PSI - Issue 19

Edrissa Gassama et al. / Procedia Structural Integrity 19 (2019) 711–718 Gassama et al./ Structural Integrity Procedia 00 (2019) 000 – 000

712

2

recommended approach for determining the life of welded joints (API 579-1/ASME FFS-1 (2016)). In an upcoming release of API 579, a new part will be introduced that is specific to the assessment of piping vibrations (Breaux et al. (2016), Bifano et al. (2018)). The fatigue criteria in the 2016 release of API 579 ensures adequate protection against high cycle fatigue in the design stage, however, when applied to in-service equipment (Fitness-For-Service) the welded fatigue methodology can be overly conservative for vibrating equipment. For vibration applications, welded components have commonly been observed to successfully operate in service for several years at high frequencies (e.g., frequencies greater than 1 Hz), surviving for more than 10 9 cycles without failure. When assessing these components using the Master S-N curve, API-579 currently recommends extending of the curve into the high cycle regime, predicting lives much less than that observed in practice. The set of data used to fit the Master S-N curve primarily consists of cases where failure was observed in less than 10 7 cycles. Few data points in the set have lives beyond 10 7 cycles, and no data points have lives beyond 10 8 cycles. As a result, the utilization of the Master S-N curve in the very high cycle fatigue (VHCF) regime can lead to overly conservative risk evaluations and costly mitigation measures. . Indeed, with the lack of available data in the VCHF regime, the true behavior is an unknown. As a result of the uncertainty related to fatigue life predictions in this regime, existing Codes and Standards differ in their recommendations. A very common approach, taken by many, is to prescribe a bilinear curve, accounting for longer lives in the VHCF regime by reducing the slope on the curve (PD 5500 (2006), BS7608 (2014), DNV-RP-C203 (2011), ASME B31.3 (2018), Nureau Veritas NT 3199 (2013), Hobbacher (2016)). While not always the case, often the bilinear curve is reserved for situations of variable amplitude loading, while endurance limits are implemented in cases of constant amplitude loading (BS7608 (2014), DNV-RP C203 (2011)). While adopting a bilinear curve to capture longer lives is a popular approach, there remains a debate as to what the second slope should be, and where the intersection of the two slopes should lie. In reality, it is not expected that there would be an abrupt kink in the fatigue life curve. Moreover, the transition from one slope to another introduces difficulties when trying to apply these bilinear curves in probabilistic approaches. The Master S-N curve, using equivalent structural stress as an index, has made tremendous strides to condense a wide range of fatigue tests to a narrow range, however, the variation in fatigue life at any level of equivalent structural stress can still span an order of magnitude. As a result of this inherent uncertainty, fatigue life predictions are often performed using probabilistic approaches (e.g. Monte Carlo). When performing a probabilistic analysis, any kink in a bilinear curve introduces a discontinuity in the standard deviation at this point. This discontinuity introduces either subjectivity or errors into the analysis, thus promoting the need for a smooth, continuous fit to the data. This paper explores one method of doing just that, in a statistically consistent manner, by incorporating the information stored within run-out test data. When performing an experimental fatigue test, it is common practice to stop the test after a fixed number of cycles. This is done for several reasons, but often it is for the practical purpose of reducing the time and costs associated with performing the test. The data points from tests that are stopped (run-outs) are often either ignored when fitting fatigue data or included as though they were points of failure. When either of these approaches is taken, the practitioner has, often unknowingly, introduced their own bias into the results and both approaches result in conservative estimates of fatigue life in the VHCF regime. Incorporating run- out or “ censored ” data (i.e., data in which the test was stopped so the true life is unknown) into a statistical analysis is not a novel approach (see, for example, Schmee and Hahn (1979), Chatterjee and McLeish (1981)). One of the most common approaches for performing a statistical fit of data containing censored lives is to use the maximum likelihood estimate approach. However, for developing standardized procedures that can be performed by all levels of practitioners, a simplified approach will be discussed herein that yields significantly the same results as the maximum likelihood estimate approach (Schmee and Hahn (1979)). In this paper, we explore one way in which run-out data points may be incorporated into a linear least-squares regression procedure, lessening the bias introduced by ignoring them or treating them as actual failures. Section 2, outlines the data fitting procedure, and discuss how the run-out data is included in the fit. In Section 3, the data collection procedure for refitting the Master S-N curve is described. Section 4 demonstrates how censored data can be included to develop a modified Master S-N curve and discusses the influence of the new fit on the predicted lives of components in the VHCF regime. Finally, Section 5 provides concluding remarks.