Roc analysis based condition indicator threshold optimization method

Abstract

In condition monitoring systems, condition indicators(CIs) are usually used to describe health state of equipment. Usually, detailed fault isolation is conducted by various diagnoses model, threshold alarming is used to detect exceptions. However, how to set threshold that performs best become a problem. Based on ROC analysis, a condition indicator threshold optimization method is proposed in this paper, optimization object function considers true positive rate and false positive rate, meanwhile, accuracy and precision are included as well; optimization method is validated by experimental signals collected from a planetary test rig.