Modelling of temperature in the aluminium smelting process using Neural Networks

Industries are aiming to become more competitive and enlarge their profits. A good management is a key factor to accomplish the company's target, however all management decisions are supported by tools that provide good information on the process. Soft Sensors have been applied in industries and its use has been growing lately. It can be adapted to any application regarding variable measurement, therefore reducing operational costs without compromising process information. In some cases, better results can be obtained. The key of its success is the intelligent computing it uses, which has been heavily used in nonlinear and highly complex process modeling. This work exploits its use with Neural Networks in a chemical process in an important Brazilian Aluminum Smelter whose process is very complex and whose measurements consume operational resources due to corrosive nature of the plant. The usage of soft sensors may reduce costs and measures' delays drastically. A case of use of the soft sensor for temperature measure is presented on this work, its design through implementation, according to a researched methodology.