The Equivalency-Based Linear Regression Method for Statistical Analysis of Creep/Fatigue Data

Materials scientists and mechanical engineers working on structural integrity are making increasing use of statistical analysis in interpreting creep/fatigue data as they contain an inherent scatter which cannot be substantially reduced even under controlled testing conditions. In practice, in most cases the uniaxial failure or cracking data can be reasonably approximated by a straight line on log-log coordinates, indicating that there is a linear log relationship with the appropriate correlating parameter. Linear regression is the most used method in statistical data analysis and is being recommended in American Society for Testing and Materials (ASTM), British Standards (BS), Det Norske Veritas (DNV) and many other engineering standards. Recently, the current practice on linear regression as adopted by the engineering standards has been critically reviewed, and the shortcomings of these procedures have been clearly demonstrated. A new statistical method based on the equivalency between all variables involved has been proposed for S-N curve analysis. In this paper, a large amount of creep and fatigue data of engineering materials collected from several well known data bases generated in US, Europe, and Japan are systematically analyzed with conventional standard and the new equivalency method. The results are compared and discussed. Finally, a recommendation to improve the fitting parameters taking into account of the scatter in both axes is presented.