Combining unsupervised and supervised fault diagnosis method with ensemble learning

This paper analyzes the limitations of the traditional supervised and unsupervised fault diagnosis methods and indicates combining both supervised and unsupervised methods is a strategy to cope with the limitations. However, the fusion models have a common defect that they will inherit all of unsupervised and supervised method false alarms. To deal with the issue, we proposed a ensemble-based method named ensemble fusion models (EFM). EFM has two advantages: i) it diversifies the modeling samples to get diverse classifier, ii) it use diverse fusion models to get general results. The methodology presented was assessed on the Tennessee Eastman Process (TEP) benchmark. These results demonstrate the validity and superiority of this method and the promising potential for the diagnosis of industrial applications.