very rough outer surface and showed significantly higher thermal stability (300uC) and less core material permeability when compared to their ureaformaldehyde capsule counterparts.
There are numerous reports of microcapsules for selfhealing polymers containing epoxy resins. With some exceptions, most of these reports have encapsulated bisphenol derivative epoxies in urea-formaldehyde shells. [97][98][99][100] One improvement to these systems was the use of a UV curable epoxydiacrylate shell material, which utilises a quicker and more efficient fabrication technique that produces microcapsules with properties comparable to, or better than, the urea-formaldehyde systems (Fig. 5). 101 A moisture sensitive epoxy hardener, BF 3 .OEt 2 , was also encapsulated with this UV curable shell material using a novel technique that avoids the hardener's decomposition, which would otherwise occur very rapidly as most types of encapsulation procedures necessarily use aqueous chemicals. In this technique, carbon dioxide bubbles were encapsulated in the shells, followed by evacuation of the CO 2 from the capsule, and diffusion of the BF 3 .OEt 2 into the empty core (Fig. 5). 102 A plethora of other resins have been encapsulated in different shell wall materials, all of which show promise for self-healing applications. For example, solvent/ epoxy solutions (Fig. 5), 103 solvent/carbon nanotube suspensions 104 and liquid paraffin wax 105 have been encapsulated in urea-formaldehyde shells; styrene 106,107 and polythiol epoxy hardeners (Fig. 5) 108 have been encapsulated in melamine-formaldehyde shells; reactive amines have been encapsulated in polyureas; 109 and diisocyanate resins were encapsulated in polyurethane shells. 110 Other optimisations of self-healing microcapsules include variations in healing agent delivery technique, such as work by Kirk et al. where, instead of microcapsules, epoxy and hardener were absorbed into nanoporous silica, which was then incorporated into a polymer. 111 Also, Pastine et al. recently reported the development of phototriggerable microcapsules, which 'burst' upon exposure to UV radiation. 112 These capsules could possibly be used to maximise the amount of healing agent released into a surface crack (for example, in a self-healing coating or film) upon exposure to artificial or even natural (solar) UV light.

Phase separated healing agents
An alternative to storing healing agent in microcapsules is the incorporation of healing agents directly into the polymer matrix as a phase separated component. One key difference between this approach and the microcapsule systems is that the protection afforded by the capsule shell is lost, therefore necessitating long term inertness between the healing agent and the matrix resin/polymer. But even though this brings about stricter healing agent requirements than that of the microcapsule based healing agents, these phase separated systems offer the potential of reduced fabrication time (i.e. no need for encapsulation procedures) and generally simpler processing. Zako et al.
were the first to demonstrate healing with phase separated healing agents. In their work, solid epoxy prepolymer healing particles (Toa-Gosei AP-700), with a melting point of 383-413 K and a mean diameter of 105 mm, were incorporated into a fibre reinforced polymer matrix. 113,114 After damage to the matrix, these composites were heated above the melting point of the epoxy prepolymer, which then caused the epoxy particles exposed on the fracture surface to melt and fill the damaged region. Upon cooling, the epoxy healing agent cures to heal the polymer, showing nearly full recovery of stiffness and even an increased fatigue lifetime, relative to the virgin material.
Cho et al. demonstrated self-healing of a vinyl ester matrix containing both encapsulated and phase separated healing agents. 115 In this system, a mixture of hydroxyl end functionalised polydimethylsiloxane and polydiethyoxysiloxane healing agents were incorporated into the matrix as the phase separated component, and a solution of an organotin polycondensation catalyst was used in embedded microcapsules. Healing in this system was relatively less efficient (24% recovery of fracture toughness) than others, but the healed regions showed excellent corrosion resistance, chemical stability and passivating ability, which are of paramount importance to the paint and coating industries. 116 Meure et al. introduced a slightly different methodology using phase separated particles of a thermoplastic polymer, polyethylene-co-methacrylic acid (EMAA). 117 After fracture and a subsequent heating cycle, the EMAA polymer infiltrated the damage regions and restored 85% of the virgin polymer's fracture toughness (critical stress intensity factor). One especially interesting aspect of the EMAA healing agent is a built-in pressure based delivery system resulting from bubbles present in the phase separated EMAA particles. These bubbles, likely filled with water resulting from reactions between the EMAA particles and the polymer matrix resin during fabrication, 118 expand during the heating/healing cycle to force the healing agent into cracks (Fig. 6).
Another thermoplastic polymer used as a phase separated healing agent is poly(e-caprolactone) (PCL). 119 PCL forms a miscible blend with uncured epoxy resin, but during the epoxy's curing reaction, the PCL undergoes a reaction induced phase separation to yield a polymer with an ordered 'bricks and mortar' morphology with spheres of epoxy polymer (bricks) interpenetrated with a network of PCL (mortar) (Fig. 7, left). After damage and a short heating cycle, the PCL melt was able to infiltrate damage and either completely recover or surpass some virgin mechanical properties (Fig. 7, right). This system is especially promising for its simple processing, micron scale distribution of phase separated PCL, and, interestingly, many of the material properties of the polymer are dominated by that of the thermoset epoxy 'bricks'.

Healing agents in hollow fibre
One drawback for many of the systems described above is that small, localised volumes of healing agent limits repeated healing of one damage site. Storage vessels capable of supplying volumes of healing agent significant larger than expected crack sizes could potentially address this problem, but increasing the size of the types of storage vessels described above could drastically diminish the virgin properties of the polymer in which they are embedded. To solve these problems, significant effort has been dedicated to incorporating healing agents into hollow fibres, which can be used both as a structural reinforcement and as a large storage vessel. Once failure occurs in composites filled with these resin infused fibres, fibres rupture and healing agent(s) can diffuse into the damaged regions (Fig. 8). These hollow fibres have been filled either through open ends, with either capillary action or vacuum assistance, or through surface pores, which need to be covered after resin infusion ( Fig. 9).
Earlier generations of composites with resin filled fibres were reported by Dry et al. [120][121][122][123] and Motuku et al. 124 to rupture in response to damage and adequately release healing agent. The quality of healing, however, was limited by poor healing agent reactivity and especially large fibre sizes that frequently acted as initiation sites for failure. Modifications to these early fibres have been reported by Dry et al. to adequately heal composites in short time periods, 125-127 but the differences between these more recent reports and the earlier systems are not described in detail. Resin filled fibres with diameters small enough to impart some structural reinforcement to the polymer matrix were used by Bleay et al., 128 but the healing agent chemistry in these systems generally required harsh curing conditions (vacuum and heat).
Bond and co-workers developed a technique to manufacture hollow borosilicate glass fibres with variable degrees of hollowness, internal diameters, external diameters and lengths. 129 After targeting the optimal parameters for self-healing applications (fibre hollowness: 50-55%, external diameter: 60 mm, internal diameter: 40 mm), these fibres were filled with a two-part epoxy resin and, in some cases, a UV fluorescent dye. Early work incorporated these fibres into reinforced polymers as groups of consecutive orthogonal plies, which showed that, after applying load, the UV dye from the fractured fibres can sufficiently fill fracture and delamination areas to aid in damage visualisation. 130,131 Additionally, after a short healing time, epoxy resin/ hardener was able to flow out of the fibres and cure, recovering nearly 75% of the flexural strength lost after damage. This system was further optimised by intermingling the filled hollow fibres at optimal pitch spacings within E-glass or carbon fibre plies, resulting in both negligible virgin property reduction and nearly full recovery of flexural [132][133][134] and compression after impact strength. 135 A number of different types of healing agents have been reported to sufficiently heal when incorporated into hollow fibres. In one approach, Liu et al. reported a coating that can heal its ability to act as a water permeation barrier. 136 In this work, hollow, water degradable poly(lactic acid) (PLA) fibres were filled with a metal oxide precursor healing agent, TiCl 4 , and subsequently incorporated into subsurface polymer layers of multilayer films. Once damage penetrated deep enough in the multilayer film to reach the PLA layers, atmospheric moisture degraded the PLA fibres, releasing TiCl 4 into the damaged region. On contact with ambient humidity, the healing agent oxidises to a TiO 2 film to form an effective water permeation barrier. Another healing agent reported to adequately self-heal a polymer when infused in hollow fibres is DCPD, otherwise well known in the realm of microcapsule based self-healing. 137 A commercially available borosilicate glass tube (external diameter: 125 mm, degree of hollowness: 64%) was filled with either DCPD or a suspension of functionalised MWNT 70 in DCPD by capillary action, sealed at both ends and coated with Grubbs' catalyst at its exterior walls. After damage and a short waiting time, healing with DCPD was able to recover 90% of its virgin tensile strength, with even higher values of strength recovery possible at low loadings of the functionalised MWNT.

