Petrobras Energia’s Genelba combined cycle plant began operation in 1999. Over several years (up until early 2005) the plant’s records showed a 20% failure rate for gas turbine start-ups. In late 2004, Genelba initiated a statistical study, and on the basis of this has taken steps that have significantly reduced the failure rate to practically 0%. The study was carried out by a multidisciplinary team consisting of engineers from Genelba and the gas turbine manufacturer, Siemens.

Start-up problem

For the purposes of the study, ‘failed GT start-up’ meant that a gas turbine start-up process had been interrupted by a ‘flame-off’ signal before the turbine synchronised with the electrical grid.

This can occur in two ways. First, the ‘flame-off’ signal appears immediately after the fuel gas emergency stop valve is opened, that is, the flame-on condition is never reached. Second, after the flame-on condition is reached, the ‘flame-off’ signal appears before synchronisation with the grid.

The former is less harmful because it only causes a delay in plant start-up, although its impact should not be underestimated. In the latter case, by contrast , the useful life of the turbine hot path components is affected by their having undergone the flame-on condition. This second situation is more damaging because parts of the hot path suffer wear.

In general, gas turbine manufacturers cause the number of hours registered on the equivalent operating hours (EOH) counter to increase whenever the fuel gas emergency stop valve is open and flame-on conditions occur (sometimes the outlet temperature value is also taken into consideration). On a failed start up, the counter increments by a predetermined amount, typically 10 EOH.

Problem history

From the moment Genelba commenced operation, frequent problems during gas turbine start-ups had been observed, but other major problems typical of a plant in the commissioning phase diverted attention from the start-up failure rate. Not long after, and thanks to the implementation of robust process optimisation programmes, the plant’s main problems were being solved one by one. Nevertheless, start-up failures still figured on the unavailability pie chart.

There were other factors that contributed to the problem being unresolved for some time: the relatively small number of gas turbine start-ups a year in base-load operation; lack of a clear pattern of failure occurrence; and several cases where successive start-up attempts failed.

As regards the first factor, the rate of gas turbine start-up was about 10 per turbine during 2004. There was a total of only four failed start-ups during the year, representing a 20% failure rate.

The lack of a clear behavioural pattern is accounted for by the fact that failed start-ups were found to occur in a variety of situations. These included start-ups at all gas turbine temperatures, those after trips and scheduled shutdowns, those after long and short shutdowns, and those with or without previous off-line compressor washing.

Regarding the cases of repeated falures, several failed atttempts were often necessary before successful start-up. During such an event, emergency decision-making was commonplace and once the plant had been started successfully it was often assumed the problem had been resolved. Unfortunately it was usually the case that after some time and several successful starts the problem would re-occur.

Consequently, in early 2005, an ‘improvement team’ was formed to analyse the failures. It comprised people from operations, maintenance, and processes as well as engineers from Siemens. As a considerable amount of time had passed since first operation of the plant, a significant quantity of relevant historical data about turbine start-ups was available. These data would provide valuable information for a statistical analysis.

Statistical analysis

At the outset the team had data from 301 start-ups relating to both gas turbines. It was particularly important to make every possible effort to interpret the available data since the plant mode of base load operation did not provide many opportunities to specify and carry out many trials.

First, the total rate of failed start-ups (Figure 1) and those particular to GT11 and GT12 were analysed (Figures 2 and 3). It was observed that both turbines behaved similarly, ruling out the possibility that a specific problem could be directly attributed to one or other gas turbine.

Next, the development of problems was analysed as a function of time. For this it was not appropriate to use calendar year time intervals but rather periods representing machine life. 25000 and 50000 operation hours, corresponding to major inspections, were used as milestones to delineate relevant time periods.

Figure 4 illustrates the start-up failure pattern over time for GT11 and GT12. An increase can be seen in the percentage of failed start-ups immediately after the 50 000 EOH major inspection. This is apparent for both gas turbines. At this stage it was decided to initiate the study. During this period some trial start-ups were carried out, so the increased proportion of failures in this period was not considered significant.

Other important aspects considered worth studying were cases of successive start-up failures (Figure 5). It is significant that only 11 of the 61 failed start-ups did not exhibit repeated failures of successive start-up attempts. The other 50 failed starts (2×7+ 3×6 + 4×2 + 5×2) were cases of repeated failed start-ups. That is, there were 28 independent failed start-up events (11 + 7 + 6 + 2 + 2) for which at least one additional attempt was necessary to achieve a successful start-up.

Consequently the team studied the relationship between a failed start-up and the immediately following attempts. Figure 6 shows an 80% probability of a failure after the occurrence of a failed start-up. The failure probability of any start-up averaged 20% according to Figures 1, 2 and 3. But once a start-up fails, the failure probability increases to 80% for the next attempt and decreases again below 20% only after three attempts (Figure 6) following the initial failure. It was concluded therefore that most of the time failed start-ups required several attempts to succeed and that these successive attempts should not be considered independent events.

