Principal component analysis as an exploratory and diagnostic tool to analyze S-N fatigue trends relevant to offshore wind foundation design

Fatigue is one of the main causes of failure in offshore welded structures, including offshore wind turbine foundations, due to repeated application of environmental and operational cyclic loads. The fatigue data gathered from the tests, which are used in the design of offshore welded structures, typically include a degree of scatter due to the presence of weld defects, weld geometry, misalignment, and other material inhomogeneities. For these reasons, the design stage of a welded structure is of critical importance. For a reliable fatigue design, engineers usually calculate the line of best fit to correlate the logarithm of the nominal stress range with the logarithm of the number of cycles to failure using least squares linear regression (LSLR) approach, followed by estimation of the scatter in the data through the residual standard error (in the field, commonly referred to as “standard deviation”), along with the coefficients of the model. LSLR is a common statistical tool that is usually employed to analyze data obtained from new experiments or from the literature relevant to a specific application. The “Stress range vs. Number of cycles to failure” (S-N) curves recommended for the design of welded structures in international standards are no exception and are usually derived by combining linear regressions with engineering judgment. In this study, Principal Component Analysis (PCA) is used to explore fatigue trends, filter fatigue data, and create new regression and classification models. A data set comprising 320 fatigue tests is analyzed on S355 double V transverse butt-welded plates relevant to offshore wind applications. Regressions on both PCA-filtered and unfiltered fatigue data are presented and compared, and since PCA describes the main directions of variance, it is shown how it can be used as a diagnostic tool to study the suitability of a dataset for further fatigue analyses.