Fault diagnosis in metallurgical process systems with support vector machines

Fault detection and identification are major challenges in process engineering and manufacturing and the key component of abnormal event management systems. Timely detection, diagnosis and rectification of abnormal or faulty process conditions can lead to savings of billions of dollars in equipment damage and lost productivity, not to mention the prevention of injury and loss of human life associated with industrial accidents. A major contributing factor to current losses in industry is the reliance on human operators to interpret high-frequency samples from hundreds or thousands of variables simultaneously. As a result, the automation of fault detection and diagnosis is seen as crucial to the successful implementation of abnormal event management, the need for which is becoming all the more urgent given the increased complexity associated with modem industrial plants. In this paper, a methodology for process monitoring that uses support vector methods to extract nonlinear features from data is discussed and applied in the diagnostic monitoring of an industrial liquid-liquid extraction column.