Predictive control of aluminum electrolytic cells using neural networks

  • PDF / 507,112 Bytes
  • 13 Pages / 612 x 792 pts (letter) Page_size
  • 105 Downloads / 195 Views

DOWNLOAD

REPORT


I.

THE ALUMINUM ELECTROLYSIS PROCESS

A brief description of the aluminum electrolytic process has been presented in a recently published article.[1] The description placed emphasis on the process control aspect, as the purpose of the article was to elaborate on the use of neural networks to recognize the operational state of the electrolytic cell in order to obtain a closed-loop control scheme. In the present work, which is a continuation of the previous work, other neural networks are used to predict the future values of the decision variables, namely, the resistance variations, in order to prevent the anode effect, an undesirable destabilizing event. For this reason, a brief description of the process is given again here, this time with emphasis on anode effects and the possible ways to counter them. The electrolytic process consists of dissolving alumina (Al2O3) in molten cryolite (Na3AlF6), commonly called the bath, and decomposing it electrolytically to obtain liquid aluminum. A high-intensity, low-voltage, constant electric current passes through the electrolytic cell from the carbon anode to the bath, then on to the carbon cathode. The latter is built in the form of a rectangular box to facilitate the gathering of the liquid aluminum produced. The oxygen freed by the electrolysis is drawn to the anode, which in the process is gradually consumed to produce carbon dioxide (CO2). The consumable anode is a typical feature of the process, known as the Hall–He´roult process, named after its two well-known early inventors. The overall reaction is given by A. MEGHLAOUI, formerly Postdoctoral Fellow, Department of Applied Science, University of Quebec at Chicoutimi, is Development Engineer, Dubai Aluminum P0B 3627 Dubai, United Arab Emirates (UAE). R.T. BUI and L. TIKASZ, Professors, are with the Department of Applied Science, University of Quebec at Chicoutimi, Chicoutimi, PQ, Canada G7H 2B1. J. THIBAULT, Professor, is with the Department of Chemical Engineering, Laval University, Quebec, PQ, Canada G1K 7P4. R. SANTERRE, Consultant, is with Alcan International Ltd., Jonquiere, PQ, Canada G7S 4K8. Manuscript submitted September 12, 1997. METALLURGICAL AND MATERIALS TRANSACTIONS B

2Al2O3 1 3C → 4Al 1 3CO2

[1]

In addition to cryolite, the bath usually contains various additives, mainly aluminum fluoride (AlF3) and calcium fluoride (CaF2), the purpose of which is to improve the physicochemical properties of the bath and lower its melting temperature. Typically, the current used in cells is on the order of 50 to 300 kA. Theoretically, 1 kA produces 8.052 kg of aluminum per 24 hours. As no real process can be 100 pct efficient, the efficiency of the aluminum electrolysis is quantified as a percentage of that theoretical productivity and it is called the current efficiency. The reason for the current efficiency to be less than 100 pct is mainly found in the ‘‘back reaction’’ where part of the aluminum formed by electrolytic reduction is reoxydized, 2Al 1 3CO2 → Al2O3 1 3CO

[2]

Alumina is the raw material of the proces