Aluminium process fault detection by Multiway Principal Component Analysis

Real-time fault detection is difficult to perform in an aluminium smelter because the continuous aluminium electrolysis is operated batchwise in terms of material additions, meaning the measurements obtained from the process are dynamic, multivariate and limited. This paper presents a new framework based on Multiway Principal Component Analysis (MPCA) to detect faults in real-time in the industrial continuous aluminium electrolysis process. This real-time fault detection system incorporates the dynamic behaviour of two important operations in the continuous aluminium electrolysis process, alumina feeding and anode changing. The methodology is demonstrated using real data from an operating aluminium smelter, and is shown to be effective in the early detection of anode spikes and anode effects.

Research highlights
► Detection of anode spikes and anode effects in a real aluminium smelter.
► An alumina feeding cycle was treated as a batch operation using MPCA.
► Monitoring numerous cells using the same reference model.