Inferência de Temperatura de Fornos de Redução de Alumínio Primário por Meio de Sensores Virtuais Neurais

Abstract

Virtual sensors have been used in industries aiming at higher profits with lower costs, since those are softwarebased sensors and, hence, are not subjected to physical damage as real sensors. Virtual sensors can be implanted in hostile environments without compromising the measurements. These successful properties have been made possible due to computational intelligence techniques, which have been widely used in modeling highly complex nonlinear processes. This work evaluates the use of virtual sensors in an important brazilian aluminum industry, whose process is very complex and the temperature measurements are hard to acquire due to the corrosive nature of the material. Specifically, this paper illustrates how a neural-network based virtual sensor performs in inferring the temperature of a furnace for primary aluminum reduction.