DETECTING ABNORMALITIES IN ALUMINIUM REDUCTION CELLS BASED ON PROCESS EVENTS USING MULTI-WAY PRINCIPAL COMPONENT ANALYSIS (MPCA)

摘要

In the aluminium industry optimal production and quality products are major process targets. One way to achieve these targets is by improving the process control of aluminium reduction cells, and this is the aim of this research. This research proposes to apply an advanced multivariate control chart to aluminium reduction cells in a manner which provides new insights into process abnormalities and their diagnosis. The proposed approach uses multi-way principal component analysis to observe the movement of data towards abnormality after process events. Preliminary results showed that using the proposed approach could detect anode spikes after anode changing or tapping. Data with anode spikes present moved in a different direction than the data with anode spikes absent. An anode spike trajectory could be set up based on this discrimination. Data which move towards the anode spike trajectory have a high possibility of having anode spikes. Therefore based on this trajectory, the cell could be monitored ahead of time for spikes, and operations may take action to search for them much earlier. This will lead to a real-time fault detection system and is expected to assist process engineers in improving the process control of aluminium reduction cells.