the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Anomaly-Based Fault Detection in Wind Turbine Main Bearings
Lorena Campoverde-Vilela
María del Cisne Feijóo
Yolanda Vidal
José Sampietro
Christian Tutivén
Abstract. Renewable energy is a clean and inexhaustible source of energy, so every year interest in the study and the search for improvements in production increases. Wind energy is one of the most used and therefore the need for predictive maintenance management to guarantee the reliableness and operability of each of the wind turbines has become a great study opportunity. In this work, a fault detection system is developed by applying an anomaly detector based on principal component analysis (PCA), in order to state early warnings of possible faults in the main bearing. For the development of the model, SCADA (supervisory control and data acquisition) data from a wind park in operation are utilized. The results obtained allow detection of failures even months before the fatal breakdown occurs. This model requires (to be constructed) only the use of healthy SCADA data, without the need to obtain the fault history or install additional equipment or sensors that require greater investment. In conclusion, this proposed strategy provides a tool for the planning and execution of predictive maintenance within wind parks.
Lorena Campoverde-Vilela et al.
Status: final response (author comments only)
-
RC1: 'Comment on wes-2022-111', Davide Astolfi, 13 Feb 2023
The manuscript entitled “Anomaly-Based Fault Detection in Wind Turbine Main Bearings” deals with a very interesting topic, which is perfectly adequate for the scientific objectives of the journal.
In a nutshell, the authors propose a PCA-based alarm raising method for diagnosing incoming damages to the main bearing of wind turbines. The method is based on SCADA data mining.
The work is well written and well presented. The workflow is very clear and presented in detail, such that it can be replicated by scholars.
The peculiarity of the work is that only exogenous variables (environmental) and the temperature of the component of interest (main bearing) are employed.
Therefore, in general I have a very positive opinion on this work. Nevertheless, there are some aspects which could be discussed more in deep.
1. A considerable number of studies has been recently devoted to this topic. Therefore, I recommend that the authors highlight more clearly the innovative contribution and the points of strength of their work.
2. The authors employ almost three years of data for model training. For the necessities of real-time wind farm monitoring, it is not obvious that such amount of healthy data is available. Could the authors discuss their models’ performance with shorter training data sets? I suggest the following reference: Turnbull, A., Carroll, J., & McDonald, A. (2022). A comparative analysis on the variability of temperature thresholds through time for wind turbine generators using normal behaviour modelling. Energies, 15(14), 5298
3. The authors obtain a result similar to that obtained, for example, in the recent paper Murgia, A., Verbeke, R., Tsiporkova, E., Terzi, L., & Astolfi, D. (2023). Discussion on the Suitability of SCADA-Based Condition Monitoring for Wind Turbine Fault Diagnosis through Temperature Data Analysis. Energies, 16(2), 620. The main bearing temperature is the most adequate target to monitor for raising an alarm, but there is an issue related to the capability of the model in locating adequately the fault. In this work, using the main bearing temperature, a fault regarding the main bearing itself and a fault regarding the gearbox are diagnosed similarly. This occurs also in the paper which I have indicated. Therefore, I am wondering if the authors have ideas for further developments regarding the issue of precise fault location.Citation: https://doi.org/10.5194/wes-2022-111-RC1 - AC1: 'Reply on RC1', Yolanda Vidal, 02 Mar 2023
-
RC2: 'Comment on wes-2022-111', Anonymous Referee #2, 13 Feb 2023
The comment was uploaded in the form of a supplement: https://wes.copernicus.org/preprints/wes-2022-111/wes-2022-111-RC2-supplement.pdf
- AC2: 'Reply on RC2', Yolanda Vidal, 02 Mar 2023
-
RC3: 'Comment on wes-2022-111', Anonymous Referee #3, 17 Feb 2023
The authors developed a method to detect wind turbine main bearing failures in an early stage, hence the work fits good to the scope of the journal. They used an anomaly detector based on principal component analysis to detect failures of a main bearing with the help of SCADA data. To train the model and to evaluate the results they used the data of 18 turbines.
The structure is clear, and the steps are described in detail. The overall quality is very good, some minor suggestions in the technical comments may help to improve it a little bit.
It is perfectly fine that the focus is on the model, the selection of data and data processing. Nevertheless, in my opinion the technical background could be highlighted more.
Special commentsE.g. the work of Carrol et. al. (DOI: 10.1002/we.1887) could help to underline the importance to prevent failures and downtimes.
To give technical details of a WT is not necessary. In my opinion the power curve in figure 1 does not give any contribution to this work. The lines from 97 to 102 could be deleted. Here a reference to other publications like Hansen would be possible as well. However, the authors do not give information about main bearings. Possible questions are: Which kind of suspension do the turbines have? Why do I need a bearing and what are possible bearing types? Maybe its not necessary to explain it in detail, but at least a reference would be welcome (Wenske 2022 DOI: 10.1049/PBPO142F or Hau ….). A cross reference to figure 2 can be done, too.
There are plenty of possible bearing damages (fatigue, wear cracks… they can occur at the rings, raceways, rollers or at the cage) which can have an effect on the bearing lifetime. This is not considered. Here I can recommend e.g. the work of Hard like DOI: 10.1002/we.2386. As a reference about bearing damages e.g. the work from Harris and Kotzalas could be used. The fact that just one main bearing failure occurs in the data, may raise the question if other main bearing failures can be detected. At least in the discussion or in the outlook I would expect a discussion on that.
The mentioned counteractions to prevent a bearing failure after detection stay very vague.
Technical comments
In Table5 and figure 5 units are missing.
In figures 11, 12, and 13 a same y-axis scale would make it easier to compare the individual turbines.
To reduce the number of plots it could be a good idea to summarize a few turbines in one plot. Different colors could be used.
Sometimes shorter sentences would increase the legibility. As one example would separate the sentence (in line 326) after the first date.
Citation: https://doi.org/10.5194/wes-2022-111-RC3 - AC3: 'Reply on RC3', Yolanda Vidal, 02 Mar 2023
Lorena Campoverde-Vilela et al.
Lorena Campoverde-Vilela et al.
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 245 | 93 | 15 | 353 | 4 | 3 |
- HTML: 245
- PDF: 93
- XML: 15
- Total: 353
- BibTeX: 4
- EndNote: 3
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1