Save Article Instructions
Close 

A Retrospective on Generator Failures

By Colin M. McDonald

The 320-mw Roxburgh project experienced significant problems related to generator failures since its commissioning in 1956. The author reviews the most severe problems with three of the plant’s units and offers insights on why they occurred.

Although many generator failures seem unique, almost every problem an engineer encounters has occurred somewhere else. Contact Energy’s 320-mw Roxburgh facility on the Clutha River in New Zealand has experienced problems with all eight turbine-generating units over its 50-year lifespan. The most severe issues or those that taught the most valuable lessons occurred in Units G2, G1, and G8.

Unit G2: resin injection, loss of fault data

In 1965, nine years after Unit G2 was commissioned, the stator suffered a turn-to-turn fault. Roxburgh personnel determined the bond between the turn and groundwall insulation systems had separated. Once this void was established, partial discharge (PD) erosion and mechanical abrasion caused the turn insulation to fail. Roxburgh personnel replaced the failed coil.

The unit operated reliably until the mid-1980s, when it experienced a series of stator faults that required coil replacements. Units G5, G7, and G8 experienced similar faults. These faults, along with information from PD tests, indicated the windings on Units G2, G5, G6, G7, and G8 had reached the end of their lives. However, instead of rewinding the units, Roxburgh personnel chose an experimental resin injection process to extend the life of the units.

Resin injection was performed on these five units in 1988 and 1989. This process involved pumping resin in around the copper strands, at the bottom end-winding coil evolute, until it came out the top evolute.

The goal of the resin injection was to extend unit life by ten to 15 years. Overall, this goal was achieved. G2 failed after ten years, and G8 failed after 12 years. Both of these failures were major events, and Roxburgh personnel chose to perform a complete rewind on these units. Both Units G5 and G6 operated successfully for 15 years, at which time the windings were replaced as part of a major refurbishment project. G7 is still in service after 18 years. However, due to water availability and system constraints, the unit only runs 13 to 28 percent of the time. This unit is scheduled for complete refurbishment in 2007-2008.

When Unit G2 failed in February 1998, the unit was tripped by both the vibration and CO2 discharge monitors. Data from these protection systems showed the main circuit breaker opened about 0.5 second after the fault was detected. At 1.965 seconds, phase currents of 5,500 amps were recorded. At 5.975 seconds, the current was still about 1,720 amps. This time sequence revealed a problem with an excitation control upgrade performed two years earlier.

During this upgrade, personnel removed the main field circuit breakers because they were no longer maintainable. This removal created a closed loop rotor exciter circuit. Without the circuit breaker to interrupt the loop during the stator fault, the current oscillates back and forth with the collapsing field until it is dissipated in the fault. This is what caused the damage to the unit.

The protection relays installed as part of the control upgrade are designed to store fault data in their volatile memory. However, during investigation of the failure of Unit G2 in 1998, Roxburgh personnel pulled the direct current (DC) fuses as part of standard unit isolation procedures. This resulted in complete loss of the valuable fault data. Personnel determined the crew that installed the new relays in 1996 had not clearly communicated all features of the new control system. As a result of this experience, Contact Energy has modified both the isolation procedures and the DC circuits for the Roxburgh units so that unit isolation requirements can be achieved while maintaining a DC supply to the relays.

An investigation after Unit G2 failed in 1998 revealed the initial fault was a turn-to-turn failure in the front leg of a coil in a yellow phase circuit. The fault vaporized virtually the entire front top end-winding and 400 millimeters of the slot portion of coil 220. The failure caused considerable damage to the back top end-winding legs of coils 211 through 219. There were also two secondary phase-to-phase faults, between the top evolutes of coils 220 and 227 and the front pole tails of coils 221 and 228. In both cases, the evolute and tail were burned through.

The top end winding, PDA couplers, and collector rings in the fault region had serious fire damage, involving 75 coils. Also, short (120-millimeter) pieces of copper strand were found up to three-fourths of the way around on top of the stator. The only evidence of damage on the bottom end-winding was a gas blowout on the front end-turn of coil 220. There was no visual evidence of other winding distortion. The stator core had distorted by up to 1.5 millimeters either way from the nominal bore.

