Unsupervised domain adaptive bearing fault diagnosis based on maximum domain discrepancy

EURASIP Journal on Advances in Signal Processing

Table 2 Fault classification results of different methods on the bearing datasets of CWRU (accuracy %)

S → T	TCA	Deep CORAL	DDC	DAN	DCTLN	Proposed method
0 → 1	62.50	98.11	98.24	99.38	99.99	99.74
0 → 2	65.54	83.35	80.25	90.04	99.99	*100*
0 → 3	74.49	75.58	74.17	91.48	93.38	*100*
1 → 0	63.63	90.04	88.96	99.88	99.99	99.71
1 → 2	64.37	99.25	91.17	99.99	100	*100*
1 → 3	79.88	87.81	83.70	99.47	100	*100*
2 → 0	59.05	86.18	67.90	94.11	95.05	*99.23*
2 → 1	63.39	89.31	90.64	95.26	99.99	99.98
2 → 3	65.57	98.07	88.28	100	100	*100*
3 → 0	72.92	76.49	74.60	91.21	89.26	*99.66*
3 → 1	68.93	79.61	74.77	89.95	86.17	*99.83*
3 → 2	63.97	90.66	96.70	100	99.98	*100*
Avg	67.02	87.87	84.12	95.9	96.16	*99.85*

The bold and italicized numbers indicate that UDA-BFD-MDD (proposed method) achieved the best transfer learning performance in the corresponding transfer task. The non bold and italicized numbers indicate that other model achieved the best transfer learning performance in the corresponding transfer task