IEEE Transactions on Power Delivery, Vol. 11, No. 4, October 1996 1789 AN ASSESSMENT OF THE RELIABILITY OB IN-SERVICE GAPPED SILICON-CARBIDEDISTRIBUTION SURGE ARRESTERS M. Darveniza, Fellow IEEE, D.R. Mercer, R.M. Watson University of Queensland Australia, 4072 Abstract - Although gapped silicon carbide arresters are no longer purchased by electricity authorities, they still form the majority of the very large number of distribution arresters in service in Australia and many other countries. Most of the arresters of this type are now over ten years old and many are much older. So the question must be asked - what is to be done with this ageing and outdated class of arresters? Extensive Australian studies in the 1960's had revealed that internal degradation resulting from inadequate seals was the predominant cause of failure of gapped silicon carbide arresters. This paper describes the results of a recent investigation. Electrical testing showed that after about 10 years of service, there is a marked upturn in the number of arresters with unsatisfactory insulation resistance, and after about 13 years of service, a marked upturn in the number of arresters with reduced power frequency sparkover level. Inspecition of the internal components of dismantled arresters confirmed that the likelihood of significant degradation increased markedly with years of service, and was evident in almost 75% of arresters with 13 years or more of service. The authors therefore recommend that all gapped silicon carbide arresters with 13 or more years of service be progressively replaced by modern metal oxide arresters. I. INTRODUCTION An earlier extensive study [ l ] of distribution surge arresters and transformers was motivated by excessive failure rates in service. That study revealed that the predominant cause of failure of gapped silicon carbide surge arresters was internal degradlation due to moisture ingress resulting from inadequate seals. The condition of the arresters was assessed by a combination of power frequency voltage withstand test, impulse voltage sparkover test, measurement of insulation 96 W M 013-3 PWRD A paper recommended and approved by the IEEE Surge Protective Devices Committee of the IEEE Power Engineering Society for presentation at the 1996 IEEE/PES Winter Meeting, January 21-25, 1996, Baltimore, MD. Manuscript submitted August 1, 1995; made available for printing November 13, 1995. resistance, and dismantling and inspection of internal components. Since then, manufacturers have improved the design and testing of arrester seals, and in-service arrester failure rates reported by electricity authorities are now generally lower than those which motivated the earlier investigation. Subsequent investigations carried out in Norrh America [2, 31 also showed that the dominant cause of degradation and failure of distribution arresters was moisture ingress. The Ontario-Hydro work [2] revealed that nearly 86% of all arrester failures could be associated with moisture ingress, while the Detroit Edison study [3] (primarily intended to determine the magnitude of lightning currents discharged through distribution arresters) showed that moisture was present in 10% of about 3000 arresters examined after about 12 years of service life. These and other studies suggested that some distribution arrester failures (about 5%) could be attributed to very severe lightning surges, eg. strokes of very high current magnitude or long continuing current durations, and/or multiple-stroke flashes with high multiplicity of stroke currents and power-follow currents. Indeed, recent laboratory studies hlave demonstrated that multiple lightning impulse currents can cause damage to distribution lightning arresters, which is not evident in standard single-impulse tests [ 5 ] . The North American studies also showed that some distribution arrester failures were caused by external contamination on the housing, and such failures were more likely for arresters without non-linear resistance grading of the internal gaps. Because of the advent of metal-oxide varistor technology, trial installations of gapless metal oxide arresters, began in the late 1970's and early 1980's, and within a few years most arrester installations were of the metal-oxide type. However, gapped silicon carbide distribution arresters still form the majority of distribution arresters in service, and represeni a significant capital investment. With a view to (1) determining the condition of current gapped silicon carbide distribution surge arresters in service on Australian distribution systems, (2) assessing the causes of their degradation and failure, and (3) assessing their average remanent useful life, almost four hundred such arresters were withdrawn after about 10 years' service in Australian distribution systems, and were subjected 0885-8977/96/$05.00 0 1996 IEEE 1790 to laboratory tests and internal inspection. This paper reports the results of the study, compares them with the findings from earlier work, and comments on the average useful length of service of such arresters. II. METHODOLOGY OF THE INVESTIGATION The Surge Arresters - Each of the participating electricity utilities was requested to recover from its distribution system