Fault Diagnosis in Aluminium Electrolysis Using a Joint Method Based on Kernel Principal Component Analysis and Support Vector Machines

As a key part of aluminium smelting, the operational conditions of aluminium electrolytic cells are of great significance for the stability of the aluminium electrolysis process. As a result, developing a effective process monitoring and multiple fault diagnosis model is essential. Traditional multi-classification methods such as neural networks and multiple support vector machines (multi-SVM) have good effects. However, the connatural limitations of these methods limit the prediction accuracies. To solve this problem, a hierarchical method for multiple fault diagnosis based on kernel principal component analysis (KPCA) and support vector machines (SVM) is proposed in this paper. Firstly, test statistics, such as the comprehensive index \( \phi \), the squared prediction error (SPE), and Hotellings T-squared (\( T^{2} \)), are used for fault detection. To separate faults preliminarily, traditional K-means clustering as transition layer is applied to the principal component scores. Next, anode effect is recognized and classified by the established SVM prediction model. Compared with multi-SVM-based classification methods, the proposed hierarchical method can diagnosis different faults with a higher precision. The prediction accuracy can reach about 90%.