When Roxburgh personnel removed the winding on Unit G2, coil dissection confirmed the failure mechanism was similar to that of the failures occurring prior to the resin injection. However, it was clear the resin injection process might have been less than perfect. The lower three-fourths of each coil showed good resin impregnation, but not all coils achieved complete impregnation of the top 300 millimeters of the slot portion and top end turn.

As would be expected, PD activity for Unit G2 always coincided with the area without resin. This activity was breaking down the turn insulation, creating a high-resistance connection through which large circulating currents flowed, in the closed loop of the shorted turns. This resulted in an arc and extreme temperatures that vaporized the copper in the region of the short. This tracked up inside the groundwall insulation to the weaker end-turn insulation, where it blew out. In this case, the initial turn short appears to have been about four lamination packets down (about 200 millimeters) from the top of the core. This is also the region where any stress due to end-winding flexing is likely to have been present.

Roxburgh personnel found evidence of recent PD activity in seven coils. In each case, the bottom end showed good resin impregnation. As expected, much of the turn mica had been pulverized in coil 220, where the fault occurred. The heat generated by the fault was so intense that it is difficult to get an exact picture of how the fault had developed. However, strands from the two turns were welded together by the arc.

When did the voids develop in the insulation of the Unit G2 stator? Roxburgh personnel had compared PD results for Unit G2 to Unit G3 and other machines in October 1997. This comparison did not indicate the presence of significant voids in the G2 winding. Therefore, the voids likely developed between October 1997 and February 1998.


The original windings at the 320-mw Roxburgh facility in New Zealand consisted of a cotton tape layer, which was where these windings started to fail. The binding resin/bitumen is gone from the cotton tape and the turn insulation. The main groundwall insulation is still intact.
Click here to enlarge image

After determining the cause of the failure, Contact Energy decided to rewind G2. To begin the rewind, Roxburgh personnel corrected core distortion problems, with the unit in place, before removing the old winding. Slot preparation involved scraping virtually all the 420 slots, to remove resin residue from spills during the injection process, winding debris, and protruding laminations. Roxburgh personnel performed this tedious task, which took five to six weeks.

Once the stator core was prepared, Roxburgh personnel and personnel from TGE Energy Services, the company hired to perform the rewind, subjected the core to a full flux test to check for hot spots. The hottest spot was 6 degrees Celsius (C) above the average core temperature of 61 C. This was considered acceptable.

The new winding was fitted in the conventional manner.

The unit has operated satisfactorily since the rewind. Contact Energy plans to replace the coil and winding in Unit G2 as part of an ongoing refurbishment of the Roxburgh station.

Unit G1: coil reversals, rewind materials

In the early 1970s, this unit – as well as Units G3, G4, and G7 – experienced multiple stator faults. At the time, this unit was more than 15 years old. In line with practice at that time, Roxburgh personnel dealt with this problem by establishing a program of phase/neutral reversals. The goal was to put the coils that were in the best condition in the region of highest electrical stress. Between October 1972 and November 1973, personnel performed coil reversals on Roxburgh’s four oldest units (G1 through G4). However, this work placed the most deteriorated coils where they were not monitored by the unit protection systems. As a result, by July 1974 both Units G3 and G4 had experienced stator faults. Repairing the faults involved replacing a considerable number of coils.

After these two units experienced the faults, Roxburgh personnel decided to re-reverse the windings in Units G3 and G1. G2 was not re-reversed because PD activity was not significant. In two newer units – G5 and G6 – Roxburgh personnel performed coil reversals to avoid faults. This work was completed by December 1974. Each unit with the reversed old windings (G2, G5, and G6) received 100 percent earth fault protection equipment, which monitors the entire winding.


This stator core shows significant loss of lamination insulation (left arrow), which led to tooth fretting and core looseness. This photo also shows Nomex packing (right arrow) inserted between the segment faces.
Click here to enlarge image

In 1975 and 1976, Roxburgh personnel rewound the stators of Units G1, G3, and G4 using a semi-rigid synthetic resin winding with the same two-turn configuration as the original, but with a 20 percent increased copper section. The same tape was used for both the turn and groundwall insulation. The tape had large mica splittings as its base, not reconstituted mica. While more expensive to manufacture, the large mica material is very resistant to PD damage. None of these units experienced another stator fault.