about 50 surge arresters which had been in service for up to 10 years, and supply them for testing. Eight Australian electricity authorities submitted a total of 379 arresters, of which 365 were suitable for inclusion i n the project. Voltage ratings ranged from 9 kV to 24kV, and about 80% had a current rating of 5 kA, while the remainder were rated at 10 kA. There were seven manufacturer's names on the submitted arresters and Table 1 provides information about arrester numbers, years of manufacture and probable length of service. The overall median year of manufacture was 1983, suggesting that the median length of service of the tested arresters was about 9 years. Service Comments - Some of the participating utilities provided comments about the service performance of their arresters. One utility quoted an annual failure rate of over 1% of their distribution arresters and commented that the cost of locating and replacing failed arresters was about six times the purchase cost of the arresters. Other utilities, particularly those in the drier areas of Australian prone to high risk of bushfires, made specific mention of the fires caused by arrester failures. One utility reported that arrester failures seemed to be disproportionally high on overhead line underground cable junctions. Yet another utility experienced a considerable number of arresters which failed by external flashover during early morning periods of heavy dew presumably associated with external contamination. The flashover problem was associated with a particular make of arrester which had a relatively short-length porcelain housing and which was supported by a metal bracket fitted directly on a transformer tank. Laboratory Tests - The arresters were tested for insulation resistance, power frequency voltage withstand, power TABLE I INFORMATION ON 365 RECOVERED ARRESTERS ~~~ ~ ~ Manufacturer No. of Arresters Range. of Years* Median Year* Median Service** 1 2 3 4 5 6 7 4 20 200 1 11 22 107 1970-1973 1965-1975 1972-1991 1991 1957-1969 1979-1983 1978-1989 1970 1970 1986 1991 1964 1980 1981 22 22 6 1 28 12 11 * year(s) of manufacture median length of service (in years) assumes the amesters were installed in the year of manufacture and were recovered in 1992 ** frequency sparkover, and impulse voltage sparkover. In addition, tests were carried out to determine the condition of seals on about 40% of the arresters, and these and some others were dismantled to allow inspection of internal components. The first group of tests were designed to follow the general pattern of the earlier investigation [ 11, in which the intention was to identify significantly degraded arresters in a cost-effective way, rather than to test for strict compliance with the relevant Australian Standard [ 6 ] . Insulation Resistance Measurement - The insulation resistance of each arrester was measured with a 2500V insulation tester, using the guard ring technique to exclude the effect of leakage across the outside surface of the arrester. The highest reading provided by the tester was 2000 MQ.. For arresters without resistance-graded gaps, the insulation resistance was deemed satisfactory if >2000MCl, and unsatisfactory if < 2000MQ. Power Frequency (PF) Tests - The Standard requires that, after two preliminary sparkovers, 5 to 10 applications of power frequency voltage shall be made, and the voltage raised to sparkover each time. An interval of 10 seconds shall be allowed between applications, and the lowest sparkover level shall be not less than a stated value related to the voltage rating of the arrester. Arrester manufacturers publish power frequency sparkover levels, usually average values, at levels a few kV above the requirement of the Standard. In the earlier work [l],a 1-minute power frequency withstand test was substituted for the sparkover measurement, so that batch testing could be employed in dealing with large numbers of recovered arresters. The voltage level chosen for the withstand test was the minimum sparkover level specified in the Standard. For this investigation, the power frequency withstand test was again adopted, but in addition, a power frequency sparkover measurement was carried out, following the procedure specified in the Standard, and the average of the last three sparkover voltages was noted. Lightning Impulse Sparkover ( L B O ) Measurement - The Standard calls for groups of five impulses of each polarity to be applied, and states that 100% operation shall be achieved at a prospective amplitude not exceeding a stated value related to the voltage rating of the arrester. If an arrester fails to sparkover once during the 10 applications, a further 10 impulses shall be applied, and if 100% operation is achieved, the arrester is deemed satisfactory. For this investigation, impulses of negative polarity were applied in groups of 10 from a portable impulse generator producing voltage impulses of the standard 1.2I5Ops waveform, and prospective amplitudes up to 8OkV. The prospective amplitudes of the applied impulses commenced at a value approximately 5 % above the levels specified in the Standard. lf the arrester failed to spark over on one of the 10 impulses, a further 10 were applied at the same voltage level. If 100% operation was not then achieved, the voltage level was increased by 5kV, and this process repeated until 100% operation was achieved. 