Incremental fault diagnosis: Exploiting unlabelled data with semi-supervised ensemble learning

Abstract

Fault diagnosis is gaining interest both in academic and industry fields, which assures machinery operational safety and reliability in terms of electrical equipments such as servo actuator systems. With a view to design a desirable diagnosis method that can gain better generalization ability of fault diagnosis with limited labeled data, a novel fault diagnosis method that utilizes tri-training architecture based semi-supervised ensemble learning is proposed, where three base learners are established and trained iteratively to reinforce the learning process by exploiting unlabeled data plus few labeled data. Besides, given consideration to both accuracy and diversity, a data editing technique for unlabeled samples is used in this study for the purpose of augmenting the differentiation of the base classifiers, where kernel principle component analysis (KPCA) maps the self learnt samples into several label vectors to further select the subsets with greater diversity instead of all for the incremental training. The proposed method aims to facilitate fault identification ability adaptively by taking advantage of unlabeled samples, which is appropriate for dealing with the diagnosis issues that only limited number of labeled data exist. Comparative experiments are included in this paper to demonstrate the effectiveness in fault diagnosis of servo actuator systems.