1 Copyright © 2016 by ASME Proceedings of the ASME 2016 Pressure Vessels & Piping Conference ASME-2016-PVP July 17-21, 2016, Vancouver, BC, Canada PVP2016-63350 A STATISTICAL APPROACH TO ESTIMATING A 95 % CONFIDENCE LOWER LIMIT FOR THE DESIGN CREEP RUPTURE TIME VS. STRESS CURVE WHEN THE STRESS ESTIMATE HAS AN ERROR UP TO 2 % (*) Jeffrey T. Fong National Institute of Standards & Technology Gaithersburg MD 20899 U.S.A. fong@nist.gov James J. Filliben National Institute of Standards & Technology Gaithersburg MD 20899 U.S.A. filliben@nist.gov N. Alan Heckert National Inst. of Stand. & Tech. Gaithersburg MD 20899 U.S.A. alan.heckert@nist.gov Pedro V. Marcal MPACT, Corp. Oak Park CA 91377 U.S.A. pedrovmarcal@gmail.com Marvin J. Cohn Intertek, AIM Santa Clara CA 95054 U.S.A. Marvin.cohn@intertek.com ABSTRACT Recent experimental results on creep-fracture damage with minimum time to failure (minTTF) varying as the 9 th power of stress, and a theoretical consequence that the coefficient of variation (CV) of minTTF is necessarily 9 times that of the CV of the stress, created a new engineering requirement that the finite element analysis of pressure vessel and piping systems in power generation and chemical plants be more accurate with an allowable error of no more than 2 % when dealing with a leak- before-break scenario. This new requirement becomes more critical, for example, when one finds a small leakage in the vicinity of a hot steam piping weldment next to an elbow. To illustrate the critical nature of this creep and creep-fatigue interaction problem in engineering design and operation decision-making, we present the analysis of a typical steam piping maintenance problem, where 10 experimental data on the creep rupture time vs. stress (83 to 131 MPa) for an API Grade 91 steel at 571.1 C (1060 F) are fitted with a straight line using the linear least squares (LLSQ) method. The LLSQ fit yields not only a two-parameter model, but also an estimate of the 95 % confidence upper and lower limits of the rupture time as (*) Contribution of the U.S. National Institute of Standards and Technology. Not subject to copyright. basis for a statistical design of creep and creep-fatigue. In addition, we will show that when an error in stress estimate is 2 % or more, the 95 % confidence lower limit for the rupture time will be reduced from the minimum by as much as 40 %. 1. INTRODUCTION In reporting engineering observations such as materials property test data involving two variables, the most common practice is to fit the data with a straight line. An example of this is given in Fig. 1, where 25 observations of variable X1 (pounds of steam used per month) vs. variable X8 (average atmospheric temperature in degree F) are plotted with a regression line representing a linear, first-order model [1]. The problem with this engineering practice is that no additional quantitative information about the scatter or uncertainty of the data is also reported, even though additional analysis methodology exists to yield, for example, 95 % confidence limits as shown in Fig. 2 [1]. This deficiency in data reporting and data compilation in engineering design and materials property data handbooks made it impossible for engineers to estimate the useful life of a component or system with evidence-based quantification of uncertainty and to conduct a subsequent risk analysis for decision-making.