Multivariate statistical inference in a radial basis function neural network

Abstract

Regression models and analysis of variance are widely used methods for analyzing manufacturing processes in the industry. Its objective is to analyze the effect of process input variables on a final quality characteristic. However, some processes have complex data that cannot be described by linear regression or have several product quality characteristics to be controlled, having several correlated responses makes it difficult to analyze and optimize the process, thus, it is common to ignore the correlation between responses and make several independent models for each response, causing problems for decision-making based on these models. This paper shows the application of multivariate statistical analysis in a Radial Basis Function neural network, considering the statistical significance between independent and dependent variables and the correlation and verifying if the assumptions for this analysis are fulfilled. The results evidence that the multivariate statistical analysis in the Radial Basis Function is a good method to analyze the process based on correlated variables: it satisfies the assumptions required for this analysis, it determines the process variation, and it also determines the most important variable that had influence in the process used to evaluate the permanent mold casting process.