Two further relationships were also studied, namely that between failed start-ups and gas turbine trip, and that between failed start-ups and prolonged (>2 days) gas turbine shutdown (Figures 7 and 8). A weak correlation was observed in both cases.

Following this analysis, it was pertinent to ask (having taken into account all the independent failed start-up events) how many of them had occurred after gas turbine trips and long shutdowns. Figure 9 shows this relationship. Apart from 4 starts that do not conform (some of them dating from 10 years ago, before detailed data was being collected), the remaining ones occurred after a turbine trip (13) or after a prolonged shutdown (12). It is worth mentioning that the 28 failed start-up events are independent failures according to the conclusions of Figure 6. These 28 independent occurences of failed start-up events, with their successive repetitions, generated the 61 failures indicated in Figures 1, 2 and 3.

In the last of their statistical analyses, the team studied which of the 61 failures added EOH (Figure 10). Of the total of 61, 31 reached the flame-on condition before the flame blew out. After a detailed analysis of the 13 cases, the team concluded that all of them occurred after a prolonged gas turbine shutdown.

As a next step, the team proceeded to construct an “affinity diagram”. This technique allows ideas to be organised and summarised into groups, which, when used with the results of the statistical analyses, can assist in finding the possible causes of failure. The groups turned out to be the four basic systems involved in the flame ignition in the gas turbine combustion chamber: fuel gas system, combustion air system, flame ignition system, and flame detection system.

Implemented measures

It was concluded that there were three main root causes of the failed gas turbine start-ups: presence of air in the natural gas piping; excess/lack of natural gas depending on the type of start-up (cold, warm or hot); and presence of water in the natural gas piping. The variety of causes accounted for the lack of a clear pattern of behaviour, which in turn, delayed finding a final solution to the problem.

With the results obtained from the statistical analyses and the affinity diagram, the team came up with various measures for implementation such as gathering information from the worldwide fleet, possible ignition with fuel gas pilots, and so on. However, three key measures were identified:

1. Adjustment of an f(x) function (the imput variable of which is the machine temperature) inside the gas turbine: this function adjusts the fuel gas control valve lift at start-up depending on the gas turbine temperature. This control, present in the original design, had not been adjusted correctly during plant commissioning.

2.The implementation of a new filling procedure for the fuel gas piping: owing to internal safety measures, every time a combustion chamber inspection is carried out a 15m piping section between the fuel gas ESV and a shut-off valve in the upstream is purged. Once the work is completed, the piping must be filled with natural gas again. The new venting procedure involves a repetitive operation with the following sequence: opening of the shut-off valve; piping pressurisation with natural gas; closure of the shut-off valve; and opening of the piping vent valve until a minimum 2 bar pressure is reached before closure of the vent valve. This sequence is repeated three times to ensure no air is left in the piping.

3.The study and development of systems to extract remaining water in the gas lines in the diffusion combustion mode, while the gas turbine is operating in premix combustion mode.

The first two measures were implemented immediately, although the third is more complex. However, Siemens has designed a specific system to address this problem, the installation of which is expected to take place at the next gas turbine minor inspection.

The results

Overall, the results were pleasing, with failed start-ups reduced to nearly 0%. Since the implementation of measures 1 and 2 during 2005 all start-ups have been successful with the exception of one event which required a further attempt. This failure can be attributed to to the fact that the third measure had not been implemented. The start-up occurred after a trip in a gas turbine that had been running for more than four months. Although this part of the problem is still unresolved, its root cause is now well known.

The results of a benchmarking programme in which Genelba participates are illustrated in Figure 12. It shows comparative reliability data for the start-up process. It is worth noting Genelba’s situation compared with other similar power plants before and after the study described in this paper.

Statistical analysis of historical data is an invaluable learning tool which should always be implemented before beginning an intuitive trial and error process. The results of these analyses often provide information that is contrary to the specialists’ a priori conclusions. The method used in this case study for solving the failed start-up problem can be applied to other problematic situations occurring every day at power plants.

Figure 1. Total start-ups (both GTs)
Figure 2. GT11 start-ups
Figure 3. GT12 start-ups Figure 4. Start-up failure rate versus running time Figure 5. Cases where there were successive start-up failures Figure 6 . Failure probability following a start-up failure Figure 7. Failed start-ups following GT trip Figure 8. Failed start-ups following prolonged shutdowns Figure 9. Relationship between failed start-ups, trips and prolonged shutdowns Figure 10. Failed start-ups which added to EOH Figure 12. Benchmarking position (before and after Genelba’s start-up improvement programme)