Instrumental errors in microsegregation measurements in multicomponent alloys complicate the treatment of randomly sampled data. In this article, two new alloy-independent data treatment algorithms are presented that are capable of separating the effects of scatter in the data from the underlying segregation trends, assigning each measurement location a unique fraction solid. These new methods are physically reasonable and result in improved estimates of segregation parameters, particularly the solute partitioning at the dendrite tip. This is demonstrated by determining the microsegregation in four successive generations of single-crystal (SX) nickel-based superalloys. Artificial, noise-induced, tails commonly seen in the microsegregation profiles are also minimized. A methodology for evaluating sorting schemes is introduced that does not depend upon a priori knowledge of the partitioning direction. Comparison is made to both other sorting methods and CALPHAD predicted partition coefficients. Implications for alloy design are reported, illustrating the interaction between solute species such as Ru, Re, Co, and W.

The random sampling approach offers an elegant yet accurate way of validating microsegregation models. However, both instrumental errors and interference from secondary phases complicate the treatment of randomly sampled microprobe data. This study demonstrates that the normal procedure of sorting the data for each element independently can lead to inaccurate estimation of segregation profiles within multicomponent, multiphase, aluminum alloys. A recently proposed alloy-independent approach is shown to more reliably isolate these interferences, allowing more accurate validation of microsegregation models. Application of this approach to examine solidification segregation of a 319-type alloy demonstrated that, for these slowly cooled castings, neither Sr or TiB 2 additions significantly affected coring of Cu within the primary a-Al dendrites. Comparison against predictions of CALPHAD-type Gulliver-Scheil models was less satisfactory. Consideration of back-diffusion and morphology effects through a one-dimensional (1-D) numerical model do not improve the agreement. Possible reasons for the lack of agreement are hypothesized.