Incorporating Heterogeneous Features into the Random Subspace Method for Bearing Fault Diagnosis

In bearing fault diagnosis, machine learning methods have been proven effective on the basis of heterogeneous features extracted from multiple domains, including deep representation features. However, comparatively little research has performed fusing these multi-domain while dealing with interrelation and redundant problems to precisely discover faults. Thus, in current study, a novel diagnostic method, namely method incorporating representative into random subspace, or IHF-RS, is proposed for accurate diagnosis. Primarily, via signal processing methods, statistical are extracted, stack autoencoder (DSAE), acquired. Next, considering different levels predictive power features, modified lasso subspace introduced measure produce better base classifiers. Finally, majority voting strategy applied aggregate outputs various classifiers enhance performance fault. For method’s validity, two datasets provided by Case Western Reserve University Bearing Data Center Paderborn were utilized experiments. The results experiment revealed that IHF-RS can be successfully utilized.