Resumo:
The quality in the manufacturing processes generally depends on the attainment of a great deal of characteristics, which sometimes are correlated. An efficient transformation of these attributes probably can be done through a reduction of dimensionality mechanism. When represented by principal components, the global indexes formed can not be enough, not representative or may present conflicts with the optimization sense of original responses, which difficulty its implementation and interpretation. In this work, it was studied the improvement of the principal components method applied to Response Surface designs, using the concept of multivariate indexes, a singular and representative measure of original set of correlated responses. In attempt to asses the method efficiency, several experimental designs involving welding and machining processes were used. Each case highlighted some particular issue of the whole method. When compared with other traditional optimization methods, this approach presented satisfactory results, what means the proposal is adequate and feasible to optimization of multiple responses.