Microvascular networks
Infusing larger storage vessels, such as the hollow fibres described above, with healing agent, will likely result in multiple healing events of one damage site, but the total volume of liquid in each fibre is still finite. A logical progression to supplying even larger volumes of healing agent to damage sites is through a series of healing agent filled interconnected channels, which could potentially be linked to an external, refillable liquid pump to deliver a constant supply of healing agent. Healing with connected networks of healing agent is mechanistically similar to that of the hollow fibre approach, and it is perhaps the most biomimetic self-healing system that will be presented in this review since the network of channels is visually and conceptually similar to the vascular systems of many plants and animals (Fig. 10). Appropriately, connected networks of flowing healing agent have been coined 'microvascular' networks.
Connecting channels of storage vessels adds several degrees of complexity not present in most other selfhealing techniques. Bejan and co-workers have addressed some network design parameters by calculating optimal two-and three-dimensional vascular architectures that efficiently deliver healing agent to multiple damage sites with maximum flow properties. [138][139][140][141][142][143] Williams et al. complemented this work with calculations to determine optimal network sizes and architectures to enhance the reliability (e.g. minimising the negative effects of network leakage and blocking) and applicability (e.g. minimum mass penalties, pumping power requirements, etc.) of microvascular self-healing systems. 144,145 However, the complexity of many of these optimal design conditions for microvascular systems would likely require extremely sophisticated fabrication techniques that are currently technologically unavailable or too elaborate to be practical. Therefore, only a few vascular architectures described in these reports have been experimentally demonstrated.
A fabrication technique utilising a robotic direct write assembly to develop a scaffold of fugitive wax was used to construct microvascular networks. The fugitive wax scaffold was infused with an epoxy/hardener resin system, which was subsequently cured, and the wax was removed at elevated temperatures to create a well defined network in the evacuated microchannels. [146][147][148] Initial self-healing work using this fabrication technique contained Grubbs' catalyst embedded in the epoxy matrix and DCPD flowing through the microvascular 9 Hollow fibres with open ends (left), 133 and hollow fibres with open surface pores (right) 136 10 Microvascular based self-healing concept: a a capillary network in the outer skin layer with a cut and b schematic of an epoxy specimen containing a microvascular network, loaded in a four-point bending configuration monitored with an acoustic emission sensor. Reprinted with permission from Ref. 149 network, which showed up to 70% recovery of fracture toughness after damage in a four-point bending protocol. 149,150 Substantial healing was observed for up to seven damage/healing cycles, after which point healing diminished, presumably due to consumption of the embedded Grubbs' catalyst that cannot be replenished as easily as the DCPD flowing through the network channels. This prompted Toohey et al. and Hansen et al. to change the healing chemistry from a system using a solid catalyst to a two-part liquid healing agent -an epoxy and an amine hardener, with each part flowing through segregated vascular networks. 151,152 Over 60% recovery of fracture toughness was achieved for up to 16 intermittent damage/healing cycles and 50-100% recovery of fracture toughness was achieved for 30 damage/healing cycles with adjacent and interpenetrating networks respectively. In another work, microvascular networks were added to composite sandwich structures by embedding PVC tubes horizontally into a polymethacrylimide closed cell foam, drilling vertical holes at appropriate locations into the foam through the tubes to form vascular networks, and sandwiching the resulting samples with an E-glass/ epoxy composite skin. 153,154 After impact damage, vascular networks filled with a premixed epoxy/hardener were able to fully restore the undamaged sandwich structure's flexural load and compression after impact (CAI) strength. When epoxy and hardener were separately added to segregated vascular channels (in which mixing of the two components must occur after they diffuse out of the networks into damage regions), full recovery of virgin properties also occurred, but only when both filled networks were ruptured.

Diffusion
The subsections below discuss self-healing mechanisms based on molecular diffusion of a mobile species to create chemical or physical adhesive linkages. An important distinction can be made between these systems and the crack filling adhesives: essentially, the diffusion based healing mechanisms in the subsections below all require the transport of the mobile species from one damage surface to another, as opposed to the crack filling systems that fill the space between damage surfaces with a healing agent. This distinction brings about several issues not present with the crack filling systems, such as both the necessity for crack closure and application of an external stimulus to drive the movement of the mobile species.

Healing via a thermoset/thermoplastic miscible blend
Hayes and co-workers developed a self-healing approach based on a thermoset/thermoplastic blend. A thermoplastic polymer (polybisphenol-A-co-epichlorohydrin) chosen to form a homogenous blend with the thermoset, both before and after curing, was dissolved into the thermoset epoxy resin, and E-glass/epoxy composites were fabricated with conventional lay-up techniques. 155 After tensile or impact loading, significant decreases in delamination area and increases in extension at failure, fracture toughness and impact strength were observed after a short heat treatment at temperatures ranging from 100 to 140uC, resulting from the thermoplastic polymer melt diffusing through and filling the damage region. The extent of healing varied with different thermoplastic polymer loadings and healing temperatures, with optimal conditions showing y70% recovery of virgin properties and significant, but slightly diminishing, healing with repeated damage/healing cycles. 156

Dangling chain diffusion healing
Another diffusion based healing technique utilises interactions between 'dangling chains' of polymer, branching off of the main polymer backbone, across damage surfaces to heal. After diffusion, these dangling chains can heal a polymer either by physical (mechanical interlocking) or chemical (reactions) interactions. One such polymer was fabricated by eliminating the sol fraction of a weak polyurethane sol-gel just beyond its sol-gel critical point. 157,158 Elimination of the sol near the critical point allowed the polyurethane gel to have a high enough cross-link density to maintain mechanical integrity, but a sufficient number of dangling chains available for healing. After cutting these polymers with a razor blade, the two damage surfaces were brought into contact, which allowed for topological interactions (interdiffusion of dangling chains) between the damage surfaces to effectively heal damage. The extent of healing was found to be strongly dependant on the number and length of the dangling chains, and samples containing optimised chain parameters demonstrated the ability to recover nearly 80% of their virgin tear strength after only 10 min of room temperature healing time.
Given that the gels discussed above have relatively weak mechanical properties (which was actually a necessity to produce the dangling chain molecular mobility required to facilitate room-temperature curing), their application base is limited. Rahmathullah and Palmese investigated the ability of thermoset epoxy/ amine polymers, which are often used for structural applications, to heal via topological diffusion mechanisms. 159 Fractured halves of compact tension test specimens were brought into intimate contact with each other, clamped at either high or low pressures, and healed with a heat treatment of 185uC for 1 h. Elevated healing temperatures were required to bring the epoxy into its rubbery state, which allowed for enhanced molecular mobility of the topological polymer chains. In cases where the polymer was initially fabricated with a stoichiometric ratio of epoxy and amine, significant diffusion across the crack plane occurred to recover up to 50-60% of the maximum load at failure, which was repeatable over numerous damage/healing cycles. In cases where the epoxy/amine ratio was not stoichiometric, over 100% recovery of virgin properties was possible due to unreacted epoxy groups diffusing through the damaged areas and undergoing various epoxy ring opening polymerisation reactions at the crack interface.
Caruso et al. also demonstrated healing of epoxy/ amine matrices with residual reactive groups. 160 Various solvents were selected and encapsulated in urea-formaldehyde microcapsules, which were embedded in the polymer matrix. Upon fracture, the microcapsules would rupture and transport solvent into the damage region, which acted to increase local molecular mobility at the crack surfaces and assist in the diffusion of residual reactive groups. This solvent assisted diffusion allowed the residual epoxy/amine groups to heal below the bulk polymer's glass transition temperature (in the case of this work, at room temperature), resulting in over 80% recovery of fracture toughness. This system was later improved upon by including epoxy resin solutes in the microcapsules, which allowed for full recovery of the virgin material fracture toughness. 161

