Structural health monitoring data analysis for ageing fixed offshore wind turbine structures
The objective of the present study is to perform a systematic data analysis of structural health monitoring data for ageing fixed offshore wind turbine support structures. The life-cycle extension of the first offshore wind farms is under serious consideration since the support structures are still...
| Published in: | Volume 2: Structures, Safety, and Reliability |
|---|---|
| Main Authors: | , , |
| Format: | Article in Journal/Newspaper |
| Language: | English |
| Published: |
American Society of Mechanical Engineers
2021
|
| Subjects: | |
| Online Access: | https://vbn.aau.dk/da/publications/d9943615-e157-4574-bffd-9e6f343961a0 https://doi.org/10.1115/OMAE2021-63007 http://www.scopus.com/inward/record.url?scp=85117084268&partnerID=8YFLogxK |
| Summary: | The objective of the present study is to perform a systematic data analysis of structural health monitoring data for ageing fixed offshore wind turbine support structures. The life-cycle extension of the first offshore wind farms is under serious consideration since the support structures are still in a condition to be used further. Big data analytics and machine learning techniques can aid to extract useful information from the monitoring data collected during the service life and build models for future predictions of an optimal life-extension. To this end, it is aimed to analyse the big data provided by embedded control systems and non-destructive inspections of ageing offshore wind turbine support structures using pre-processing techniques, including denoising, detrending, and filtering to remove the noise of different nature and seasonality as well as to detect the signal-specific contents affecting the structural integrity in the time and frequency domain. The effectiveness of the Welch method is investigated in terms of dealing with noisy signals in the frequency domain. Besides, the principal component analysis is carried out to reduce the dimensionality of the data and to select the most significant features that are responsible for most of the variance in the structural health monitoring data. Moreover, nonparametric statistical methods are used to test whether the data before noise being added and the data after cleansing the added noise came from the population with the same distribution. Further, permutation (randomisation) testing is performed to predicate that the results of the nonparametric test are statistically significant. The outcome of this study provides refined evidence that enables to feed the condition monitoring data into the training of the deep neural network to be able to discriminate different structural conditions. |
|---|