During 2006, Unit G1 was completely refurbished as part of an ongoing station upgrade. (For more information on the Roxburgh refurbishment program, see the sidebar on page 18.) When being unstacked in July 2006, the stator exhibited the typical red dust signs of fretting, particularly at the segment joints and at key bars. In addition, the tooth tip lamination insulation loss was typical of the Roxburgh units. The extent of erosion penetration was about 10 millimeters but was as much as 70 millimeters in the worst cases. Given that the machine is 50 years old, the extent of loss is not too bad.

An interesting discovery in the G1 stator during the refurbishment was evidence of slot discharge PD activity having eroded away the core lamination insulation. However, the effect of the PD was much more significant on the slot side packing and coil. Roxburgh personnel determined that, during the mid-1970s rewind, the winding was not packed tightly enough. The lap coils (the last to be fitted) are difficult to side pack effectively, and they had not been positioned to coincide with the neutral end of the winding.

Unit G8: stator core separation, repair

As mentioned earlier, this unit experienced a series of stator faults in the mid-1980s that required coil replacements. Unit G8 also was one of the five Roxburgh units on which resin injection was performed in 1988 and 1989 to extend the life of the units.

During a biennial visual inspection of Unit G8 in June 1996, Roxburgh personnel observed a separation of the stator core from the frame. The bottom half of the core had migrated 5 millimeters into the 11-millimeter air gap.

To repair this problem, Roxburgh personnel used a specially constructed jig and hydraulic jack to push the core into shape. The unit ran smoothly during the bearing run. However, upon application of excitation, it immediately began to resonate. A local consultant using a handheld probe measured vibrations of the floor plates and other accessible areas at 100 Hertz (Hz) and 200 Hz. However, the Bently-Nevada proximity probes installed on the shaft did not detect vibrations. Air gap measurements at 12 points throughout the unit showed the stator had distorted by up to 1.5 millimeters.

Roxburgh personnel consulted two generator experts, who believed the problem was related to a gap developing between the stator segment joints. The gap was preventing the forces in one segment from being transferred to the adjacent segment. This would leave a section of core relatively unsupported and susceptible to air gap problems. Neither expert could provide a guarantee for a successful repair.

Roxburgh personnel used feedback from the two experts and techniques developed on site to craft a repair strategy. The procedure performed involved cutting off all angle iron/steel gusset key bar anchor brackets, one segment at a time, and pushing and pulling the stator core segment back into shape. Work was carried out by two local fitter/welders and one trades assistant. Care was taken to avoid damaging/cracking the stator winding. The original angle iron bracket design was discarded in favor of 12-millimeter-thick horseshoe-shaped anchor plates. The six stator core segment joints were packed with Nomex material coated with epoxy resin, so the core would act as an annulus. The materials were supplied by Electropar, a New Zealand firm. The stator bore and back iron were coated with a “weeping” epoxy resin, to lock up the laminations.

As a result of the problems with Unit G8, Roxburgh personnel inspected Units G5, G6, and G7. Similar core migration defects were discovered in all three units. Total outage time per repair was about six weeks. Cost of each repair was about NZ$70,000 (US$49,400).

In June 2000, four years after the core repair, Unit G8 experienced a major stator failure. As a result of the lessons learned with G2, Roxburgh personnel were able to download fault data from the integrated protection system. The data showed the main circuit breaker opened 134 milliseconds (seven cycles) after the fault, with red and blue phases involved, and that the fault occurred just as the red phase current reached a cycle peak. Maximum peak-to-peak current flowing during the fault was 15,000 amps (normal full load current is 2,300 amps), with the stator current continuing for some seconds after the fault, due to the closed loop rotor-main exciter circuit.