Viscoelastic healing
Kalista, Jr and co-workers demonstrated a healing technique that takes advantage of the inherent viscoelastic behaviour of polymers. Thin films of commercially available EMAA, either neutral or partially ionised, were subject to high energy impact or sawing damage, after which point the damage was autonomically healed to withstand pressures of up to 3 MPa. 162,163 This healing was attributed to a viscoelastic response of the EMAA, facilitated by a transfer of energy from the damage event to temporarily bring the damaged region of polymer into a melt. More specifically, the transfer of energy during impact or sawing elicited a localised melt of the damage region that allowed the inherently elastic response of EMAA to physically close deformations. Subsequently, the viscous behaviour of the polymer melt allowed for diffusion of polymer chains to seal the damage site. This healing mechanism was shown to be applicable only when the damage event was of high enough energy to bring the local damage region into the melt. Also, healing was only observed over a defined temperature range, 230-60uC, since at lower temperatures the damage event could not transfer enough energy to heat the polymer above its melt temperature, and at higher temperatures, the damage energy was dispersed more easily throughout a larger area of polymer, also resulting in a failure to heat the localised damaged region into a melt.
Studying the detailed mechanism of this viscoelastic healing has proven difficult as the healing response occurs almost instantaneously (attempts to visually monitor the healing were unsuccessful, even with high speed camera analysis capable of 4000 frames/s 162 ). To address this problem, a quasi-static test method was developed to mimic the high impact ballistic damage in which a preheated, disc shaped object was rapidly (10 3 -10 5 mm min 21 ) pulled through the EMAA polymer in a controlled manner. 164 This technique allowed for the contribution of the elastic and viscous components of the healing process to be monitored independently. It was proposed that the ionomeric clusters along the EMAA chain enhances healing, relative to neutral EMAA, by increasing the elastic modulus and therefore aiding in efficient damage closure. The viscous component of healing was improved by heating the localised damage region to higher temperatures and maintaining the heat for longer times. 165

Bond reformation
Polymers with controlled reversible polymerisation processes have shown great potential for self-healing. These unique polymers, sometimes called 'mendomers' or 'dynamers', contain specific bonds that are reversible in response to a generally mild external stimulus (e.g. heat, light, acidic or basic conditions, etc.). This unique characteristic has profound implications for self-healing polymers: a bulk mendomer has the potential to repeatedly heal itself in such a way that the virgin material is fully restored, even at the molecular level. A great number of different mendomers have been reported in the literature, so the entire scope of these reversible polymers is too great to fully address here; interested parties are directed to several recent reviews that give due credit to the breadth of this field. [166][167][168][169][170] Instead, we focus our discussion only on reversible polymers that have been shown to heal bulk damage in solid state polymers.
By employing rational molecular design, the reversible bonds in mendomers can be engineered to behave as 'weak links' that preferentially break in response to stress, preserving the intactness of the irreversible bonds. This is especially important for self-healing applications: if large scale damage can translate to the molecular level exclusively as cleavage of the 'weak links', then molecular structure can be completely restored upon employing the external stimuli to reform these broken 'weak links'. However, it is probably unreasonable to expect macroscopic damage of a bulk polymer to occur exclusively through these 'weak links' (although we admit that, to the best of our knowledge, there have been no literature reports to confirm or refute this claim). For this reason, many of the polymers presented in this subsection have only been shown to heal after a depolymerisation/polymerisation cycle, which allows intact 'weak links' at the damage site to break and subsequently reform with 'weak links' on opposite damage surfaces. Also, the depolymerisation step can provide additional molecular mobility for the 'weak links' to partially fill the crack volume and diffuse into opposite crack surfaces to find bonding partners.
One inherent limitation in these systems is that applying the required external stimulus to initiate the depolymerisation/polymerisation cycle of damaged mendomers requires manual intervention, and the healing is therefore not fully autonomic. However, clever solutions to this problem have recently started appearing in the literature. For example, mendomers requiring thermal stimuli can be heated by passing electrical currents through reinforcing graphite fibres, 171,172 although incorporation of these fibres still cannot be considered truly autonomic healing without an efficient detection system that can locally heat damage regions. Also, mendomers requiring UV irradiation to reverse bonding can potentially receive this external stimulus via sunlight. But there are still significant questions that need to be addressed before the sun can be considered a feasible self-healing UV source, such as how to exclusively expose sunlight to damage regions in order to not constantly depolymerise 'healthy' polymer.

