Methodical Approaches

Abstract

Orthogonal arrays with a size to assign almost any number of variables, followed by the procedure to calculate the SNR indices, combined with the ability for efficient pattern recognition opens a way to inverse the formulation. That means, if in a well-known set of data composed of variables and their respective characteristic outcome of objects, the values of the variables will change, then the corresponding new characteristic outcome can be calculated with reasonable accuracy. Consequently, experiments or simulation calculations are unnecessary. The generic term prognosis is appropriate as it is valid regarding almost any number of variables, more or less correlated, digital or continuous, with or without deviation. This inverse application makes use of the SNR indices and defines its numerical values as weight coefficients for prognosticating unknown characteristic outcome values. Incidentally, the number of variables can by far exceed the number of variables. Another approach makes use of the Mahalanobis algorithm that can provide more accurate results in case of nonlinearities. Both procedures are applicable and can be evaluated to select the superior one. Depending on the results, combinations can be used as well to even further improve the accuracy of a prognosis.