Bearing Fault Detection in Varying Operational Conditions Based on Empirical Mode Decomposition and Random Forest

Roller bearings play a significant role in kinds of machine. In most cases, it won't work in steadily operational conditions. The paper proposed a method which combines empirical mode decomposition and auto-regressive model to extract features of faults in various operational conditions and uses random forests to set an effective pattern recognition model. In addition, the paper compares the result of random forests with that of some other classification method. The bearing vibration data comes from Case Western Reserve University Bearing Data Center. The result indicates that the method is effective and can be used in actual situations.