1791 Seal Test - The seal was tested by placing the arrester in a chambier which could be sealed and connected to a vacuum pump. Times from pump start were noted as the pressure reached 10, 5, 2, 1.5, and 1 Torr respectively. Free volume within the chamber was minimised by packing with steel pieces. For each class of arrester, pass/failure criteria for assessing the condition of the seals were selected from the time-pressure data for 91 arresters which had satisfactory insulation resistance and which obviously had sound seals. Altogether, 154 arresters, representing 42% of the arresters included in the project, were given seal tests. Inspecition - All of the arresters which underwent seal tests were subsequently opened and inspected internally. Some others, mainly those with very good or very poor results in the electrical tests, were also opened and inspected. All indications of departure from presumed original perfection within the arrester were recorded. However, in assessing the results of the inspection, minor arc marks on gap electrodes and block faces, signifying the passage of normal discharge currents, were regarded as acceptable. Any other recorded evidence of change was treated as deterioration, and summarised as moisture-related damage or as electrical damagle, according to the presence or absence of indications of moisture ingress or block damage such as surface flashover, puncture or rupture. No. Is10 IR Seal Inspection 2NT lP, 1F 5 5 ~ 49F 262 262P 262P 207P 16F 36 36F 36P 20P 66 III. PF 66P 66F 66P RESULTS AND DISCUSSION 1NI, 27F, 21P 4P 4F 160NT 159N1, 1F 40F 3N1, 21F, 16P 7P SF, 2P 7NT 3N1, 4F 8F 8F 1P 1F 7NT 3N1, 2F, 2P 13F 1NI, 11F, 1P 34NT 33N1, 1F 27F 2N1, 8F, 17P SP 2F, 3P 1NT 1P Fig. 1 shows the results of electrical tests on the 365 arresters investigated, and the results of seal tests and inspections where applicable. The results for each arrester are arranged in a horizontal strip across the diagram. Thus, the first block represents 262 arresters, all of which passed the power frequency and impulse tests, Fifty-five (21%) of these arresters had less than satisfactory insulation resistance (IR). Of this group, 2 were opened and inspected without undergoing the seal test, and 1 failed the inspection. The remaining 53 were seal-tested, and 49 (92%) failed the test. One of the 49 was not inspected, and of the remainder, 27 (56%) failed, while 21 (44%) passed. The 4 arresters which passed the seal test all failed the inspection. Of the 207 which had satisfactory IR, 160 were not seal-tested, and 159 of these were not inspected. The one which was inspected was found to have significant deterioration. Forty (85%) of the 47 which were seal-tested failed the test, a slightly lower percentage than among those with unsatisfactory IR (92%). Three of the 40 were not inspected, and of the 37 inspected, 21 (57%) failed and 16 (43%) passed. The percentage of arresters which failed the seal test and were subsequently found 1.0 have significant internal deterioration was almost the same among those with unsatisfactory IR (59%) as among those with satisfactory IR (60%), but moisture-related deterioration was found in 94% of the former, compared with 74% of the latter. The median length of service for the 262 arresters was 9 years, and there was a slight difference in this regard between those with unsatisfactory IR (10 years), and those with satisfactory IR (8 years). 1 1P 1F 1P - PF : power frequency tests LIS0 : lightning impulse sparkover IR : insulation resistance NT : not tested NI : not inspected P : passed F : failed FIG. 1 Test Results for 365 Arresters The second block in Fig.1 represents 36 arresters which failed one or both power frequency tests, and passed the impulse test. Sixteen (44%) had unsatisfactory IR, so the occurrence of unsatisfactory IR was much higher in this block than in the first block (21%). Of the 16 with unsatisfactory IR, 9 were seal-tested, and 8 (89%) failed the test. All of the 20 arresters with satisfactory IR were seal-tested, and 13 (6546) failed, again a lower percentage than aimong those with unsatisfactory IR. As in the first block, bhe percentage of arresters which failed the seal test and subsequently revealed significant deterioration on inspection was not greatly different for unsatisfactory IR (100%) and satisfactory IR (92%), though the actual percentages were much higher. Similarly, moisture-related deterioration occurred in a higher percentage (75%) of the arresters with unsatisfactory IR than 1792 among those with satisfactory IR (45%). However, it will be noted that these two percentages are much lower than the corresponding figures for the first block, indicating a higher percentage of arresters with damage due to excessive discharge current, without evidence of moisture ingress. The median length of service of the 36 arresters (20 years) is much greater than that for the 262 arresters which passed the power frequency tests, and again the arresters with unsatisfactory IR had a slightly greater median length of service (21 years) than those with satisfactory IR (18 years). The third block