Multivariate analysis with covariance matrix applied to separative power modeling of a gas centrifuge

In this work, the least-squares methodology with covariance matrix is applied to determine a data curve fitting to obtain a performance function for the separative power δU of an ultracentrifuge as a function of variables that are experimentally controlled. The experimental data refer to 460 experiments on the ultracentrifugation process for uranium isotope separation. The experimental uncertainties related to these independent variables are considered in the calculation of the experimental separative power values, determining an experimental data input covariance matrix. The process variables, which significantly influence the δU values, are chosen to give information on the ultracentrifuge behaviour when submitted to several levels of feed flow rate F, cut θ and pressure in the product line, Pp. After the model goodness-of-fit validation, a residual analysis is carried out to verify the assumed basis concerning its randomness and independence and mainly the existence of residual heteroscedasticity with any explained regression model variable.