Thesis
Harnessing data for wind turbines : machine learning digital-enabled asset management strategies
- Creator
- Rights statement
- Awarding institution
- University of Strathclyde
- Date of award
- 2024
- Thesis identifier
- T17007
- Person Identifier (Local)
- 201868896
- Qualification Level
- Qualification Name
- Department, School or Faculty
- Abstract
- As interests in offshore wind farms continue to grow, so does the demand to reduce the cost of energy (COE). Maintenance cost and downtime can reduce the COE through greater information on offshore wind assets concerning the operational loads and structural integrity. This has had a significant impact on the interests of digital-enabled asset management (DEAM) using digital twins. Digital twins’ technologies can replicate operational assets computationally, providing more information and increasing one’s confidence in operations and maintenance (O&M). DEAM is a multi-disciplinary field and making advances in this field requires aspects of multiple modelling domains, this thesis aims to develop this and help aid in the future of DEAM. The work carried out in the thesis has been validated against operational data recordings from offshore structures. This provides value and confidence to the results of the state-of-the-art models for real-world engineering systems. This research presents a portfolio of four research areas that have been published in a variety of peer-reviewed journal articles and conference papers. The areas are: 1) A proposal for standardisation of pre-processing data. Current standards have not addressed how to deal with data for machine learning, and this paper aims to begin this discussion with an example. This work implements a trend condition monitoring model that makes predictions on the power of an offshore structure using supervisory control and data acquisition (SCAD) data. There are 5 different machine learning (ML) models used and the data is validated using unused data with the modelling errors quantified. 2) A novel approach to dealing with the limitations of small data sets. This is an innovative way of transferring information from a homogeneous population to increase the accuracy of an artificial neural network (ANN). The ML model is a comparison of a conventional ANN compared to the proposed hard-parameter transfer ANN model. The ML model makes a classification of the error signature from the gearbox using both SCADA data and condition monitoring system (CMS) data. The validation of the comparison uses unseen data during the training process and the errors are measured. 3) Is a case study on Wikinger offshore wind farm population homogeneity where the operational and environmental conditions are compared for all three wind turbines. This case study provides a framework to follow when investigating an offshore wind farm population. This uses operational data from three wind turbines with both SCADA, CMS data, and processed data from RAMBOLL. The outcomes from this paper are used to determine the type of ML model used in the last study. 4) Is the model development of a population-based structural health monitoring (PBSHM) model. This study investigates three domain adaptation techniques suited to strong homogeneous populations. The ML model takes SCADA data as an input and predicts the damage equivalent moments (DEM) on the jacket foundation structure. To validate the PBSHM model data from a structure that is not used during the training of the model is used to quantify the precision of the model. The individual contributions of the developments in each of the constituent areas relate to an overall improvement in modelling approaches that are necessary for DEAM and aid in the realisation of true digital twins. All the areas relate to offshore wind ML and are related to operational data. The link between the measured data and the individual models aid in gaining more information and greater insights into the O&M.
- Advisor / supervisor
- Brennan, Feargal
- Kolios, Athanasios
- Resource Type
- DOI
Beziehungen
Objekte
| Miniaturansicht | Titel | Hochladedatum | Sichtbarkeit | Aktionen |
|---|---|---|---|---|
|
|
PDF of thesis T17007 | 2024-06-26 | Öffentlich | Herunterladen |