on Fig.1 represents 66 arresters which passed the power frequency tests but failed the impulse sparkover test. All of these arresters had satisfactory IR, and their median length of service was 7 years, appreciably less than the preceding groups. The failure rate on seal test was also lower at 84%, and the incidence of significant visible deterioration of internal components, in those arresters which failed the seal test and were inspected, was much lower, at 32%, and the proportion with moisture-related deterioration was 75%, as in the second block. The fourth block on Fig. 1 represents one arrester with 19 years of service, which failed power frequency and impulse tests, but this does not merit further discussion. The lack of association between failure on impulse sparkover test and failure on other electrical tests is quite striking, and was also encountered in the earlier investigation [ 11. The rate of occurrence of failure on impulse sparkover test is greater than i n the earlier investigation, but this is due to the much lower tolerance allowed. The earlier investigation was dealing with a population of arresters which dated back to the early years of silicon carbide arresters, in the 1930's, and it was considered necessary to permit a tolerance of some 25% above the maximum impulse sparkover voltages permitted by the Standards current in the 1960's. For the present investigation, a tolerance of 5 % above the levels set by the current Standard was adopted. In the mature phase of the earlier investigation, for several years arresters were recovered from service when they were suspected of failing to perform satisfactorily, or when the associated plant items were being changed to larger size. Of 1702 such arresters which were less than 15 years old, 84% passed all electrical tests, 5 % failed the insulation resistance test, 3% failed the power frequency voltage test, 5% failed the impulse voltage sparkover test, and 3% failed more than one test. In the current investigation, if the 25% tolerance on impulse sparkover is applied to 301 arresters less than 15 years old, the corresponding figures are: 77% passed all tests, 17% failed the insulation resistance test, 2% failed power frequency voltage test, 2% failed the impulse voltage sparkover test, and 1% failed more than one test, Comparison of the IR test results of the two investigations is of limited value, as many of the arresters in the current investigation hadresistance-graded gaps (manufacturers 6 and 7),and that feature was rarely present in the earlier work. With that qualification, comparison of the electrical test results of the two investigations suggests that the performance of distribution lightning arresters has changed little in the intervening 25 years. All the arresters in the third block had impulse sparkover levels which exceeded the level allowed by the Standard by more than 15%. In earlier investigations, eg. [l], it was noted that high impulse sparkover was sometimes accompanied by high internal pressure, and that arresters subjected to multipulse operating duty tests sometimes exhibited raised impulse sparkover voltage afterwards [SI. Evidence of raised internal pressure was observed in some instances i n the recent investigation. The median year of manufacture of the arresters with raised impulse sparkover voltage was 1985, and the majority were manufactured from 1980 to 1988, mainly by one manufacturer, as discussed later in relation to Table 3. However, the earlier investigation showed that raised impulse sparkover level was not a significant factor in relation to failure rates of distribution transformers in service. The apparent significance of length of service which emerges from Fig. 1 is dealt with more directly in Table 2 and Fig. 2. There is clearly an upturn in the incidence of unsatisfactory IR after about 8 to 10 years of service, and this is followed by a more marked upturn in failure on power frequency tests at about 13 to 15 years of service. The incidence of visible deterioration of internal components, as revealed by inspection, begins to appear significant after only a few years of service, and rises steadily thereafter. Because of the small numbers involved, the further subdivision of fault rates into "moisture-related'' and "purely electrical" fault rates is somewhat less reliable, but is indicated in different ways at the bottom of Table 2 and in Fig. 2. The failure rates of the tested arresters are subdivided in Table 3 by manufaturer. As is evident, most of the arresters were from manufacturers 2 (ZO), 3 (ZOO), 6 (22) and 7 (107). There are clear differences in the results for these four manufacturers' arresters. Some of the differences are attributable to different ages of the arresters, eg. in respect of TABLE 2 FAILURE RATES / LENGTH OF SERVICE 0 to 6 Years (1992-1986) 7 to 12 Years (1985-1980) More than 13 Years (1979-1957) Total 119 157 70 Low IR Failied PF test(s) Failed LISO. No. seal-tested Failed seal test No. inspected Failed iiispection Moisture-related Electrical 7 (6%) 4 (3%) 24 (20%) 39 34 (87%) 39 10 (26%) 9 (23%) 1 (3%) 40 (25%) 5 (3%) 35 (22%) 74 64 (86%) 73 37 (51%) 33 (45%) 4 (6%) 21 (30%) 23 (33%) 4 (6%) 36 34 (94%) 43 38 (88%) 32 (74%) 6 (14%) 1793 