Thermally reversible self-healing
For a large number of polymers that heal via bond reformation, heat is the external stimulus that begins the healing process. The reason for this ubiquity can perhaps be attributed to the fact that a thermally reversible chemistry: the Diels-Alder (DA) reaction (between a diene and a dienophile), is ideal for selfhealing for a number of reasons. First, when a good diene and dienophile pair is chosen (such as furan and maleimide derivatives, respectively), the DA reaction can occur at room temperature without the need for any external reagents or solvents, which is an invaluable quality when preparing bulk polymers. Additionally, the controlled retro-Diels-Alder (rDA) reaction of furan/ maleimide derived polymers occurs in the temperature range of approximately 90-120uC, which is low enough such that thermal degradation will not compete with the cycloreversion, but high enough for some structural applications. And finally, the DA adduct is often believed to act as the 'weak link' during failure, cycloreverting to reform the diene and dienophile preferentially to random bond scission. This implies that, should the 'broken' diene and dienophile parts be brought together again in a solid state DA reaction, damage can be repaired on the molecular level.
Chen et al. developed thermoset polymers based on the DA reaction between the multifuran and multimaleimide monomers shown in Fig. 11, which resulted in a polymer with tensile, compressive and flexural mechanical properties similar to commercially available epoxies and unsaturated polyesters. 173 Original reports demonstrated that this polymer, when fractured in a compact tension protocol, could be heated to above 120uC and recover 57% of its original fracture load upon cooling (Fig. 11). Using differential scanning calorimetry and solid state 13 C NMR spectroscopy healing was observed to predominately result from the solid state DA reaction, and this healing was repeatable for multiple damage events. 174 Later reports showed that full recovery of virgin mechanical properties could be achieved for repeated damage events when fracture surfaces were accurately aligned and clamped with 0?35 MPa of pressure during healing. 175 Healing of polymers with structurally different multifuran and multimaleimide monomers has also been met with similar levels of success. [176][177][178] Murphy et al. have developed a series of self-healing DA based monomers that use cyclopentadiene as both the diene and dienophile. In this novel approach, a macrocyclic derivative of DCPD, which is the DA adduct of two molecules of cyclopentadiene, was synthesised using several short organic linkers in such a way that the rDA reaction of the macrocycle would yield an a,v-bis(cyclopentadiene). 179 Upon cooling to below rDA temperatures, this linear molecule will preferentially undergo intermolecular DA reactions, as opposed intramolecular recyclisation, to form the polymer backbone (Fig. 12). These resins are unique in that the one-component monomer greatly simplifies polymer fabrication, relative to multipart resins that require homogeneous mixing of two or more liquids or solids. Healing efficiencies varied with the different mendomers shown in Fig. 12, but all systems showed full recovery of compression strength over several damage/healing cycles, as high as 60% recovery of fracture strength, and visual disappearance of both fracture and indentation damage after heating at or above 120uC. 171,172 While the above systems show great potential for healing at the molecular level, the economic drawback of scaling up synthesis of these custom made DA monomers would likely outweigh the benefit of healing, at least for large scale, industrial applications. One recently investigated approach to reduce the economic severity of thermally reversible healing is to functionalise well known polymers, made from resins that are already cheap and commercially available in massive quantities, with DA based thermally reversible cross-links. And in addition to healing, when these cross-links are 'broken' at elevated temperatures via the rDA reaction, the resulting functionalised resins and polymers can be recycled using standard thermoplastic processing technology. To this end, several well known polymers (such as polyamides, 180 polyketones, 181 PMMA 182-185 and polyesters 186,187 ) have been functionalised with pendant or telechelic furan groups and cross-linked with small bis-or tris-maleimides. The functionalised resins were either commercially available or obtained in 1-2 step syntheses from commercially available resins. All crosslinked polymers were shown to have thermal and mechanical properties superior to their non-cross-linked polymer counterparts, and after a heat treatment to cleave the cross-linking groups, they could be processed using standard solution or melt processing techniques. Crack healing behaviour was observed in all systems, with nearly full recovery of virgin material properties possible after the rDA/DA heat treatment. 12 Three mendomers (above), which form a self-healing polymer after a heat treatment (below) In addition to the thermoplastic polymers above, epoxy resins have also been prepared with thermally reversible furan/maleimide cross-links (Fig. 13). Since epoxies traditionally behave as thermosets, these polymers are unlike the DA cross-linked thermoplastics above in that they likely will never have the ability to be melt or solution processed, but they can still function to heal damage. In one example of this work, Tian et al. synthesised a furan functionalised diepoxide resin that, when cured with an anhydride hardener and cross-linked with a commercially available bis-maleimide, possessed thermal and mechanical properties similar to or better than traditional epoxy resins. 188 After damage and a 20 min heating cycle at >119uC, a decrease in crack size was qualitatively observed.
Peterson et al. developed a thermally reversible epoxy gel, which was intended for use as a crack filling healing agent (similar to the above microcapsule or phase separated healing agents) that could be added to a traditional, unfunctionalised polymer matrix as a secondary phase. 189,190 The healing agent was synthesised from a furan functionalised epoxy prepolymer in N,N'-dimethylformamide solvent, which was crosslinked with a commercial bis-maleimide to form the gel. The addition of this thermally reversible healing agent gel was hypothesised to allow for a self-healing polymer matrix that does not contain thermally reversible cross-links (and would therefore not lose mechanical integrity while heating), but maintain the ability to repeatedly heal on the molecular level (through the discreet gel phase filling damage regions). The initial reports of this technique focused mainly on manually applying healing agent gel to the crack surfaces of fractured specimens, which demonstrated the ability to partially heal the polymer for at least five damage/ healing cycles.

UV initiated self-healing
Conceptually similar to the thermally reversible systems, bond reformation of damaged polymers can also be achieved by employing UV light as an external stimulus. One type of chemistry identified by Cho et al. as suitable for this type of healing is the [2z2] cycloaddition of cinnamoyl groups. Several tricinnamates were synthesised and photoirradiated to form cyclobutane containing cross-linked polymer films, and it was observed by FTIR, UV-vis spectrophotometry, fluorescence spectrometry and fluorescence microscopy that after crack damage of these films, the cyclobutyl groups at the damage regions cyclorevert to reform cinnamoyl functionalities (Fig. 14). 191 After 10 min of reirradiation with UV light, the cinnamoyl groups were able to recyclise to recover as high as y25% of the virgin film's flexural strength. 192 This relatively low quality of healing may limit cinnamate based polymers to film or coating applications, for which mechanical integrity is not always a primary concern. Furthermore, the most appealing UV sources (e.g. sunlight) may have small penetration depths unable to reach damage deeper than the small sample thicknesses typical of polymer films.
Polyurethanes incorporating a chitosan based crosslinker were also shown to heal in the presence of ultraviolet light. 193 Chitosan is a partially deacetylated derivative of the polysaccharide chitin, which is the primary structural component of arthropod exoskeletons and available in massive quantities worldwide. In these polymers, a commercially available chitosan saccharide was first functionalised with pendant fourmembered ring oxetane groups and subsequently cured with a trifunctional isocyanate and a polyethylene glycol chain extender to form a cross-linked polyurethane network. After damage and exposure to UV light, these polyurethane films were found to effectively heal scratches (Fig. 15). Through a series of control tests and ATR-FTIR and IRIR imaging of damaged regions, the authors hypothesised a two-step healing mechanism. First, mechanical damage opened the oxetane pendant rings to create two reactive ends. Then, exposure to UV radiation cleaved urea linkages in the chitosan backbone, which combined with the oxetane reactive ends to form new cross-links. This healing mechanism was qualitatively observed to both close and seal scratch damage. One promising feature of this system is that rapid healing (y30 min) of micron sized cracks is possible using UV light of similar wavelength and power density as the sunlight that reaches Earth's surface, but detailed information not present in this initial report (such as the polyurethanes virgin mechanical properties, quality of healing and repeatability of healing) should be addressed before the practicality of this technology can be judged.

