Table 7.

Diagnosis accuracy of AICNN-1D with input-corrupt training and global average pooling, with clean-input training and with fully-connected layers.

Method	SNR (dB)
	−2	0	2	4	6	8	10	12
DCAE-1D + AICNN-1D (model A)	96.65	99.23	99.42	99.58	99.63	99.72	99.75	99.78
DCAE-1D + AICNN-1D (model B)	93.54	97.32	98.89	99.19	99.25	99.30	99.60	99.75
DCAE-1D + AICNN-1D (model C)	94.26	97.89	99.15	99.26	99.37	99.58	99.68	99.78