TABLE 3 FAILURE RATES / MANUFACTURER Manufacturer 1 2 3 4 5 Total 4 20 200 1 11 22 107 Median length of service (yrs) Low IR 22 3 Failed PI? test@) 3 11 44 (41%) 5 0 No. seal-tested Failed seal test 3 2 6 8 (4 %) 15 (8%) 62 (31%) 64 1 28 12 1 1 7 (32%) 0 5 0 Failed LIS0 22 7 (35%) 9 (45%) 0 No. inspected Failed inspection 3 3 Moisture-related 2 Electrical 1 12 12 (100%) 13 13 (100%) 10 (77%) 3 (23%) 59 6 0 (1 5 10 5 10 (100%) 7 10 7 1 (10%) 5 1 (100%) 2 0 0 3 1 0 (92%) 1 64 27 0 (42%) 0 17 (63%) 0 10 (37%) Comparisons were made of test results for arresters from electricity authorities in different climates, but with somewhiit similar median lengths of service. These showed that, although there appeared to be a small effect on seal failure rates for service periods of about 7 years, for longer periods of service the effects due to length of service were much greater than any effect due to climate. This suggests that replacement of arresters on the basis of age would be a cost-effective strategy, and that the age criterion need not be different for different areas. 7 (5%) 1 It is of interest to attempt comparisons between the above data for Australian arresters and those for North American arresters reported in [21 and [3, 41 from studies conducted about fifteen years ago. The Ontario-Hydro results [2] were for failed arresters, the Detroit Edison resulls [3] were for inservice arresters withdrawn and inspected (primarily to determine the magnitude of lightning discharge currents) and for failed arresters inspected to identify the cause of failure [4]. In all of the studies, moisture ingress was the dominant cause of arrester degradation or failure. The current Australian investigation is i n accord with this - 74 (about 48% of the 155) arresters opened for inspection (Table 2) showed evidence of moisture-related degradation (internal moisture and/or clear moisture effects such as verdigris or severe corrosion was evident in about one-third of these, ie. about 16% of the 155). It would be fair to suggest that these percentages are biassed upwards, because the inspected arresters were mainly opened because they had already exhibited unsatisfactory electrical and/or seal test results. [f the moisture-related degradation and the moisture-evident numbers are expressed as a percentage of the whole sample of arresters included in the project (365), then the percentages of 20% and 7% agree well with the two Detroit-Edison studies [4, 31. Block damage such as surface flashover and block puncture or rupture was also evident in small but 58 48 (83%) 62 45 (73%) 40 (89%) 5 (11%) insulation resistance, arresters of make 3 had a low IR failure rate (4%) compared to 32 to 41% for makes 2, 6 and 7, but then the median age of make 3 arresters was 6 years compared to 11 to 22 for the others. The same type of comment can be made (but less clearly so) in respect of the results of power frequency tests. Reference has already been made to the fact that most of the arresters with high lightning impulse sparkover were of one make (3, with a failure rate of 31%). The seal failure rates of all arresters subjected to seal tests were uniformly high among nearly all the makes. The inspection results of the four dominant makes were mostly similar - high percentages (63 to 100%)showing evidence of moisture ingress and significant percentages (1 1 to 3 1%) exhibiting electrical damage. A - 119 arresters, 0-6 years; - 157 arresters; 7-12 years; C - 70 arresters, more than 13 years B 100% I 75% 50% 25% 0Yo 50 Hz Test@) IS0 Test E l A IR Test m Seal Test B FIG. 2 Test Failure Rate vs Years Service Inspection m C 1794 significant percentages of North American and Australian arresters. The current investigation's percentage is 5.7% of the 365 arresters (13.5% of the inspected arresters) - figures which are comparable to the 5.9% for the Ontario-Hydro failed arresters [2] but low compared to the 40% for the Detroit-Edison failed arresters [4]. It is of interest to note that both types of block damage (surface flashover and puncture or rupture) were also caused by multipulse operating duty cycle tests conducted on similar arresters in the laboratory [ 5 ] . Finally, both the Ontario-Hydro and the Detroit-Edison studies reported some occurrences of housing flashovers attributable to external contamination. In the current Australian investigation, 9 of the 365 arresters showed external flashover marks - 8 were associated with one particular make of arrester with a relatively small length of porcelain between the line terminal and the (earthed) mounting bracket . VI. REFERENCES [ 1J M. Darveniza and D.R. Mercer, "Service Performance of Distribution Lightning Arresters and Transformers", Elect. Eng. Trans. Inst. Eng. Aust., vol. EE2, Sept. 1966, pp 97-112. [2] M.V. Lat and J. Kortschinski, "Distribution Arrester Research", IEEE Trans. Power Apparatus and Systems, vol. PAS-100, no. 7, July 1981, pp 3496-3505. [3] G.L. Gaibrois, "Lightning Current Magnitude Through Distribution Arresters", ibid, vol. PAS-100, no. 3, March 1981, pp 964-970. [4] G.L. Gaibrois, "Small-Block Distribution Arresters Win In Field Performance", Transmission and Distribution, vol. 32, no. 9, Sept. 1980, pp 36-40. IV. CONCLUSIONS (I) In a