An Intelligent Fault Severity Diagnosis Method based on Hybrid Ordinal Classification

Abstract

Feature selection and classification are widely used in fault diagnosis. In fault severity diagnosis, different fault severity can be expressed as normal, slight, moderate and severe. And some fault features monotonically change with the severity. How to use the potential ordinal information in the data is the key to the accuracy of diagnosis. Considering the existence of ordinal information, this paper presents an intelligent method for fault severity diagnosis based on ordinal classification. Furthermore, this paper discusses the method of optimal division of monotonic and non-monotonic features. First, we apply a feature selection algorithm to obtain an optimal feature subset. Then, we train a diagnosis model by using the partially monotonic decision tree (PMDT). Then, the proposed model is applied to classify different crack levels of the gear. Experimental results demonstrate that the proposed diagnosis method can reduce the feature size and improve performance of fault severity diagnosis.