Metal-ligand dissociation/association healing
As mentioned earlier, ideal design of polymers that heal via bond reformation lies with the placement of 'weak link' moieties, sacrificial bonds designed to break in response to stress, thereby preserving the generally irreversible organic covalent bonds. And with the appropriate chemistry, these 'weak links' can recombine to reform the virgin polymer network and heal damage. Coordination polymers, which contain metal-ligand bonds in their backbone, may satisfy these requirements given that metal-ligand bonds are generally known to be much more labile than their organic covalent counterparts, but, if the metal and ligand are chosen wisely, have a high enough thermodynamic affinity to rapidly reassociate.
As an important segue from molecular design to the demonstration of self-healing in a bulk polymer, a detailed understanding of the precise mechanochemical (i.e. chemical response to mechanical stress) dissociation of metal-ligand bonds in coordination polymers is required. Knowing how mechanical force affects metal-ligand bonding motifs is vital to ensure that the reversible 'weak link' in the polymer backbone is indeed preferentially broken during failure. As mentioned earlier, while macroscopic damage will probably not exclusively break these weaker bonds, it is intuitive that 'weak links' with weaker bond strengths and higher densities throughout the polymer backbone can better preserve the integrity of the irreversible bonds they are designed to protect. Yet despite the complexity of this effect, with potential implications towards quality and repeatability of healing, the precise mechanochemical response of the 'weak links,' relative to the irreversible bonds in the polymer, has gone largely overlooked in the bond reformation self-healing systems described outside of this subsection. But with regards to the mechanochemical bond cleavage in coordination polymers, which has been extensively studied, metal-ligand bonds are generally known to break at much lower mechanical loads than the polymer's irreversible covalent bonds. In one example, Kersey et al. studied the mechanical behaviour of a palladium/pyridine based coordination polymer using single molecule force spectroscopy. 194 Separate poly(ethylene glycol) polymers were attached to both the tip of an atomic force microscopy (AFM) probe and a SiO 2 substrate, which were end capped with pyridine based ligands. The substrate was then flooded with a dimethyl sulphoxide (DMSO) solution of a bispalladium complex, and the pyridine ligands coordinated to the metal to form polymer extending from the AFM probe to the sample surface (Fig. 16). When the functionalised AFM probe was retracted to the contour length of the polymer, bond rupture was consistently observed to occur through metal-ligand dissociation. Solutions of polymer functionalised N-heterocyclic carbene ligands have also been shown to efficiently cleave from silver and ruthenium complexes when subject to ultrasound induced mechanical stress, 195,196 which is especially noteworthy given that these carbene ligands are known to be excellent s-donors, often leading to high metal-ligand bond strengths. [197][198][199] In other work, Paulusse et al. synthesised diphosphine telechelic poly(tetrahydrofuran) polymers, and the phosphine endgroups were complexed to palladium or platinum halides to yield polymers with multiple metal coordination sites along the polymer backbone. [200][201][202] Stress was applied to a solution of the metal-phosphine coordination polymers via ultrasonication, and chain scission was observed to occur predominantly through metal-phosphine dissociation. Bulk healing of coordination polymers has been demonstrated with several different systems. In one example, linear polymers with pyridine pendant groups were cross-linked with bis-palladium or bis-platinum complexes to form hybrid polymer gels, which were observed to rapidly undergo ligand dissociation and reassociation in response to shear stress. 203 Varghese et al. synthesised lightly cross-linked, hydrophilic polymer gels containing carboxyl groups that could be healed by dipping fractured gels into a CuCl 2 solution. 204 When the damage surfaces are brought into contact, the carboxylate groups on the polymer chain effectively coordinate to the copper complex, resulting in significant recovery of mechanical properties. Kamplain and co-workers developed conducting organometallic polymers comprised of bis-N-heterocyclic carbenes and nickel, palladium or platinum metals. 205 These polymers exhibited conductivities on the order of 10 23 S cm 21 , and, after immersion in DMSO vapour at 150uC for 2 h to facilitate the metal-ligand association, were qualitatively observed to heal microcracks. 206 Healing via supramolecular assembly Polymers made from supramolecular self-assembly have also been shown to exhibit self-healing behaviour. The connectivity in these polymers partially relies on noncovalent interactions (e.g. hydrogen bonding, p-p stacking, etc.), which surely satisfies the omnipresent molecular design requirement of a 'weak link' throughout the polymer backbone.
Montarnal et al. reported the bulk synthesis of a hydrogen bonding supramolecular polymer by first reacting dimeric or trimeric fatty acid derivatives with diethylenetriamine, followed by subsequent reaction with urea. 207 This synthetic technique provided a batch of different monomers containing numerous hydrogen bond donors and acceptors (Fig. 17) that formed a supramolecular polymer through molecular recognition of these hydrogen bonding sites. Variations in the ratio of reactants were able to produce materials with a wide array of properties ranging from polymers behaving like semicrystalline thermoplastics to elastomers to associating liquids, 208 although self-healing tests were only conducted on an elastomer-like material with a glass transition temperature of 28uC. When this elastomer was fractured and the damage surfaces brought together, nearly full recovery of the virgin material's elongation at break was achieved after 180 min. Healing was only able to fully recover virgin properties when the fracture surfaces of broken specimens were brought into contact with each other immediately after fracture, with maximum possible healing decreasing at longer waiting times. This was attributed to the fact that broken hydrogen bonding groups eventually find partners within the broken part, leaving less hydrogen donors and acceptors available to associate with those across damage surfaces. 209  4,-diyl-iminocarbonyl-1,4-phenylenemethylene chloride), which was able to quickly recover the virgin gel's shear modulus after destruction of the network under high shear stress. 210 Based on the unlikelihood of p-p stacking of the aromatic groups in the oligomer backbone, due to the positive charge on the main chain, and an activation energy for gel reformation (210?4¡0?6 kJ mol 21 ) being significantly lower than expected for hydrogen bonding through water or amide groups, the authors proposed that the healing took place through chlorine ion mediated hydrogen bonding with water. Greenland et al. and Burattini et al. developed healable supramolecular self-healing polymers formed with p-p stacking interactions. 211,212 These materials consisted of a blend of low molecular weight polymersa chain folding polydiimide and a pyrenyl end capped polyamide or polysiloxane -that assembled to form flexible, self-supporting films with glass transition temperatures higher than 100uC. Rapid and reversible self-assembly of the electron poor aromatic groups along the backbone of the polydiimide and the electron rich telechelic pyrenyl groups (Fig. 18) was observed through model compound studies, computational modelling, and 1 H NMR, UV-vis and fluorescence spectroscopy. When solution cast films were cut, the broken pieces brought together, and healed at elevated temperatures, full and repeatable recovery of tensile modulus was possible in y5 min at 50uC and only a few seconds at 80uC. This healing was also qualitatively observed by ESEM to close and seal microcracks, albeit only above ambient temperatures, while control tests with the polydiimide and a phenyl end capped polymer (as opposed to the telechelic pyrenyl group) exhibited no healing (Fig. 19). This control test was vital to rule out viscous behaviour of the polymer melt as a major contributor to healing. Instead, healing was attributed to a disruption of the pp stacking interactions at elevated temperatures that allowed the polymer to fill the damage region and reform the p-p non-covalent interactions upon cooling. This was confirmed by rheometric analysis of the films, which showed a significant drop-off in modulus and melt viscosity above a critical temperature, which in the case of this work was y40uC. 213