significant proportion of the modern silicon carbide distribution arresters examined in this investigation, deterioration of internal components begins after a few years in service, and the proportion of arresters so affected increases steadily with increasing length of service. (2) After about IO years of service, there is a marked upturn in the number of arresters with unsatisfactory IR. (3) After about 13 years of service, there is also a very marked upturn in the number of arresters with reduced power frequency sparkover level. (4) Evidence of moisture ingress has been found, but much less frequently than in the earlier work reported in [I], suggesting that arrester seals have improved, but not sufficiently to completely overcome the problem of moisture ingress. ( 5 ) In about one-eighth of significantly deteriorated arresters, inspection of internal components reveals damage by high currents andlor multipulse currents, without indications of moisture ingress. (6) No single test gives a comprehensive evaluation of the condition of a recovered gapped silicon carbide arrester. One of the conclusions reported in the recent work on the effect of multipulse lightning strokes on arresters was that conventional tests were not particularly sensitive to some forms of deterioration within arresters. (7) Most gapped silicon carbide distribution lightning arresters of age in excess of 13 to 15 years have characteristics which render them no longer suitable for service on a distribution system. Recommendation - It is recommended that all silicon carbide distribution arresters older than 13 years be progressively replaced by modern metal oxide arresters. V. ACKNOWLEDGEMENT The authors grateful acknowledge the contributions of the 8 Australian electricity utilities who participated in this project. [ 5 ] M. Darveniza and D.R. Mercer, "Laboratory Studies of the Effects of Multipulse Lightning Currents on Distribution Surge Arresters", IEEE Trans. Power Delivery, vol. 8, no. 3, July 1993, pp 1035-1044. [6] A.S. 1307-1986, "Surge Arresters, Part 1 - Silicon Carbide Type for A.C. Systems", Standards Australia, North Sydney, Australia 2060. Mat Darveniza (F1979) and Ron Watson work in the High Voltage Laboratory in the Department of Electrical and Computer Engineering at the University of Queensland. Doug Mercer is a Honorary Research Consultant of the University and was formerly with the Queensland Electricity Supply Industry. 1'795 Discussion further study of the tabulated results discloses the following: From 160 arresters that were inspected following the leak test, 60% failed and 40% passed the visual examination. But of the 17 thlat The data presented in the paper very well supports the passed the leak test, 70.6% failed the visual examination. And of 130 that failed the leak test and were visually examined, 42% were conclusions that allowing for different tolerances specified at judged as passing the visual examination. Clearly the effectiveness various times, the basic electrical characteristics of gapped of the leak test method used should be reconsidered. silicon carbide distribution class arresters have changed little from those manufactured in the 1930s up to early 1990s. Arresters are usually factory sealed at atmospheric pressure with Insulation Resistance : This test appears to be the main in- controlled gasket compression As the arresters age in service, the service test done by most Utilities. Looking at the results of elastomeric gaskets may assume some permanent deformation, the IR (Megger) tests done on the healthiest batch of relatively unable to assume their original thickness should the mechanical young arresters (median age 9 years), it can be noted that it loading be relaxed Subjecting the arresters to a vacuum causes the end sealing hardware to be loaded in an unintended manner gave wrong results 40% of the time when it indicated a failure, causing loss of some of the compressive force Perhaps some of and inlerestingly wrong again 55% of the time when it these arresters became "leakers" during the lea6 test, such as the 2 1 indicated a pass! This is about a precise as flipping a coin! of line 2 and 16 of line 5 of the first block , and the 17 of line 2 With older arresters, it appears 100% correct when it indicates of the third block A sparkover test before and after the vacuum a failure, but unfortunately 35-80%0 wrong when it indicates a test may indicate if the arrester gasket opened and then resealled pass! It appears that when testing older arresters IR can pick after the seal test Did the authors consider such procedure to validate the vacuum leak test method used? up the stronger fail signal, but it not a good pass predictor. Lightniing Impulse Sparkover : This test has predicted failure Measurement of intemal ioni~ationor radio mflueiice voltage (RI'V) correctly only 35% of time. It also appears to predict pass is used by many as a repeatable and inexpensive test to detect gap results correctly only 10-40% of the time. This test is degradation Did the authors consider this method as an alternate definitely the least reliable predictor. to impulse sparkover? Power Freqnency Sparkover : Manufacturers consider this the better predictive test. When this test