Rolling Bearing Diagnosis Based on Composite Multiscale Weighted Permutation Entropy.

Area of Science:

• Nonlinear dynamics
• Signal processing
• Machine learning for fault diagnosis

Background:

• Rolling bearing faults present complex nonlinear dynamic characteristics.
• Accurate fault diagnosis is crucial for industrial machinery reliability.
• Existing methods may struggle with the intricate nature of bearing fault signals.

Purpose of the Study:

• To propose a novel method for evaluating nonlinear time series complexity.
• To develop an effective rolling bearing fault diagnosis approach using the proposed complexity measure.
• To validate the proposed approach on experimental fault data.

Main Methods:

• Composite multiscale weighted permutation entropy (CMWPE) was developed to quantify time series complexity.
• A fault diagnosis framework integrating CMWPE, joint mutual information (JMI) for feature selection, and k-nearest-neighbor (KNN) classification was established.
• CMWPE extracts features, JMI selects sensitive ones, and KNN classifies bearing conditions.

Main Results:

• The CMWPE method demonstrated its advantage in analyzing simulated signals.
• The CMWPE-JMI-KNN approach successfully identified different rolling bearing conditions in experimental datasets.
• The proposed method effectively captures the complex nonlinear dynamics of fault signals.

Conclusions:

• The CMWPE metric provides a robust measure for nonlinear time series complexity.
• The integrated CMWPE-JMI-KNN approach offers an effective solution for rolling bearing fault diagnosis.
• This methodology holds promise for enhancing predictive maintenance in rotating machinery.