Diagnosing faults in aluminium processing by using multivariate statistical approaches

Abstract

Multivariate statistical approaches are expected to detect and diagnose faults effectively for complex materials processing including steel, iron, copper and aluminium processing. This advanced monitoring of materials processing is essential to address abnormal or faulty conditions in a timely manner. In the aluminium-smelting process, late diagnosis of abnormal conditions such as an anode effect can result in an increase of energy consumption and emission of greenhouse gases. In this article, a new statistical framework is proposed that is based on hierarchal diagnostic approach to diagnose two groups of faults, anode faults and non-anode faults. The system diagnoses faults by predicting the type of fault based on continuous, non-linear and multivariate process data using discriminant partial least squares (DPLS). The new system goes beyond the typical multivariate system in that it also includes the dynamic behaviour of the process during anode changing and alumina feeding. The results of performance evaluation of the new diagnosis system tested using real-data show that the system can diagnose the two groups of faults.