Mejora de la calidad de procesos con redes neuronales, algoritmos genéticos y lógica difusa Mejora de la calidad de procesos con redes neuronales, algoritmos genéticos y lógica difusa

Abstract

In many cases, when trying to improve the quality of products or processes, it is necessary to simultaneously optimize multiple responses. A classic approach is to apply the Design of Experiments (DOE), multiple regression models to estimate the relationship between the responses and the controllable factors; then, combine the different responses with a desirability function and finally the controllable factors are optimized. However, it may happen that the relationship between controllable factors and responses is too complex to estimate the relationship with these methodologies; for example, a highly nonlinear relationship. A proposed alternative approach is the use of artificial neural networks (ANN) to estimate response functions; in the event of having qualitative variables, they are processed with fuzzy logic (FL), and in the optimization phase genetic algorithms (GA) are used. An example of optimizing a process of multiple responses is presented to validate this proposal.