Electrical ageing of composites: an industrial ... - Schneider Electric

Technical collectionElectrical ageing of composites:an industrial perspective2007 - Conferences publicationsS. W. Rowe

Schneider Electric 2007 - Conferences publicationsand 40µm, even though we had convinced ourselves asto their absence. In other words considering fig1. we arealready at time t 3 whilst we thought we were at t 0 .Shortly after manufacturing, the gas pressure in a voidmight be less than atmospheric and might not be air.This situation would be even worse, as the mean freepath would be greater. Bigger undetectable voids couldthus exist. They would however gradually fill up toatmospheric pressure via air diffusion, and thereafterdischarge violently. This might take days or evenmonths. This point is worth pondering… Increasingsensitivity and/or applied field, improves things, butremember that for an avalanche to grow normally, theelectrons must be able to freely follow the direction ofthe electric field lines. If this is not the case, such as insinuous/tortuous channels, even longer “elongatedvoids” might not be detectable.One is justified in wondering whether the precedingdiscussion has any real practical importance. Considertherefore, the following case of a 6µm length void filledwith atmospheric air. This corresponds to a maximumof 3 mean free paths. In the pure gas we would thus notexpect to produce more than 8 electrons by collisionalionisation, no matter how high the applied field.However each of these electrons will strike thedownstream surface with an energy corresponding tothe voltage drop across the final accelerating step. For asmall gas void in epoxy with an applied electric fieldequal to 2kv/mm, the “average” impact energy will beabout 20eV. This might be enough to do damage to thepolymer chain structure. Under 50HZ ac conditions, thisbombardment will continue as a repetition of “bursts”throughout the life of the insulation. Assuming anaverage of only 2 discharges per cycle, leads to about3×10 10 impacts per year or 10 12 over the 30 year lifespan. Hence an almost “negligible” and virtuallyundetectable discharge might have a strong cumulativeeffect in practical applications. In conclusion, we cangenerally not be sure that our samples do not containvoids in the 0 – 30µm range.Electrical ageing driven by this might be occurringwithout us ever suspecting. However, the basic theorysays that ageing is via extension of the initial microvoid,so at some instant in time the PD level should riseabove the noise level of the experimental set up, andcontinue to rise. This experiment is so simple to do thatone can wonder that it has not been done many times,because without this result the “standard scenario” cannot be validated. To try to obtain this validation, wehave had an accelerated ageing experiment running nowfor about 18 months at 6 times the normal applied field,with samples at temperatures of 40°C, 80°C and 120°C.So far we have observed no evolution of the PD level.ELECTROLUMINESCENCEWe also decided to study the electric field induced lightemission of our filled samples, to see how this couldhelp us in our quest. We rapidly discovered that photonemission exists and rises rapidly with electric field,Fig.7. We also discovered a strong thermally stimulatedluminescence for higher temperatures, which was linkedto oxidation of the hardener, (removed in the curvesshown), [4]. Although plotting on a linear scale seemedto indicate the presence of a threshold, this becomesquestionable when plotting with Log-log coordinates asshown. For ambient temperature the signal falls belowthe noise level of the detector for about 2kV/mm.In reality the overall photon detection “efficiency” is notsimple to determine. As photons are emitted in alldirection we can only hope to capture a smallpercentage depending on the design of the opticalarrangement. Furthermore the efficiency of the detectorsthemselves is less then 20%, so the sample could beproducing more than ten times as many photons as thosewe actually “count”. Also we are intrinsically assumingthat photon absorption by the polymer itself isnegligible for our 1mm thick samples. All the same let’smake the following rough estimation. 1 photon detectedper second corresponds to about 10 9 photons per year.Fig.7: Electric field induced light emission for a filled sampleThe question we now have to ask ourselves once moreis, “has this light emission got anything to do withelectrical ageing”? For instance, a light emitting diode,(LED), produces extremely high numbers of photonspermanently, but we do not normally consider that theproduction mechanism involved actually “ages” thecomponent. Knowing the emission wavelength andusing Planck’s constant, we can estimate the energyassociated with a single photon. For a wavelength of500nm, this turns out to be about 2.5eV. To decide ifthis has anything to do with electrical ageing we need toknow what mechanism is responsible for liberating thisenergy as light, and how it was gained from the electricfield. Taking an average hopping length of 5nm, we findthat for our nominal conditions, the voltage differenceper hop, is about 10mV. The average carrier impactenergy is thus about 0.01eV which seems unlikely toproduces the observed effect.So where does this, “2.5eV” light come from ? Carrierrecombination ?, Deep trap filling ? One option is that itis linked to the presence of micron size “voids”. In thiscase some carriers would be able to make “jumps” ofseveral microns, rather than a few nanometers and404

