ASIA-PACIFIC JOURNAL OF CHEMICAL ENGINEERING Asia-Pac. J. Chem. Eng. 2007; 2: 355–361 Published online 6 August 2007 in Wiley InterScience (www.interscience.wiley.com) DOI:10.1002/apj.065 Research Article Pseudoresistance entropy as an approach to diagnostics and control in aluminium production Alex J. Hughes, 1 * Mark R. Titchener, 2 John J. J. Chen 1 and Mark P. Taylor 3 1 Department of Chemical & Materials Engineering, University of Auckland, New Zealand 2 Bioengineering Institute, University of Auckland, New Zealand 3 Light Metals Research Centre, University of Auckland, New Zealand Received 12 December 2006; Revised 9 February 2007; Accepted 9 February 2007 ABSTRACT: The advent of deterministic information theory (det-IT), which allows quantification of the T-entropy (similar in a sense to thermodynamic entropy) of time-series data, presents novel insights into the state behaviour of the aluminium reduction cell. The pseudoresistance signal has the potential to expose a number of abnormal/undesired process attributes (excessive magnetohydrodynamic (MHD) oscillation and short-circuiting most importantly), and the identification of these attributes in order to improve cell diagnostics has remained an area of interest for a number of years. Some of the possible applications of process entropy techniques are explored here using specialised visualisation software. Two diagnostic metrics proposed in the literature are compared to analogous, but ultimately preferable entropic properties. Segments of a pseudoresistance trace at a frequency of 1 Hz are analysed in this paper, with recurrent process patterns identified through an elegant state-segregation technique. It is noted that this procedure could be carried out on a continuous basis, allowing early warning of a change in process behaviour. The power and versatility of the graphical interface employed is demonstrated through spectral analysis of the same data streams, leading to further insight into the nature of MHD oscillation in particular. Present limitations on the techniques are described and future opportunities are also discussed.  2007 Curtin University of Technology and John Wiley & Sons, Ltd. KEYWORDS: Hall–H´ eroult; pseudoresistance; entropy; state identification; signal processing INTRODUCTION The unforgiving physical and chemical environment of the aluminium reduction cell necessitates the analysis of ‘indirect’ data sources in the observation of process behaviour. Pseudoresistance, a quantity derived from continuous measurements of the voltage and current across each cell on the smelter floor, provides a means of observing several physical attributes of the reduction process (e.g. magnetohydrodynamic (MHD) oscillation in the metal pad and short-circuiting). In recent times, there has been tentative exploration of the potential to incorporate control action based on the linkage between these physical processes and the ‘noise’ associated with the pseudoresistance signal (Banta et al ., 2002; Banta et al ., 2003; Berezin et al ., 2003; Keniry et al ., 2001). However, implementation to date has been limited to the alteration of the anode beam height depending on overall cell signal amplitude (Bearne, 1998). *Correspondence to : Alex J. Hughes, Department of Chemical & Materials Engineering, University of Auckland, New Zealand. E-mail: ahug030@ec.auckland.ac.nz Encouragingly, some effort has been applied to the identification of process ‘states’, solely through analy- sis of pseudoresistance variation (Banta et al ., 2003; Berezin et al ., 2003). As mentioned above, several physical phenomena define these states and produce characteristic patterns in reduction cell data streams, whose identification on a continuous basis has been tar- geted by Banta et al . through the development of two ‘metrics’ (Banta et al ., 2003). These metrics are first compared in this paper to two novel quantities arising from process entropy considerations. The entropy mea- sures are then applied to actual pseudoresistance data in an attempt to derive distinct states of cell behaviour. DETERMINISTIC INFORMATION THEORY AND PROCESS ENTROPY In an attempt to build upon the more conventional signal processing techniques commonly applied in the aluminium industry, a powerful tool has been proposed to evaluate the absolute complexity of time-series data. An algorithmic technique developed by M. R. Titchener allows calculation of the ‘T-entropy’ of a finite string  2007 Curtin University of Technology and John Wiley & Sons, Ltd.