indicated a failure it The described power frequency test includes a one minute voltage was about 90% valid. However if a pass was predicted, it was withstand test near the minnnum sparkover voltage lebel Tlirs correct about 40-65% of the time. Hence only the "pass" voltage is several times the normally expecred system Iine-toresults from a PF test must be treated with some caution, ground voltage, which can overstress any non-linear resistors used mainly because this test is not able to pick up damage to S i c for gap grading Did the authors notice any evidence of gap grading resistor damage caused by this withstand test7 blocks <andis not always sensitive to moisture ingress. Sealing Reliability : In 1987 AS 1307.2, written for the new Answers to the above questions may serve to reinforce the correct gapless Metal Oxide types, established a more rigorous sealing conclusions reached by the authors type test that was in effect applied to gapped silicon carbide types still being made at the time. Unfortunately, the data shows 89% of the 100 relatively young distribution class Manuscript received February 15, 1996. arresters tested were unable to demonstrate gas tightness. We may conclude that compliance with the 1987 sealing type test has no1 in itself been sufficient to guarantee any worthwhile degree of seal reliability. However, it must be said that not all porcelain insulated designs on the market, particularly the Lambert Pierrat, Senior Member, IEEE (General Technical station class types, have unreliable seals. Division, Electricit6 de France, 37 rue Diderot, 38040 Grenolble Modern gapless metal oxide polymeric distribution arresters tend to have designs which eliminate intemal air volume. Cedex, France). Fabrice Perrot, Non-member (EA Technology They are expected to have much better long term sealing Ltd, Technical Business & Services, Capenhurst, Chester CH1 performance and far more stable electrical characteristics. 6ES, U.K.). The data collected and analysed by the authors present a valuabie piece of information. Effectively, the field experience for gapped Sic surge arresters can also be used for other network components, particularly for gapless ZnO surge arresters. This discussion concerns firstly the arresters failure modes which are partly connected to their engineering technology John B Posey (Westfield Center, Of3) As the authors have stated, and secondly the estimation of the residual failure risk of multiple tests should improve the diagnosis However, comparison components in service. of the iesults of leak testing and visual examination shows little correlation suggesting a reexamination of the leak test procedure 1) - Arrester failure modes : The authors determined that the may be appropriate predominent cause of the arresters failure wits due to inadequate seal performance leading to moisture ingress and therefore Since the text describing results of testing and examining the third deterioration of the arrester active components. Damage thought block of 66 arresters does not agree with the tabulated values. David lioby (ABB Power Transmission, Australia 2170) : 1796 to be the result of high current and/or multiple currents was also reported. a) - Could the authors comment on whether ferro-resonance overvoltages could have been involved in some cases of arrester failure due to excessive current absorption; especially when one of the utilities mentionned disproportionally high arrester failure rate at overhead line / underground cable junctions? b) - The authors have conducted 4 types of electrical test, a seal test and a destructive examination. The seal test was conducted under vacuum. The authors postulate that arresters with satisfactory insulation resistance must necessarily have sound seals. Surely this is not quite the case, as the results of the investigation showed that a high percentage of the arresters which failed the seal test did not necessarily failed the insulation resistance test. Should a test involving the immersion of the arrester in a liquid for a temperature cycle be more appropriate and closer to field conditions in order to determine the mechanical performance of the seal under mechanical stress? c) - The authors report the striking lack of association between failure on the impulse sparkover test and failure on other electrical tests. We are qestionning the relevance of the impulse sparkover test due to the known erratic behaviour of spark gap systems under impulse conditions after a few years in service. Can the authors comment on this issue? d) - We believe that most of the arresters tested in this study were of the porcelain housing type. Did the authors test any polymeric housed gapped Sic arresters and if yes could they comment on them? 2) - Estimation of the residual failure risk : The authors recommend that all gapped Sic distribution arresters older than 13 years be progressively replaced by modern metal oxide arresters. It is possible to analyse the failure risk of the whole of the arresters by considering firstly the age distribution of the arresters and secondly their failure distribution as a function of age. The age distribution is approximately exponential over an average value of the order of 2 p 7 years (see data