Affiliations: Key Laboratory of Mechanism Theory and Equipment
Design of Ministry of Education, Tianjin University, Tianjin, China
Note: [] Corresponding author: Guofeng Wang, Key Laboratory of Mechanism
Theory and Equipment Design of Ministry of Education, Tianjin University,
Tianjin 300072, China. Tel.: +86 022 2740 6951; Fax: +86 022 2740 6260; E-mail: [email protected]
Abstract: Condition monitoring of rolling element bearing is paramount for
predicting the lifetime and performing effective maintenance of the mechanical
equipment. To overcome the drawbacks of the hidden Markov model (HMM) and
improve the diagnosis accuracy, conditional random field (CRF) model based
classifier is proposed. In this model, the feature vectors sequences and the
fault categories are linked by an undirected graphical model in which their
relationship is represented by a global conditional probability distribution.
In comparison with the HMM, the main advantage of the CRF model is that it can
depict the temporal dynamic information between the observation sequences and
state sequences without assuming the independence of the input feature vectors.
Therefore, the interrelationship between the adjacent observation vectors can
also be depicted and integrated into the model, which makes the classifier more
robust and accurate than the HMM. To evaluate the effectiveness of the proposed
method, four kinds of bearing vibration signals which correspond to normal,
inner race pit, outer race pit and roller pit respectively are collected from
the test rig. And the CRF and HMM models are built respectively to perform
fault classification by taking the sub band energy features of wavelet packet
decomposition (WPD) as the observation sequences. Moreover, K-fold cross
validation method is adopted to improve the evaluation accuracy of the
classifier. The analysis and comparison under different fold times show that
the accuracy rate of classification using the CRF model is higher than the HMM.
This method brings some new lights on the accurate classification of the bearing faults.
Keywords: Fault classification, conditional random field, rolling element bearing, hidden Markov model
