Predictive maintenance of a wind farm / Overheating detection
News - 18 January 2021
The current maintenance strategy for wind farms is mainly periodic, based on a calendar of operations. However, between two operations, it is not possible to foresee the appearance of a major malfunction in the near future: these defects are therefore generally dealt with at the most critical moment, which generates heavy logistics costs and a consequent machine downtime (as an example: the cost of replacing a large component such as the gearbox is more than €450,000, and the operation can extend over several weeks). The origin of this problem is linked to the lack of knowledge of the evolution of the degradation leading to the breakdown: what is the source of the breakdown? When is it relevant to prepare a maintenance operation?
In recent years, VALEMO’s engineering department has developed solutions with the ultimate aim of adopting a predictive maintenance strategy: based on a relevant exploitation of targeted SCADA (Supervisory Control And Data Acquisition) data generated by the PLCs present on the machines, the algorithms developed must enable the operator to intervene early enough to prevent a malfunction in the near future, in order to minimise maintenance costs by avoiding a much heavier maintenance operation.
A method for monitoring the condition of a machine has therefore been developed with a focus on the generator bearings. The principle is to use the variables having a causal link between them (a modification on one of the variables implies a modification on all the others) in order to have the most accurate possible estimate of a variable to be monitored. In the case of the generator, the bearing temperature (front and rear) is estimated using the active power, the speed of rotation of the fast shaft and the temperature of the nacelle, and is then compared with the actual measured value. An indicator is then obtained, to which a threshold is assigned, above which it is considered that the turbine being monitored is faulty and that a maintenance operation on the generator is necessary.
In the literature, previous studies use a single-turbine or multi-turbine approach. In the first case, only the data of the monitored turbine is considered, which has the advantage of having results that are not very sensitive to changes in operating mode (from a certain wind speed, the power stabilises at its nominal value). In the second case, the data of the whole park are considered (usually the median is used), which allows to get rid of the influence of seasonal cycles. The originality of the method proposed by VALEMO is to merge these two approaches in order to benefit from both advantages (2).
The solution was first established and then validated on a park, which we will later name A, with three localised defects on the generator, one of which was accompanied by a complete replacement of the generator. The periods are specified in Table 1. The results are very promising since in all three cases the anomaly (threshold overrun) is detected with a considerable advance (at least one week) in relation to the recorded production stoppage date.
To facilitate decision support, as well as to give the operator a certain freedom of choice, in order to better understand the effects of his decision (1), a hierarchy of alarms has been set up, illustrated by a simple colour system :
- Yellow (C3): For information purposes, no intervention is necessary,
- Orange (C2): Potential fault, no intervention required,
- Red (C1): defect, intervention necessary.
In Figure 1, we can see that our algorithm makes it possible to detect a fault as early as the first week of January, i.e. more than a month before the turbine shutdown date (27 February), which would have made it possible to set up a preventive maintenance operation early enough to anticipate the replacement of the generator and thus the weeks of shutdowns that were caused.
Following these initial results, it was necessary to verify, a few years later, the flexibility and reliability of the solution: the monitoring tool was thus applied to all the parks operated by VALEMO over their entire operating life.
On this occasion, a more global representation was adopted (Heatmap), displaying for a given fleet and operating period the number of C1 alarms (represented by a red colour ranging from the lightest to the darkest) per month and per turbine. This representation gives an overview of a new fleet to be studied, and above all allows you to quickly locate any fault periods to be studied. On Figure 2, which illustrates the application of this representation over 5 consecutive years of operation of a fleet B made up of 6 machines, a rapid diagnosis can be made: a significant number of C1 alarms are present during three periods identified by a bright red colour, but it is in October 2018 on the T6 turbine that the most alarms are observed (the darkest red colour). Following this example, the corresponding indicators would then be displayed for the periods under consideration, and these observations would be validated using the available monthly operating reports.
Finally, the performance of the solution developed by VALEMO has been evaluated in terms of detection advance in days (period between the date of first detection and the date of machine shutdown). The performances of the A fleet are presented in Table 2. For the 3 faults considered, a period of more than one week is available for decision-makers to plan an appropriate maintenance operation.
|Default||Rear bearing E6||Front bearing E9||Winding E11|
|Advance to detection||29||10||14|
As the monitoring tool offers very promising results at the generator level, we are now going to focus our developments on the complete diagnosis of the wind turbine, in order to obtain a complete and real-time monitoring of the state of the wind farms, leading to a gain in productivity and optimisation of maintenance operations in terms of time and costs.
- Castanier B, Zhu W, Bettayeb B. A decision-support tool for planning the maintenance operations of an offshore wind turbine. Congress Lambda Mu 20 Maîtrise Risques Sûreté Fonctionnement, 11-13 October 2016 St Malo Fr. 2016;
- Lebranchu A. Analysis of monitoring data and synthesis of fault and degradation indicators for predictive maintenance support of wind turbine farms.