Multivariable model-based control of the non-alumina electrolyte variables in Aluminium smelting cells

Abstract

The focus of this thesis is on improving the regulatory control of the electrolyte variables of a reduction cell. Bath temperature, aluminium fluoride concentration (%excess AlF3), liquidus temperature and superheat are termed the electrolyte variables in this thesis. Also note that the commonly used term bath is used in this thesis to refer to the molten electrolyte in the cell. The electrolyte variables are important to control because of the impact they have on aluminium smelter profitability. his thesis investigated the potential to improve the control of the electrolyte variables through two areas: 1. Implementation of multivariable model-based control. The rationale was that control performance might be improved by acknowledging the likely process interactions; currently, control systems are typically a collection of single-input/single-output loops. Despite a limited number of published attempts to apply multivariable model-based control, no one had adequately determined whether it is necessary or beneficial. In this thesis, the application of Linear Quadratic Gaussian control (LQG control) was specifically investigated. While it is Model-based Predictive Control (MPC) that is of particular long-tenn interest to this project (due to its performance and widespread use in other industries), LQG control is equivalent to MPC without constraints. 2. Implementation of recently developed sensors, for use by a multivariable model-based controller. The rationale was that a possible barrier to effective process control might be the limited number of measurements that either can be or are routinely performed. Specifically, this thesis investigated the potential use of alumina concentration and individual cell duct flowrate, temperature and heat loss measurements, as feedforward variables. The value of these measurements for control of the electrolyte variables had previously not been determined. As a component of these investigations, this thesis also determined the appropriate control structure for control of the electrolyte variables (i.e. the controlled and manipulated variables and their interconnection; interconnection required with control of bath height, metal height and alumina concentration; and required feedforward variables), as this could not be completed based on theory and literature contributions. In addition, procedures for new and existing measurements were developed, with the aim of minimising control-relevant measurement errors (as the literature identifies several factors that, if not accounted for, can increase the level of control-relevant measurement errors), and then the likely impact of measurement errors on controller performance was assessed. New sensors were implemented and procedures developed for these and for existing measurements. A systems identification experiment was designed and carried out at a smelter, on modem reduction cells (i.e. high amperage, point-fed and magnetically compensated technology). From the data, dynamic predictive models were identified. Using these models, a control structure analysis was conducted and LQG controllers were designed for a number of the candidate control structures. The likely performance, stability and robustness of these (to model/plant mismatch, loop gain variation or failure and opening/closing of loops) were assessed using simulations and other analyses. From these results, the benefits, limitations and risks of the LQG control options were defined by comparing them with the existing controllers. The issue of appropriate control structure was further considered by comparing the various LQG control options. Where appropriate, other experimental measurements and data analyses were performed to support the main work of this thesis. The overall conclusion of this thesis is that the implementation of multivariable model-based control at the smelter in this study, should improve the control of the electrolyte variables (bath temperature, liquidus temperature, %excess AlF3 and superheat). In addition, the control of bath height should also be improved. As the smelter in this study uses control systems typical of the industry, these conclusions could potentially apply to other smelters as well.