Virgin property strengthening
A drastically different methodology relatively new to the field of self-healing polymers is the development of materials that apply local strengthening mechanisms in response to stress, but before virgin material failure.
These polymers incorporate mechanophores (mechanochemically active units) along their backbone that are designed to remain dormant when unperturbed, but impart additional polymerisation and/or chemical crosslinking to localised portions of the bulk polymer that are under stress. While this research is still in its early stages, and to date bulk polymers that conclusively demonstrate virgin property strengthening have yet to be reported, the current efforts to develop these systems are discussed below. We will mainly focus our discussion of mechanophore chemistry on current efforts to impart a virgin property strengthening mechanism to bulk polymers; several good reviews encompassing a larger scope of mechanochemistry are available elsewhere. [214][215][216] Many mechanophores are designed to generate active metal catalysts when placed under stress. In fact, the coordination polymers mentioned above (that heal via reversible metal-ligand association/dissociation) can potentially be used as mechanophores, [194][195][196][200][201][202][203]205,206 assuming that the force induced dissociation of their metal-ligand bonds activates an otherwise dormant precatalyst. Sijbesma and co-workers have reported two mechanophore linked polymers that satisfy this requirement. 196 In the first case, solutions of silver complexes of polymer functionalised N-heterocyclic carbenes were able to efficiently catalyse the transesterification of vinyl acetate and benzyl alcohol after sufficient ultrasound induced force was applied to cleave one of the metal-carbene bonds (mechanophore 1 in Table 2). And second, solutions of ruthenium bis-carbene complexes with polymer functionalised ligands were able to initiate olefin metathesis reactions, both ring closing metathesis and ring opening metathesis polymerisation, after stress induced ligand dissociation (mechanophore 2 in Table 2). These reports have only focused on the catalysis of small molecules in solution, and have not yet been used to demonstrate selfstrengthening. It would be interesting to determine if the ultrasound stress induced chemistry of metal catalyst mechanophores like these can translate into virgin property strengthening of bulk polymers, potentially by triggering cross-linking reactions on nearby polymer chains.
Mechanophores that form stable, well defined radicals under stress are especially attractive for self-strengthening applications. In unsaturated polymers, for example, if well defined radicals can be mechanochemically formed before material failure, they can potentially cross-link neighbouring double bonds through radical polymerisation. One type of chemistry investigated for this purpose is the Bergman cyclisation of enediynes to form arene diradical intermediates, which are known to produce high molecular weight polymer in the presence of common unsaturated monomers. 217,218 The Bergman cyclisation is generally known as a thermally activated rearrangement that, when carried out on a strained 10membered ring enediyne, often occurs under mild heating conditions. But COGEF (COnstrained Geometries simulating External Force modelling has predicted that under mechanical load, the enediyne ring can be distorted towards its cycloaromatisation transition state, lowering the activation barrier enough to spontaneously cyclise at ambient temperatures (mechanophore 3 in Table 2). 219 In an attempt to empirically prove this model, enediyne rings were incorporated into bulk poly(methyl methacrylate) as a cross-linker, and the cross-linked polymer was swelled with methyl methacrylate monomer. It was hypothesised that, if the dimensional expansion of the swollen polymer provided adequate force to elongate the enediyne cross-links enough to spontaneously cyclise, the resulting cyclised adducts could initiate detectable amounts of polymerisation of the methyl methacrylate monomer. However, after failing to detect any significant polymerisation of the liquid methyl methacrylate, it was concluded that swelling provided inadequate mechanical force to induce the spontaneous cyclisation. In other work, radicals were generated from the ultrasound induced homolytic extrusion of nitrogen from an azo-functionalised poly(ethylene glycol) (mechanophore 4 in Table 2). 220 This mechanochemistry was not demonstrated in a bulk polymer, and the ability of the resulting radicals to initiate vinyl polymerisation was not evaluated, but considering the efficiency of the radical formation and the structural similarity of these azo moieties to that of the well known radical vinyl initiator azoisobutylnitrile (AIBN), this mechanophore may be useful to initiate radical crosslinking mechanisms in bulk unsaturated polymers. Although the ongoing development of different mechanophores does indeed look promising for future applications, several issues related to shifting the technology from solution based systems under ultrasound induced forces to bulk polymers under macroscopic damage must first be addressed. For example, while ultrasound induced stress activation of the polymer linked mechanophores mentioned above has been well documented, it is not known how well this behaviour will translate to bulk polymers. In fact, there is very little precedent at all for the precise mechanochemical transformations of any type of mechanophore in a bulk polymer under mechanical load. In one of the few promising examples, PMA and PMMAlinked spiropyran mechanophores were developed that, when placed under load, undergoes a reversible electrocyclic ring opening reaction to form the corresponding merocyanine (mechanophore 5 in Table 2). 221 This reaction was not designed to initiate any additional polymerisation or cross-linking mechanisms, but a marked fluorescence and visual colour change of the activated mechanophore allows for an easy approach to map the mechanochemical transduction throughout a polymer. And indeed, when bulk polymers containing the spiropyran mechanophore were fabricated, both tensile and compression load induced a colour change concentrated at regions of high stress (e.g. Fig. 20 shows a yellow to red colour change at the neck region of a dogbone specimen, which becomes more prominent at higher tensile loads). 222 But while this precedent does bode well for the prospect of incorporating reactive, self-strengthening mechanophores into bulk polymers, it is still unknown exactly how different mechanophores would respond to macroscopic stress or how effective their solid state strengthening mechanisms would be.
The desired periodicity of mechanophores along a polymer backbone may also differ in bulk polymers. In many of the studies utilising ultrasound induced forces, only one mechanophore is placed near the centre of the polymer chain. This is done predominately for analytical reasons: the presence of only one mechanophore greatly simplifies data interpretation, and the solvodynamic shear stress resulting from ultrasonication is generally localised near the centre of the polymer. 223,224 But in bulk polymers, stress distribution along polymer chains may follow different trends, and surely cross-linked polymers, without a traditional midpoint, would not greatly benefit from specific mechanophore placement. Hence, it is prudent to carefully weigh the advantages of varying the number of mechanophores present in a polymer. Lower densities of mechanophores along a polymer backbone, such as in many of the reports using ultrasonication, may lessen the scale-up obstacles of 20 A bulk PMA linked spiropyran dogbone specimen under tensile stress, which mechanically converts the spiropyran moiety to the red coloured merocyanine at increasing loads. Reprinted with permission from Ref. 222 222 synthetically challenging mechanophores. Furthermore, sparse placement of mechanophores along the backbone would probably reduce their effect on the virgin properties of the polymer and allow for conventional processing techniques to be used. However, higher densities along a polymer chain would intuitively increase the amount of mechanophores activated under stress, thus increasing initiation sites for additional polymerisation and cross-linking. Lenhardt and coworkers have taken the first steps towards answering these questions by investigating the activity of higher densities of non-scissile mechanophores in a single polymer chain. 225 In their work, reaction of polybutadiene with aqueous NaOH in chloroform conveniently afforded abundant quantities of gem-dichlorocyclopropane mechanophores (as high as 72% conversion of olefins to cyclopropyl groups) along the backbone of the polymer. When subjected to ultrasound induced force, these mechanophores undergo electrocyclic ring opening to form 2,3-dichloroalkenes (mechanophore 6 in Table 2), which, due to the high density of mechanophores, was easily monitored by 1 H-NMR (Fig. 21). The utility of the resulting dichloroalkenes mostly lies with their ability to map stress distribution along the polymer chain, but a high density of self-strengthening mechanophores with similar scalability would surely be of practical interest.