The initial event was a phase-to-phase fault in the bottom end turn region of the winding. The fault caused significant damage, with copper loss in seven coils, shorted turns in 11 of the 44 rotor poles, and stator core distortion consisting of four segments pulled in and two pushed out.

The fault location was centered 210 millimeters below the bottom of the core stack, on the sloping section of the coil end turn and just above the surge ring/surge bracing. Seven coils sustained copper loss damage, involving all three phases. The main arc damage was on the bottom end turns of the coils in slots 89 to 93, with the fire damage on the bottom end winding region from slot 25 through slot 109. There was no damage evident at the top of the stator, nor was there evidence of any winding distortion.

Globules of molten copper were blown out from the fault area. Despite the copper loss and fire, there was a substantial amount of resin surrounding the turn insulation. The resin had to be chipped off to survey the copper loss, so PD activity was considered unlikely to have caused the failure. While coils 89, 90, 91, and 92 straddle the 4 o’clock stator joint, there was no evidence of significant radial distortion across the joint that could have stressed the insulation.

The Unit G8 failure mechanism was most probably due to a crack in the end-winding insulation. Roxburgh personnel determined that the insulation damage could have resulted from several causes. The first was suspected impact by a 3-inch-long, 1-inch-thick piece of steel bar of unknown origin. This bar was found in the bottom endwinding cover and looked as it if had arc damage. The second cause could have been a stress crack introduced as the result of core movement during the earlier stator core repair or system disturbances (disruptions in the system resulting in voltage and frequency out of their normal range). Finally, a combination of insulation damage, surface contamination, and a voltage spike arising from a system disturbance would be sufficient to initiate the fault.

During disassembly of the Unit G8 stator, most of the strands peeled out clean, leaving much of the groundwall insulation glued in the slot. Roxburgh personnel made several observations. Only one coil had significant corona activity. All the PD was in the top of the coil zone, which equates to the top of the core. The stator windings in the unit had a reasonably large number of examples where old signs of corona activity had been encapsulated in resin. Also, the insulation had a very dry appearance. This means resin had filled the voids during the earlier injection process but had not replaced the original binder between the mica flakes.

Based on visual observations, it appeared the 1996 core repair withstood the fault well. There were signs of fretting and vertical movement of the stator core laminations on the key bars and minor packing migration in the bottom of a joint. Core distortion was confirmed by measurement, with segments 3, 4, 5, and 6 having been pulled in toward the rotor and segments 1 and 2 having been pushed out. The distortion was up to 1.5-millimeter variation from “as left” after the repair and exceeded the 10 percent air gap variation allowance. However, the rotor’s center of rotation was still within 0.3 millimeter of the 1996 center.

Roxburgh personnel were not certain how the unit would react to a weld repair. They were afraid contractions caused by the heat might result in failure of the dovetails tabs on the laminations, leading to a complete core failure. As a result, Roxburgh personnel opted for a full core replacement.

The decision was made to refurbish all the poles, using resin-rich Nomex paper as the turn insulation.

Discoveries made during removal of the winding validated the decision to replace the core. Resin leaks had glued the coils into the core slots, and a snatch block and the overhead crane were needed to extract each coil. With the amount of material left in the core slots, it would have taken months to scrape each slot clean. By contrast, the core was dismantled within a couple of weeks.

The most critical task of a stator core assembly is the core and radial positioning and alignment of the keybars. The keybar placement dictates final shape of the stator bore and the ease of stacking. Each keybar must be accurately set vertically (both radially and tangentially) and with the correct spacing between the bar (angular offset). Allowance has to be made for the movement that occurs during weld contraction. TGE Energy Services performed the Unit G8 work in situ, using the existing stator frame without the thrust bracket removed. The keybars were required to be aligned about the center of rotation and to the unit’s natural lay, as distinct from the true vertical. The center of rotation was established and surveyed back to reference points prior to the machine disassembly.

The new cores were assembled as a continuous ring. During the stacking process, TGE Energy Services compressed the core three times and subjected the core to a full flux ring-flux test before beginning winding installation, to shake down the core and check for core defects (hot spots). The calculated core loss results for the new cores were about 1.6 watts per kilogram (57 kw) at 1.1 Tesla at 60 Hz. No hot spots were identified.