Schneider Electric 2007 - Conferences publicationshence gain much more kinetic energy. We do not needto assume the development of a large scale electronavalanche or even strong, gas phase ionisation because1 photon per second on average is only one per 50cyclesof power frequency voltage which is not very much…Clearly, if voids of less than 40µm existed, avalanchesof up to 10 6 electrons could exist, as described above. Anumber of these electrons could give rise to lightemitting mechanisms in the gas or at least when theybombard the internal surface of the void. Depending onvoid shape, on average they might be able to gain 20eVwith the tail of the statistical distribution containingeven more energetic electrons. ALL these electronsmust give up this kinetic energy on striking the innersurfaces of the void. Under these circumstances onemight expect a considerable amount of emission havinga spectrum corresponding to one, or both, of the twomechanisms mentioned above.Recent experiments with porous polypropylene samplesshows that partial discharges of this type are observable.Once again, at the end of the day we come to theconclusion that even these result do not really help inpinpointing the origin of electrical ageing. We need toAN ALTERNATIVE SCENARIOAt this stage, the convergence of a number of ideas andindications, led us to reconsider the validity of thestandard ageing scenario described above, for our highlyfilled samples. Clearly the extremely heterogeneousnature of the structure is not taken into account, Fig 10.It also neglects the well documented presence of largequantities of absorbed water, (0.1 – 5% weight).The results of dielectric spectroscopy on filled andunfilled samples which have absorbed an identicalamount of H 2 O relative to the resin part, show thevalidity of such concerns. These results point to a multiparametermaxwell-wagner type interfacial polarisationmechanism, Fig.8. Other data lead us to consider thepossibility of an inter-grain mechanism of carriertransport and interface accumulation, Fig10.One of the best known influences of water incomposites is plastification and the correspondingreduction of the glass transition and various physicalproperties. Working with both standard and nano-filledsamples we note a reduction of Tg linked directly to thequantity of water “in the resin part”. This reductionreaches 20°C for saturated samples, (corresponding to aTg reduction of about 30% for our samples, tg = 65°C),Fig.9.Furthermore, the principal mechanism considered to beresponsible for the mechanical failure of “joined”materials is the action of water at the interfaces, [5].Consequently, our latest results, which indicate thegrowth of water layers of about 2nm around each offiller grain, are of particular significance [6] [7].Accordingly we are now concentrating our efforts intesting an “Alternative Ageing Scenario”.This scenario is specific to highly filled materials and isbased on the progressive degradation of the filler/matrixinterfaces by water molecules.Fig.8: Dielectric parameters with absorbed H 2 Oknow exactly the physical mechanism behind thephoton production, before we can decide if it hasanything to do with electrical ageing. To confirm thesethoughts, we have initiated a series of experiments onphoton counting using artificial voids in the 10 – 100µmrange.Fig.9: Glass transition versus atmospheric H 2 O contentN9 =9% nano-filler, LD41=59% micro-fillerOur basic hypothesis is that a small fraction of the waterwhich diffuses into our composites, reaches thepolymer/filler interfaces where it accumulates. The405

Schneider Electric 2007 - Conferences publicationspresence of these water molecules reduces themechanical binding between the polymer matrix and thefiller grains.mechanism still escapes us. Without knowledge, of this,technical progress is at best a case of trial and error.This is no longer acceptable in the modern industrialenvironment and we are therefore very hopeful thatfollowing the “Alternative Ageing Scenario”, will bringus closer to this elusive objective.THANKS100 µmFig.10: Xray tomographic “slice” from a 3D scan of atypical sample. The white regions are filler particlesMy sincere thanks go to the numerous people who havetaken part in this long trek through lands where themaps available were often of approximate precision andsometimes even misleading. Starting with the youngest,thanks then to, Emily Brun, Eddy aubert, Chen Zou,Naum Pérez, Christophe Guillermin, Olivier Gallot-Lavallée, Hoang Nguyen, Stephane Robiani, BrigitteOhl, Francois Trichon, Gilbert Teyssedre, Pascal Rain,Gisele.teissedre, Alain Sylvestre, Christian Laurent andJohn Fothergill…This accumulation of water thus modifies both themechanical strength, the dielectric behaviour and theelectric strength of these tiny regions. This degradationis “diffuse” and occurs at ALL interfaces at the sametime. Gradually a network of semi inter-connectedpathways builds up through the labyrinth of fillerparticles.Finally “percolation” of these degraded regions occursand an electrical path appears through the sample.Conduction current is then initiated and a runawaymechanism leads to electrical breakdown, Fig. 11. Weare initially assuming that the water molecules are“bound” and that heating due to current flow does notdry the minute interface regions and lead to stabilisationwhich would inhibit full breakdown.This “Alternative Ageing Scenario” for explaining theelectrical failure of highly filled HV insulation is thusfundamentally different from other explanations. Inparticular, the ageing precursor has nothing whatever todo with the electric field or current or charge etc. Theelectrical field only steps in, when the damage has beendone by the water, which is inevitably always present.In other words, if this scenario were proved exact itwould imply that “electrical ageing does not exist”.Time will no doubt clarify, which, if either, of thesescenarios is the correct one.CONCLUSIONOver the years a considerable amount of research hasbeen carried out aimed at uncovering the processesresponsible for electrical ageing of the filled polymers,used for HV applications. The humbling conclusion isthat although we have discovered many interestingproperties of these materials, we are not much nearerour goal than at the outset. The principal drivingFig.11: Representation of the “Alternative AgeingScenario”, showing H 2 O degraded, filler interfaces.REFERENCES[1] C. Guillermin Thesis;"Vieillissement électrique etthermique d'un composite résine époxyde-silice"University of Grenoble I -Mai 2004[2] P Rain C Guillermin, and S W Rowe«Characterization of electro-thermal aging of a filledepoxy resin IEEE ICSD 2004 pages 351-354,[3] C Guillermin, P Rain and S W RoweTransient and steady-state currents in epoxy resinJ. Phys. D: Appl. Phys. 39 (2006) 515-524[4] Olivier Gallot-Lavallée, These, University PaulSabatier Tououse III, 2004[5] A. J. Kinloch, In “The mechanics of adhesion” EdD. A. Dillard, Ensevier, 2002[6] Chen Zou, J. Fothergil and S. Rowe, ICSD,Winchester, 2007[7] Emily Brun, Ibid406