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

Table 2

Comparison of OWTKNNs for multiple fault signals.

Fault 1	Fault 2	OWTKNN classification effect of every 100 sets of data for the 0 HP case	OWTKNN multiclassification effect for every 10 datasets for the 1 HP case
The bearing was normal	Inner ring 0.007 in.	In Figure 9, the red ∗ and blue × for the normal data from the bearing and fault data of the inner ring are smoothly divided into two parts. The boundary of the two types of data is clear, with obvious classification effects and a correct diagnosis rate of 100%.	In Figure 9, for the two sets of classified data, the normal bearing and inner ring fault represented by red ○ and blue □, respectively, can be correctly identified with a recognition rate of 100%. This means that the fault category can be effectively recognized.
Inner ring 0.007 in.	Ball 0.007 in.	In Figure 13, the red ∗ and blue × for the fault data of the inner ring and ball are well divided into two parts. The boundary of the two types of fault data is clear except for 2 data points in close proximity, with obvious classification effects and a correct diagnosis rate exceeding 99%.	In Figure 13, for the two sets of data to be classified, the faults of the inner ring and the ball represented by red ○ and blue □, respectively, can be completely correctly identified with a recognition rate of 100%. This means that the fault category can be well recognized.
Inner ring 0.007 in.	Ball 0.014 in.	In Figure 14, the red ∗ and blue × for the fault data of the inner ring and ball are well divided into two parts. The boundary of the two types of fault data is clear except that 2 blue × ball fault data points are improperly classified. The classification effect is obvious, and the correct diagnosis rate exceeds 98%.	In Figure 14, to classify the two sets of data to, the faults of the inner ring and the ball represented by red ○ and blue □, respectively, can be correctly identified with a recognition rate of 100%. This means that the fault category can be recognized quite well.
Inner ring 0.007 in.	Inner ring 0.014 in	In Figure 15, the red ∗ and blue × for the fault data of the inner ring and ball are divided into two parts quite well. For the data points, except for 2 red ∗ inner circle faults (0.007 in.) and 5 blue × inner circle faults (0.014 in.) which are improperly classified, the other data boundaries are clearly defined, the classification effect is obvious, and the correct diagnosis rate over 95%.	In Figure 15, to classify the two sets of data to, the fault (0.007 in.) and the fault (0.014 in.) of the inner ring represented by red ○ and blue □, respectively, can be correctly identified with a recognition rate of 100%. This means that the fault category can be effectively recognized.