Over the past decades, the wind energy sector has been growing significantly and efforts are being made to minimise the overall cost of a wind farm. One of the main cost drivers is directly related to operation and maintenance (O&M) actions. Current tendencies in O&M practice are shifting from rather costly corrective strategies to preventive and predictive approaches. Crucial for setting up these cost effective strategies is to understand profoundly when and how wind turbine (WT) components fail. Moreover, the failure severity, in terms of caused downtime and repair cost, as well as the frequency of failure occurrences need to be known. These can be obtained from analysing historical failure databases and maintenance logbooks provided by manufacturers and operators. The components and their sub-assemblies have to be classified regarding their physical location and functionality, using a so called taxonomy or component breakdown. Then, the frequencies of component failures and the resulting WT downtimes are derived from the failure database. The outcome of the analysis can then be used to build reliability models and failure prediction tools in order to estimate the WT component degradation over time and to anticipate failures. However, there are still serious problems regarding recent practices and one of the main issues is lack of available failure data. Due to the lack of available data, many reliability models and maintenance decision tools are based on assumed failure rates – not representing sufficiently well the reality. Supervisory Control and Data Acquisition (SCADA) alarms are proposed to complement available failure logs with additional information. Most modern WTs are equipped with SCADA systems, generating a huge amount of information that can be obtained mostly free of additional cost.
In order to investigate the correlation between the alarms extracted from the SCADA system and the actual failure occurrences, the data from the most widely installed modern technologies had to be analysed. As older turbines are not necessarily equipped with SCADA systems or only operate relatively limited ones, they were excluded. The different technologies are indicated by their rated power and drive train setup – being either direct drive or geared WTs. As for confidentially reasons no manufacturer names can be published, the WT makes are indicated by the letters A to G. The respective SCADA system used within these turbines is referred to with the numbers 1 to 5. Turbine types A, D, E, F, G are equipped with a DFIG and types B and C with a synchronous generator. In total 440 WTs were analysed over a period of three years, resulting in 1320 operational years. An overall number of 653 failures and 1345036 alarms were registered and processed. The failures and alarms per turbine are displayed as rounded values.
Figure 1 displays the possible alarms for each SCADA system. Figure 2 shows the composition of the alarms actually recorded for each system within the given observation period. They consist of alarms related to a specific WT component, alarms due to extreme environmental conditions, and others that could not be assigned to any component, e.g. grid restrictions. Comparing the two figures, shows that for WT types A, B, C and D many weather related alarms were recorded, indicating extreme conditions, which could be responsible for certain component failures. Turbines B, C and D showed a quite similar share of the three alarm categories recorded. Also the share of possible alarms of these two SCADA systems are alike.