RESUMO
The question of the statistical accuracy of EBSD data for global texture calculation was re-explored on the basis of a very large grain population (83 000 grains measured on a recrystallized low-alloyed Zr sheet). Previous works aimed mainly at identifying and quantifying the main texture components and were based on much smaller data sets. The present work attempts to quantify the accuracy of the complete texture, including low-density regions of the orientation space. For that purpose, a new statistical parameter, V(Delta), based on the calculation of texture difference functions is proposed. This parameter has two main advantages: it is equally sensitive to both high and low peaks of the orientation density function (ODF), and it has a physical interpretation because it is the material volume fraction corresponding to the difference between a given ODF and a reference ODF (considered, or known to be close to the truth). Two main variables were studied: the number of grains taken into account and the peak width phi(0) of Bunge's 'Gaussian' model density used as kernel for the actual analysis. The orientation distribution functions were computed by nonparametric kernel density estimation with harmonics up to the order of 34. Minimizing the value of V(Delta) serves as the objective function for optimizing the peak width phi(0) as a function of the number of grains. The properties of the V(Delta) parameter also allows for the definition of a method for estimating the accuracy of a given texture that has been obtained from a limited number of grains, without knowing the true texture of the investigated material.