on Table 1). The estimation of the rate of failure in service is based upon the data shown in tables 2 & 3 and figure 2. The results of the arrester examination show that the increase of the failure rate is linear; which correspond approximately to a Weibull distribution. (shape parameter P=2 and characteristic life duration : 8 p 7 years). Consequently, the Mean Time To Failure (MTTF) of the arresters would be in the order of &/2.z0.e0=43 years. This value could appear to be high, but one must consider its important dispersion associated (coefficient of variation cv > 1). The residual risk of failure of the arresters is approximately given by a distribution of Weibull (shape parameter P < 1; characteristic life duration = 43years). For 13 years old components (value recommended by the authors) the failure risk would be of the order of 30%. If the dubious components were not taken out of service, the rate of failure of the arresters would decrease with time. However, this would correspond to a progressive extinction of the number of arresters after all their failure: obviously, this catastrophic hypothesis is not realistic and we therefore agree with the recommendations formulated by the authors. The removal of 15 years old arresters from service should stabilise the arresters failure rate to acceptable levels. We thank the authors for their contribution and their answers to our questions and comments. Manuscript received February 26, 1996. .Darveniza, D.R. Mercer: The authors are grateful for the widespread interest shown in our paper by Utilities in many countries and for the contributions from the discussers. David Roby’s points highlight the fact that an assessment of the condition of anything (be it a person by a doctor, or an item of electrical equipment by a test engineer) requires the use of multiple diagnostic tools (tests). It is always unreliable to make an assessment from the results of any one diagnostic test. It is always best to use the results from as many tests a s i s practicable. John Posey (and Lambert Pierrat and Fabrice Perrot) questioned the value of the impulse sparkover test. We have sympathy for this, but included it in our set of diagnostic tools precisely because the impulse sparkover voltage is a direct measure of a gapped arrester’s capacity to limit lightning overvoltage and so to protect equipment. In our earlier studies, we did use partial discharge (PD) tests (which are equivalent to RIV tests) and found them to be sensitive to the presence of both internal moisture and of damaged internal components. We did not persist with PD tests because we believed that our other diagnostic tools (particularly IR and PF tests) were also adequate in identifying arresters with those deficiencies. All the discussers commented on the deleterious influence of intemal moisture on arresters and noted the importance of an adequate seal test as a diagnostic tool. We agree that subjecting a distribution arrester to a full (external) vacuum is a harsh test of the seal. But we do not believe it to be unrealistically harsh, as it has been our experience that most new arresters pass such a test (of course we did encounter a (very) small number of new arresters with faulty seals and believe that these either escaped notice during factory tests or were physically damaged later). There are of course other ways for testing the seals and we make no claim that our method is the “best” - the matter of importance is that an adequate seal test must be included in the set of diagnostic tools. Some other specific questions were also asked by the discussers; to Mr Povey - we did not observe damage to grading resistors attributable to the PF tests; to Messrs Pierrat and Perrot - the only gapped S i c arresters available to us had 1'797 porcelain housings, and yes we think ferro-resonance could be the cause of arrester failures at overhead linehnderground cable junctions (though we have not made a specific study of this). Finally., we thank Messrs Pierrat and Perrot for adopting a more systematic statistical approach to the problem of arrester reliability. It has been our experience that Weibull statistics provide a formidable tool for analysing the failure risk of system components as they age and approach the end of their useful life. Colleagues at our University have successful used Weibull statistics in a number of such applications, but we did not do so for distribution arresters. We are grateful that Messrs Pierrat and Perrot have enhanced the value of our results by illustrating the applicability of the Weibull distribution to the analysis of the failure risk of aged arresters. Finally, we would comment that the next phase of our work will address the reliability of polymeric - lhoused metal oxide arresters. This work has begun and is being carried out in two strands (i) identification of failure modes by laboratory tests, and (ii) examination of arresters which have failed or have become degraded in service. The latter requires (and has) active collaboration by electricity utilities in Australia. We would also like to hear from utilities in overseas countries with failure experiences of polymer-housed metal oxide distribution arresters. Manuscript received April 1, 1996.