Fábio M. Soares

Estimation of aluminium fluoride concentration in aluminium reduction cells through a soft sensors approach

This work exploits a model for Aluminium Fluoride Concentration Measurement in the Aluminium Smelting process. This process variable is usually measured every 50-100 hours since it requires long laboratory analysis. This variable has a strong influence on the whole process, thus it should be controlled in a shorter basis. In order to prevent the long time between measurements, we developed a soft sensor based on neural networks which allows estimating the fluoride concentration at any moment by querying against the available process database. This database encompasses the consolidated knowledge on this chemical process and thus can be used both to build and validate the model.

Soft sensor for fluoridated alumina inference in gas treatment centers

The Gas Treatment Center performs a key role in the aluminum smelting process, since it strongly influences the chemical and thermal stability of the electrolytic bath through fluoridated alumina. Therefore this variable should be considered to keep the bath chemistry under control. However, the fluorine concentration measurement in fluoridated alumina is very time-consuming and that information becomes available only after a while. By using Artificial Neural Network we developed a Soft Sensor capable to estimate the fluorine concentration in fluoridated alumina, and to provide that information to plant engineers in a timely manner. This paper discusses the methodology used and the results of an implemented Soft Sensor using Neural Networks on fluorine estimation in fluoridated alumina from a Gas Treatment Center in an important Brazilian Aluminum Smelter.

Simulations on The Bath Chemistry Variables Using Neural Networks

In this work we present a single layer neural network based model for bath chemistry variables in the aluminum smelting process. This model is designed to be simulated with real data as if it worked online in parallel with the process. The model is built using a very fast machine learning algorithm, the Extreme Learning Machines, which provides excellent results in regression problems in a very short time. Also we applied statistical analysis for data collection, preprocessing and filtering and for validation we performed several simulations to attest the neural model’s capability to respond to new data. A comparison of this model against linear and traditional nonlinear structures is performed to show how single layer neural networks can be applied on the bath chemistry modeling.

Applied Statistical Analysis on the Calcination Process in the Ferronickel Production

Ferronickel is mainly produced by the RKEF (Rotary Kiln Electric Furnace) process. The ore is extracted, crushed and dried before calcination. Rotary kilns are usually about 100 m long and rotate to facilitate material flow. Due to this length, in addition to rotation speed, the material takes a variable time to cross the whole kiln, and thereby changing the chemical and temperature profile, making the control of calcine temperature a great challenge. However, statistical analysis are a great tool for finding interesting patterns that are of valuable help to control the kiln variables, that are very sensitive to inertia caused by rotation and changes in temperature profile. We present a study based on real data taken from a processing plant, whereby we applied data mining techniques to extract information on which variables have influence on kiln’s key performance index variables, such as calcine temperature.

Clustering Aluminum Smelting Potlines Using Fuzzy C-Means and K-Means Algorithms

Aluminum smelting potlines usually have a big number of cells, producing aluminum in a continuous and complex process. Analytical monitoring is essential to increase the industries’ competitive advantage, however, during their operation, some cells share similar behaviors, therefore forming clusters of cells. These clusters rely on data patterns that are usually implicit or invisible to operation, but can be found by means data analysis. In this work we present two clustering techniques (Fuzzy C-Means and K-Means) to find and cluster the cells that present similar behaviors. The benefits of clustering are mainly in the simplification of potline analysis, since a large number of cells can be summarized in one single cluster, which can provide richer but compacted information for control and modelling.

Fuzzy Control Applied to Aluminum Smelting

Aluminum is a modern and new metal, since it has been produced for industry no earlier than 1886, when Hall and Héroult concurrently found out a method to produce free Aluminum through electrolysis (Beck, 2008). In 1900, the Aluminum production worldwide had reached a thousand tons. Nevertheless, at the beginning of the 21st century, global production reached 32 million tons encompassed by 24 million of primary Aluminum and 8 million of recycled material. This fact puts Aluminum at the second place in the list of the most used metals on earth. The world without Aluminum became inacceptable: the businessmen, the tourists, the delivery offices fly over the world in airplanes made of Aluminum, as well as many enterprises and industries are strongly dependent of this metal. Figure 1 shows in a widely perspective where Aluminum is most used.

Sensor Virtual Neural para Estimação de Flúor em Alumina no Processo de Fabricação de Alumínio Primário

AbstractThe Gas Treatment Center of an important Brazilian primary aluminum industry performs fundamental role in the aluminum smelting process. The main role of this plant is to produce fluoridated alumina that is used to maintain the chemical and thermal stability of electrolytic bath. Through a Computational Intelligence technique known as Artificial Neural Network, we found a Soft Sensor able to estimate the fluorine concentration in fluoridated alumina. Soft Sensors are software based, thus, its measurement is far easier than the real measurement (by laboratory) of the target variable. In this context, this paper presents how a soft sensor, based in neural network, operates in estimation of fluorine in fluoridated alumina from Gas Treatment Center. To certify the generalization capacity of the proposed soft sensor, we used data from another Gas Treatment Center, performing estimated result consistent with the target data.