OPTIMIZATION OF QUALITY-CONTROL WITH MULTIDIMENSIONAL INSTRUMENTS

The article describes algorithms for obtaining simultaneously the most informative ensemble of signals from a multidimensional spectral inst rument, the calibration of the instrument, and an algorithm for its op eration. It shows that by selecting a pair of closely correlated signa ls the probability of an error in separating two close adjacent concen trations of the inspected product can be reduced by more than six orde rs of magnitude. The essence of the algorithm consists in constructing ellipsoids of correlation of signals with equal probability density f or groups of samples with values of the inspected parameter that are d ifferent for different groups and constant within each group, and with arbitrary varied values of the noninspected influence parameters. The idea of reducing the probability of errors is based on taking into ac count the dimensions and mutual relative positions of the ellipsoids.