Thesis
Failure diagnosis for offshore wind turbines with low availability of run-to-failure data
- Creator
- Rights statement
- Awarding institution
- University of Strathclyde
- Date of award
- 2023
- Thesis identifier
- T16543
- Person Identifier (Local)
- 201867626
- Qualification Level
- Qualification Name
- Department, School or Faculty
- Abstract
- Despite the efforts to achieve a through-life reliable design and the attempts to control the failures of wind turbines, some system failures are inevitable. The inherent requirement for cost, material, and weight optimisation, together with the extreme operating conditions, can lead to unexpected failures. This is true for land-based turbines and has an even greater impact on offshore wind systems, where the harsh environment and the high cost of the assets and logistics increase the importance of a proactive approach to the system’s maintenance. The smart management of an asset starts with the identification of the health status of its systems, to take cost effective decision on how and when maintain it. The first level of the detection of an anomality in the system comprises the recognition only of the failed status of the asset (level I). Following, the location of the failure should be identified (level II), followed by the detection of its degree of severity (level III) and consequences (level IV). Depending on the availability of continuous monitoring data, historical databases, and advanced numerical models, different frameworks can be established for the failure diagnostics and prognostics. This thesis investigates on the use data-driven, model-based, and digital twin solutions to support the diagnosis of failure events of offshore wind turbine systems characterised by a low availability of run-to-failure data. This topic is of major concern for either the current installations - for which the collection of data is restrained either to only few assets or to more cost-effective temporary monitoring campaign – and the new offshore wind technologies (e.g., floating wind, large-MW structures), for which no or only a limited amount of operating data has been gathered. The mechanical failure of the components of the offshore wind speed conversion system can have a significant impact to the operational expenditure and can be associated to a significant loss of production of the offshore wind farm. The detection of their incipiency has been extensively investigated by machine and deep learning techniques on big sets of condition monitoring and operational data. By contrast, this research explores the implementation of transfer learning to detect anomalies in an offshore wind gearbox with low availability of representative failure data. To move towards the quantification of the consequences of such a failure (level IV), a case of study is used to explore then most suitable the model-reduction techniques to be applied to a full aero-servo-elastic model of the offshore wind turbine. Such a numerical model is the basis for the development of digital twin technology; it is aimed at capturing the only the essential dynamics while targeting the degree(s) of freedom indicating the presence of the failure mode. The presence of a damage in the offshore wind foundation is not commonly recorded, yet structural failures can either lead to catastrophic consequence or considerably increase the cost of maintenance for the planning of expensive subsea inspections. In particular, the fatigue-driven offshore wind jacket foundation designs are sensitive to extreme site conditions, and their expected lifetime can decrease considerably if exposed for a long time to undetected phenomena such as scour and corrosion. This research demonstrates the feasibility of a vibration-based diagnosis (level II) of several damage scenarios for a jacket substructure of an offshore wind turbine. Considering than only a percentage of the assets in the farm are likely to be instrumented with a high-frequency structural health monitoring system, the feasibility of the detection (level I) of a structural failure mode via low-resolution operational data is additionally explored. These virtual monitoring frameworks are supported by the deployment of the digital twin technologies for their setup and their future field application.
- Advisor / supervisor
- Smolka, Ursula
- Kolios, Athanasios
- Resource Type
- DOI
- Funder
Relations
Items
| Thumbnail | Title | Date Uploaded | Visibility | Actions |
|---|---|---|---|---|
|
|
PDF of thesis T16543 | 2023-04-12 | Public | Download |