Virgin property reduction
In order for self-healing materials to be considered feasible for real applications, it is important to understand how the self-healing components affect the virgin properties of polymers they are meant to heal. But while self-healing studies are very much applied research, and the implications of virgin property reduction are therefore of paramount importance, such information is often lacking in the literature. Here, we summarise and discuss general trends observed with virgin polymer property changes when different types of healing components are added. Table 3 outlines much of what is known regarding these property changes.
In most reports of microcapsule based self-healing mechanisms, increasing loadings of liquid filled capsules significantly toughens the composite matrix, relative to  29,226 Epoxy resin filled capsules/solid imidazole catalyst 37 Epoxy resin/mercaptan filled capsules 43 Phase separated pEMAA particles 117 DCPD filled capsules/solid, wax encased Tungsten catalyst 85 Shape memory alloy wires 79,80 Epoxy resin filled capsules/matrix dissolved imidazole catalyst 38,40,41 Matrix dissolved thermoplastic polymer 155 Strength Epoxy resin-filled capsules/matrix dissolved imidazole catalyst (CAI) 42 PDMS containing capsules (tear) 46 Epoxy resin filled hollow fibres: large pitch spacing (flexural) [133][134][135] Epoxy resin filled microvascular network (flexural) 153 Epoxy resin filled hollow fibres: small pitch spacing (flexural) [132][133][134] Epoxy resin/mercaptan filled capsules (flexural, tensile) 43  infused sisal fibre (flexural) 45 Matrix dissolved thermoplastic polymer (storage) 156 Epoxy resin/mercaptan filled capsules (flexural, Young's) 43 DCPD filled capsules/solid Grubbs' catalyst (shear) 227 Phase separated Epoxy polymer particles (tension) 115,116 Phase separated pEMAA particles (flexural) 117 Phase separated poly(caprolactone) (storage) 119 Tg Phase separated poly (caprolactone) 117 Matrix dissolved thermoplastic polymer (storage) 156 Mass penalty Epoxy resin filled microvascular network 153 the neat polymer matrix, until a critical loading is reached, after which point further increases in capsule loading decreases toughness. In DCPD filled microcapsules, for example, this toughening was attributed to hackle markings and subsurface microcracking at fracture surfaces that were significant only with good capsule matrix adhesion. 226 But this effect does not seem to be universal, as several microcapsule systems reported minimal changes 85 or decreasing values of fracture toughness 38,40,41 with increasing capsule loading. However, data in these reports suggest that this behaviour may be the result of a complex interplay between the toughening effect of the microcapsules and that of other toughness decreasing components (e.g. solid or matrix dissolved catalysts), resulting in overall decreases in fracture toughness. Reports on the effect of microcapsules on other virgin properties are scarce, but evidence suggests that the capsules have negative effects on modulus 43 and strength, 227 which is consistent with what is expected for hollow fillers and voids. The effect of resin filled hollow fibres or microvascular networks on virgin polymer properties depends strongly on the placement of the vessels throughout the polymer matrix. For example, hollow fibres with small pitch spacings throughout plies have a detrimental effect on the strength of the composite. 132 Increasing this pitch spacing, or analogously increasing the spacing between vascular channels, can eliminate the effect these vessels have on the polymer's virgin properties, but this often comes at the cost of diminished healing ability. [133][134][135]228 Additionally, an increased mass penalty is observed when incorporating microvascular networks throughout a polymer matrix, 153 but mass penalties can potentially be reduced by adjusting the vascular network architecture. 144,145 Several approaches to self-healing involve the development of new polymer matrices where the healing functionality is inherently part of the virgin material (for example, see the above section on polymers that heal via bond reformation). While these systems do not contain traditional healing additives, and therefore cannot be discussed in the context of virgin property changes, it is at least important to consider whether or not their virgin properties are adequate enough to substitute for the traditional engineering polymers they intend to replace. The DA based thermoset polymer shown in Fig. 11 is appealing in this respect, given that its Young's modulus and tensile, compression and flexural strengths are comparable to that of commercial epoxies and unsaturated polyesters. 173 Also, most of the epoxies and common thermoplastic polymers with reversible DA based cross-links possess mechanical properties superior to their respective non-cross-linked polymers. [180][181][182][183][184][185][186][187][188][189][190] However, several other custom made polymer matrices outlined above are either inferred or explicitly stated to have significantly weaker mechanical properties than conventional structural materials and may have limited applications.

Healing evaluation
The quality of healing, generally referred to as 'healing efficiency' and denoted as g, is most often defined as the per cent recovery of a virgin material property (equation (1), P5material property). Details regarding the specific material properties used for this calculation are discussed below, but first it is important to mention some factors that may complicate any direct interpretation of healing efficiency values. First, it should be noted that some results are biased by the fact that very brittle polymers are chosen as reference materials. That is, healing efficiencies using polymers employed in real applications, which often utilise stronger and tougher polymer matrices, will likely be lower due to their superior material properties. Also, the effect of the healing additives on virgin polymer properties may erroneously increase or decrease healing efficiency. As seen in Table 3, most healing additives have either a beneficial or detrimental effect on the virgin material properties of the polymers in which they are incorporated, which makes full recovery of these properties either more or less difficult respectively. This is especially significant when comparing healing efficiencies at different loadings of healing additives. For example, it is often the case that higher loadings of healing components imparts both decreasing virgin properties and increasing healing ability to a polymer, which can greatly exaggerate the effect of adding higher amounts of healing components. For this reason, alternate definitions of healing efficiency are sometimes used that normalise healed material properties by the virgin properties of polymers without added healing components (equation (2), P5material property). At any rate, virgin property reduction and material properties of both the virgin and the healed polymer should be reported concurrently with healing efficiency in order to adequately evaluate the quality of healing g~P Healed P Virgin |100% (1) g~P Healed P NoHealing |100% In most cases, evaluation of healing requires a controlled and measurable application of damage to the virgin polymer, followed by a similar application of damage to the healed polymer. However, there lacks a unifying standard of how to best apply this damage, and as such, several different damage methods have been used to evaluate healing. Most often, damage is induced through various different tensile, compression or bending test protocols (for example, see Fig. 22), but numerous different damage modes have also been utilised: impact, cutting, scratching, sawing, needle puncture, nail puncture, hammering and indentation. Furthermore, different studies employ different extents of damage using these testing protocols, which have ranged from applying only enough stress to induce measurable cracking and delamination to catastrophic failure of specimens into multiple pieces. Healing efficiencies reported on samples that are catastrophically failed are probably a more reliable measure of healing given that the virgin failure and subsequent healing event can be easily controlled along a single fracture path. But causing only partial damage may be the best way to mimic the more frequently observed and realistic failure conditions such as microcracking and delaminations deep within the material, the healing of which was the original intended purpose for self-healing polymers. 28,229 Table 4 lists many of the material properties that have been used to quantify healing efficiency. Which healing efficiency definition is chosen may depend on the virgin polymer properties, desired failure mode, self-healing mechanism, etc. For example, defining healing efficiency as the recovery of fracture toughness allows for a measurement of mode I crack opening, which is similar to the mechanism of microcrack growth that is often observed in real applications. Healing efficiency based on tear strength or strain energy is a better fit for elastomers or plasticised polymers that undergo highly ductile failure. Recovery of impact, compression after impact or flexural after impact strength may be more appropriate metrics for evaluating fibre reinforced polymers since impact damage is well suited to target delamination and fibre-matrix debonding failure mechanisms.
As mentioned above, healing efficiencies of specimens only partially failed must be evaluated with some scrutiny since the reported material properties of healed specimens may be a combination of both healed and virgin portions of the polymer. To better illustrate this point, we examine work done by Pang and Bond,131 in which healing agent filled hollow fibres were incorporated into a fibre reinforced composite. After low energy impact indentation and a short healing time, testing specimens were evaluated with a four-point bend flexural testing protocol. It was observed that samples without impact damage, samples with impact damage but no allotted healing time, and samples with impact damage and healing had mean flexural strengths of 623?9, 548?1 and 603?6 MPa respectively. The traditional definition of healing efficiency shown in equation (1)