Refurbishment of the rotor poles for Unit G8 was too big a job to attempt on site. Each complete rotor pole assembly was shipped to TGE Energy Services in Perth, Australia. TGE Energy Services removed all but one pole from the rotor rim within four days. To remove the 44th pole, the company had to drill two holes through the unloading bay floor, construct a special jig, and jack the pole out of the rotor rim using a 50-ton force. The problem was that the two pole wedges that had been worked on were both ends of the same wedge. During an earlier attempt to remove the pole, it appears one pole wedge had broken off in the slot.The maintenance crew then abandoned the removal attempt and drove the broken piece back into the slot. What actually happened was that the full wedge remaining in the slot had sprung over, and the broken wedge was inserted, on the opposite side, to its original position.

While the pole assemblies were at the contractor’s, Roxburgh personnel and TGE Energy Services carried out experimental tests to see if there were shorts that would only appear when the unit was at speed. This was done by progressively increasing the pressure on the winding and taking volt drop readings at each step. No turn shorts were found in the sample of windings tested, which indicated there were no faults that would occur when the unit was in service.

The unit has operated satisfactorily.

Reasons for the generator problems at Roxburgh

Roxburgh was designed as a base load facility. Because of a power shortage in New Zealand, the eight units were assembled and put directly into service without much factory or commissioning testing. The first two units were commissioned on consecutive days in July 1956! The next two units were commissioned in just two days each in August and December 1956. Other units were commissioned in the intervening years, with the final unit commissioned in July 1962.

Obviously, commissioning was limited to functional checks and did not include heat runs. One unit underwent a test run in the factory. However, during this factory acceptance test, the manufacturer blocked through-rotor flow to keep windage and friction losses within the allowable 10 percent over the guaranteed value. The original generator design was to supplement the fan-/pole-generated air flow with through-rotor air flow. This modification to the through-rotor flow reduced air flow in the generator, with the coolers in the closed circuit, to 19.5 cubic meters per second (cms), or about 90 percent of the design flow. As a result, the Roxburgh generators operate at higher temperatures than most other hydro generators in the country. Through trial and error, plant personnel determined this thermal problem was the underlying cause of the multiple stator failures discussed in this article.

Making modifications

Roxburgh personnel believe the work performed to date has addressed the stator cooling problem. Modifications performed to correct overheating of the units include: a partial return to the original design of through-rotor cooling; changing the way water passes through the cooler and tube fin spacing; changing the core air duct configuration; and making the coolers slightly larger (wider). All this has increased the volume of air circulating within the unit by about 28 percent, or to about 25 cubic meters per second. s

Colin McDonald, generation engineer at Contact Energy Limited’s 320-mw Roxburgh hydro project, has worked in the generation industry for more than 40 years.

Mr. McDonald may be reached at Contact Energy Limited, Fruitgrowers Road, P.O. Box 25, Clyde 9341 New Zealand; (64) 3-4400324; E-mail: colin.mcdonald@contact-energy.co.nz.

© Refurbishing Roxburgh

In 2002, Roxburgh personnel implemented a major refurbishment program for all eight units. This involves complete refurbishment, including installing a new stator core and winding, reinsulating the rotor poles, refurbishing the turbine runner and wicket gates, replacing the wearing ring on the turbine shaft, repainting the penstock and head gate, regalvanizing the intake screen, replacing the stator air coolers, and refurbishing other mechanical components as required.

The turbine runners only require refurbishing, which is basically cavitation repair and returning runner bands to design tolerances. Replacement cannot be justified because the runners are in good condition for their age.

As of January 2007, five machines (G3, G5, G6, G4, and G1) have been completely refurbished. G8 has undergone an electrical refurbishment, and the mechanical refurbishment is scheduled for 2008. The sequence of work is based on partial discharge analysis and dissection results, plus the condition of the core anchor plate and the mechanical components.


To access this Article, go to:
http://www.hydroworld.com/content/hydro/en/articles/print/volume-15/issue-4/articles/cover-story/a-retrospective-on-generator-failures.html