Page 293 - Pipelines and Risers
P. 293

266                                                              Chapter I5


           The following uncertainties are introduced (Bai and Song (1997)):

           Model  uncertainty, XM. Model  uncertainty  is  introduced  for  the  criteria  to  account  for
           modeling and methodology uncertainties. It reflects a general confidence in the design criteria
           for a real life in-situ scenario. The model uncertainty is calibrated from the test results listed
           below. A normal distribution is applied to fit this uncertainty.


           Uncertainty for pressure,  Xp.  The  characteristic  value  of  the  normalized  pressure  Xp  is
           obtained by substituting safety factors, characteristic values of  the other parameters into the
           design  equation.  In  general,  the  annual  maximum  operating  pressure  is  higher  than  the
           nominal operating pressure. This is reflected by the mean bias in Xp. A Gumbel distribution is
           used.


           Uncertaintyforflow stress, Xf. The Xf mainly reflects the material property. Uncertainty of Xf
           is largely dependent of the material grade. A log-normal distribution is assumed to fit the data
           in the existing database.


           Uncertainty for  dent  depth,  XD. The  uncertainty  in  the  dent  depth  is  associated  with
           inspection. A normal distribution is assumed for XD based on judgement.


           Uncertainty for crack length, XL.  It is similar to the discussion of XD. Normal distribution is
           used for XL.


           Uncertainty for geometry function, XY. Considering the uncertainties in  geometry function
           estimation, a log-normal distribution is applied for XY.


           Uncertainty for pipe  wall-thickness, Xt. The uncertainty in pipe wall-thickness is considered
           by bias Xt following a normal distribution.


           The statistical values for the above biases are given in Table 15.2 as below.
           15.4.3  Reliability Analysis Methods
           Generally, LSF is introduced and denoted by g(Z). Failure occurs when g(Z)SO.  For a given
           LSF g(Z), the probability of failure is defined as:
                PF(t)  = p[g(z)   01                                          (15.34)


           The results can also be expressed in terms of a reliability index p, which is uniquely related to
           the failure probability by:
                P(t) = -W(PF(t)) = a-q- P F ( t))                             (15.35)


           where:
   288   289   290   291   292   293   294   295   296   297   298