|100%
(4) Jones et al. approached this problem not by evaluating a traditional material property, but instead by quantifying the growth of delamination area. 155,156 This was accomplished with an image analysis program that, after impacting composite panels, was able to measure the damage area before (A Impacted ) and after a healing cycle (A Healed ), and healing efficiency was calculated as in equation (5) g~A Healed A Impacted The traditional definition of healing efficiency shown in equation (1)  In the first, which is shown in equation (6), fatigue life extension l is evaluated as a function of the total number of cycles to failure for a sample with self-healing capabilities (N Healed ) normalised by the number of cycles to failure of an otherwise identical specimen without healing (N Control ). [30][31][32] The second approach, developed by Lewis and co-workers 33,34 and shown in equation (7), is similar to the first, but instead utilises the mean fatigue crack propagation rate (FCPR)  192 ) and the presumably low stiffness of Ghosh and Urban's UV healable chitosan/ polyurethane based polymers (the mechanical properties of this system were not explicitly stated, but inferred from their ability to readily undergo unassisted crack closure 193 ) may further limit them to coating applications, where structural integrity is not of the upmost importance. Similarly, other self-healing polymers known to have relatively low healing efficiencies or weak virgin mechanical properties may also be limited to film or coating applications. On the other hand, several self-healing systems can only be used in bulk polymers or composites. For the most part, this restriction comes from the mass and volume requirements of certain healing additives being greater than typical thicknesses of thin films. For example, the thinnest reported specimens containing microvascular networks are on the scale of 5-7 mm thick, 150,153 which is at least 1-2 orders of magnitude thicker than required for many thin coatings. Healing agent filled hollow fibres are also limited to bulk polymers for similar reasons. Microcapsule based self-healing polymers, however, are not as limited as the microvascular and hollow fibres systems. While microcapsules for self-healing polymers are most efficient on the 100-200 mm size scale, 72 which may be too large for thin films, recent reports of healing agent containing nanoscale sized microcapsules 89 may open the field of microcapsule based self-healing to thin coating applications.
Although probably best suited for bulk applications, fabricating self-healing systems containing phase separated thermoplastic polymers as healing agents into thin films may not have as many obstacles as several of the above systems. Although this has not yet been reported in the literature, the phase separated EMAA and epoxy particles reported by Meure et al. 117 and Cho et al. 115,116 respectively, could presumably be reduced to the nano-or micron size scale using standard melt processing techniques, before their incorporation into a self-healing polymer matrix. Luo et al.'s epoxy polymers containing reaction induced phase separated PCL are especially translatable into thin film applications, as the distribution of the PCL inherently phase separates on the micron scale, 119 thereby eliminating the need for any additional processing steps to reduce the healing agent size.
Healing in presence of structural reinforcements A significant failure mode in fibre reinforced polymers, in addition to the brittle microcracking behaviour of their polymer matrices, is their susceptibility to delamination damage. Given that this delamination disrupts the matrix-fibre bonding requisite for composite materials to take advantage of the superior structural properties of their embedded fibres, a self-healing functionality is extremely attractive for composite materials. 232 However, the addition of structural fibres to a polymer often changes the healing performance, relative to that of the neat matrix. Below, we discuss these known effects of structural fibres on healing capabilities. While healing in fibre reinforced polymers is prevalent in many of the different self-healing methodologies presented above (notably those containing epoxy/hardener filled microcapsules, [38][39][40][41][42] phase separated thermoplastic polymers, 115,116 and healing agent filled hollow fibres 128,[130][131][132][133][134][135], the most comprehensive comparisons and discussions of the effect of the presence of structural reinforcements on healing have been reported with systems that heal via embedded DCPD filled microcapsules and Grubbs' catalyst particles. One difference between fibre reinforced polymers and neat polymer matrices containing a healing functionality is the effect of the healing components on the virgin material properties. Generally, adding structural fibres to a neat polymer matrix increases fracture toughness by forcing crack growth to follow a more tortuous path around the fibre architecture. Fibre reinforced polymers with self-healing functionality are no exception to this, with higher fracture toughness values for complex fibre weaves than unidirectional fibres. 233 Based on what is known about the toughening that DCPD filled microcapsules impart to neat polymers, 226 one would initially expect that adding microcapsules to fibre reinforced polymers would further increase their toughness. However, the opposite trend, a marked decrease in fracture toughness, was observed when adding the same microcapsules to fibre reinforced polymers. 234 This was attributed to the microcapsules increasing the thickness of the interlaminar layer, relative to microcapsule free layers, which has been observed elsewhere to similarly affect composites' fracture toughness. 235 The average fracture toughness based healing efficiency of a fibre weave reinforced epoxy polymer containing DCPD filled microcapsules and Grubbs' catalyst was reported as 38%, 234 considerably lower than the 90% healing efficiency reported elsewhere for an otherwise similar neat polymer matrix. 29 While this decrease in healing is likely not as dramatic as a direct comparison of the two values would indicate, due to the virgin property differences between the fibre reinforced polymer and neat polymer matrix, it is clear that incorporation of the fibre weave is detrimental to the healing mechanism. It was concluded that the lower healing efficiency was related to a lack of healing additives (catalyst and microcapsules) present at the fibre/matrix interface, where a majority of the damage occurs in the form of delamination. This resulted in less overall healing agent in the damage area, and consequently, lower healing. Furthermore, maximum healing was achieved in the fibre reinforced polymer after y48 h, which is significantly longer than the time required to reach maximum healing in a neat epoxy polymer (y12 h 29 ). The longer healing time required for fibre reinforced polymers indicates that the curing kinetics of healing agent is much lower than in the neat polymer, which is partially related to the aforementioned lower amounts of catalyst present at the damage surfaces of the delaminations. Also, the slower reaction kinetics may be related to the high thermal mass of the fibre weave contributing to a lower local temperature at the delamination surfaces, relative to the neat polymer's fracture surfaces, where the healing initiated. This longer healing time in fibre reinforced polymers can potentially allow the liquid DCPD healing agent more time to either volatilise or diffuse into the polymer matrix before the onset of significant reaction, which may also be contributing to the lower fibre reinforced polymer healing efficiency.
Intuitively, smaller sized healing components with better dispersion throughout the polymer matrix and around the fibres could improve both the healing efficiency and the healing time in fibre reinforced polymers. 234 Sanada and co-workers attempted to better distribute the self-healing components around the fibres in a unidirectional fibre reinforced epoxy using a different fabrication technique than in the systems discussed above. 236,237 The fibres were first coated with an epoxy resin mixture containing Grubbs' catalyst and relatively high loadings of DCPD filled microcapsules (as high as 40 wt-%). The coated fibre strands were partially cured, placed in a mould and impregnated with an epoxy/ hardener mixture (not containing healing components). This technique allowed for a high concentration of healing components localised around the damage prone fibre/matrix interfaces without the need for high overall loadings of catalyst and microcapsules throughout the bulk of the polymer. In another approach to this problem, Grubbs' catalyst was not directly embedded in the composite matrix, but instead coated on the outside surface of the fibres. 137 Additionally, recent advances in fabricating smaller microcapsules 89 have already proven fruitful in increasing healing efficiencies in fibre reinforced polymers. 238,239

Conclusions
The multidisciplinary field of autonomic healing materials has provided several different techniques to impart a self-repairing function to polymers and composites. In this review, we have summarised these current research thrusts and discussed several issues related to translating the technology to practical applications, such as virgin polymer property changes as a result of the added healing functionality, healing in thin films v. bulk polymers and healing in the presence of structural reinforcements. There are a number of variations in the self-healing systems described above that are beginning to garner significant interest, but have thus far been reported infrequently in the literature and therefore was not discussed in great detail herein. [240][241][242][243][244][245][246][247][248] Additionally, the burgeoning field of computational modelling of the different healing mechanisms is continually providing insights into ideal polymer and composite design parameters for, among other things, improved scalability and healing capabilities. [249][250][251][252][253][254][255][256][257][258][259][260][261][262][263][264][265][266][267] While future endeavours will undoubtedly improve current healing mechanisms towards efficient, fully autonomic and biomimetic healing materials, as well as yield other approaches to imparting autonomic repair, future research thrusts need to concentrate on issues related to employing self-healing materials in industrial applications. Several companies are beginning to lead the efforts to market and produce self-healing polymers (such as the company Autonomic Materials, which is developing microcapsule based self-healing elastomers, thermosets and powder coatings, 268 and Arkema Inc., which is currently producing polymers that heal via supramolecular assembly 269 ), but several issues that are rarely discussed in the literature, such as economic feasibility and long term 'healability' of the different healing mechanisms, need to be addressed before selfhealing materials can begin to replace current polymers and composites.