The Pursuit of Causality in Multivariate Statistical Control of Aluminium Smelting Cells

Abstract

Aluminium smelting cells are operated at the limit of their capability by the adaptation of increasingly higher line current and the requirement of maximum flexibility in energy input and raw materials used. Poor controllability and observability limit both energy efficiency and productivity of pot lines under these varying inputs. Based on several key hypotheses which are developed and tested here, a new control approach is demonstrated for smelting cells, to tackle the cell variability over a wide range of time scales and in a way which reduces this variation over time. In the present thesis a novel control scheme has been tested on 10 industrial cells. These results are compared to the period before the test and to a group of 10 reference cells. The control philosophy is based on the determination of cell-specific natural behaviour envelops using Hotelling T2 statistic and cause-specific detection of abnormalities. A sludge cycle is identified as one of the main self-accelerating destabilising mechanisms in smelting cells driving substantial variations in liquid bath mass up to 50%. This overrides the self-regulating side wall ledge mechanism, and moves the liquid bath mass and chemistry out of the optimum operating range. The natural behaviour envelops are based on a moving window of the interchanging of the bath temperature, liquidus point and the rate of alumina accumulation to allow maximum flexibility. The size and shape of these ellipsoidal envelops are characteristic of the presence or absence of bath mass variation through the above sludge cycle. These envelops allow multivariate detection of abnormalities and this detection is combined with online Pareto and root cause analyses. This cell control scheme has been demonstrated to improve the quality of decision-making. Identification of causes of abnormality and their correction or removal through this system allows continuous improvement on a daily basis. For the test cells, the employment of the new control philosophy presented in this thesis has resulted in a higher current efficiency (+2.0%, p=0.075) and significantly lower energy consumption (-0.42 kWh/kg, p=0.034) for 2008 compared to 2007. The new control scheme has been implemented for all cells on both pot lines at Aldel in 2009. In the future, further control system improvements to alumina feeding, along with extension of the multivariate control approach and incorporation of human decision guidance will have even greater implications for the flexibility of smelter operation under the constrained global electricity market conditions prevailing.