Fault Diagnosis of Bearings with Adjusted Vibration Spectrum Images

Table 10

Comparisons for multiclass fault detection between current work and some published work.

Method in refs.	Training data	Testing data	Classification accuracy (%)
RPS and GMM in [18]	Load 1 (3840)	Load 1 (3840)	99.95^a
DNN-based method in [9]	Load 1 (1000)	Load 1 (1000)	99.95
	Load 2 (1000)	Load 2 (1000)	99.61
	Load 3 (1000)	Load 3 (1000)	99.74
	Load 1–3 (3000)	Load 1–3 (3000)	99.68
LWPT and binary tree system in [19]	Load 0 (400)	Load 0 (200)	99.53
MFE and MSVM in [20]	Load 0 (840)	Load 0 (360)	94.50
TR-LDA2 and kNN classifier in [21]	Load 1 (200)	Load 1 (600)	98.00
	Load 2 (150)	Load 2 (650)	97.65
Multiple ANFIS combination in [22]	Load 0–3 (300)	Load 0–3 (300)	91.33
MKMFA and kNN in [23]	Load 0–3 (500)	Load 0–3 (500)	97.45
Dataset G2 in present work	Load 1 (100)	Load 1 (400)	100
Dataset G3 in present work	Load 2 (100)	Load 2 (400)	100
Dataset G4 in present work	Load 3 (100)	Load 3 (400)	100
Dataset G1 in present work	Load 0 (100)	Load 0–3 (1900)	99.13
Dataset G2 in present work	Load 1 (100)	Load 0–3 (1900)	99.99
Dataset G3 in present work	Load 2 (100)	Load 0–3 (1900)	98.07
Dataset G4 in present work	Load 3 (100)	Load 0–3 (1900)	95.72
Dataset G5 in present work	Load 0–3 (400)	Load 0–3 (1600)	100
Dataset G1 with nonadjusted spectrum^b	Load 0 (100)	Load 0–3 (1900)	84.97
Dataset G2 with nonadjusted spectrum^b	Load 1 (100)	Load 0–3 (1900)	92.30
Dataset G3 with nonadjusted spectrum^b	Load 2 (100)	Load 0–3 (1900)	89.66
Dataset G4 with nonadjusted spectrum^b	Load 3 (100)	Load 0–3 (1900)	90.72
Dataset G5 with nonadjusted spectrum^b	Load 0–3 (400)	Load 0–3 (1600)	100

^aThis classification accuracy is computed based on Table 2 in [18], which is the average of the ten classes. ^bThe test settings are same with that in G1–G5 using the adjusted spectrum images.