An Orthogonal Wavelet Transform-Based K-Nearest Neighbor Algorithm to Detect Faults in Bearings

Table 1

Comparison of OWTKNNs for different feature extraction methods.

Algorithm	Every 100 sets of data OWTKNN classification effect 0HP
EMD algorithm feature extraction	In Figure 11 (EMD), the red∗normal data and the blue×inner circle fault data are divided into two parts in the figure. The boundaries of the two types of data are unclear, cross-distributed, and the classification effect is not obvious. Each consists of over 20 parts, the fault data is misclassified, and the correct rate of diagnosis only reaches 20%.
VMD algorithm feature extraction	In Figure 11 (VMD), the red∗normal data and the blue×inner circle fault data are divided into two parts in the figure. The boundary of the two types of data is unclear, cross-distributed, the two types of data cannot be distinguished, and the diagnostic accuracy rate is only about 50%.
K-centre clustering algorithm (KCA)	In Figure 12 (KCA), the red ∗ normal data and the blue × inner circle fault data are divided into two parts in the figure. The boundary of the two types of data is unclear, the cross distribution, and the classification effect is inconspicuous, each with more than 20 parts the fault data is misclassified, and the diagnosis accuracy rate is only about 80%.
Rms extraction	In Figure 12 (rms), the red∗normal data and the blue∗inner circle fault data are divided into two parts in the figure. The boundary of the two types of data is unclear, and the boundary is cross-distributed, and the classification effect is not obvious, each with 10 multiple fault data are misclassified, and the diagnosis accuracy rate is only about 90%.
Kurtosis
Extraction	In Figure 12 (kurtosis), the red∗normal data and the blue×inner circle fault data are divided into two parts in the figure. The boundary of the two types of data is not very clear, cross-distributed. The two types of data cannot be distinguished, and the diagnostic accuracy is only around 30%.
Impulse factors feature extraction	In Figure 12 (impulse factor), the red∗normal data and the blue∗inner circle fault data are divided into two parts in the figure. The boundary of the two types of data is unclear, cross-distributed, the two types of data cannot be distinguished, and the diagnostic accuracy rate only reaches about 30%.