Multiscale Time-Frequency Sparse Transformer Based on Partly Interpretable Method for Bearing Fault Diagnosis

Table 1

The ablation study of different hyperparameters.

Label												Parameters number (M)	Average accuracy (%)
Baseline		512	256	128	64	128	8	6	6	0.1	0.5	1.53	99.34
A	A1	256	128	64								1.57	94.23
	A2	1024	512	256								1.53	92.15
	A3	128	256	512								2.07	88.16
B	B1				128	128						4.87	99.46
	B2				64	64						1.27	97.13
	B3				128	64						4.51	98.74
C	C1						16					1.53	95.77
	C2						4					1.53	83.12
D	D1							8				1.68	99.42
	D2							10				1.90	99.17
	D3							4				1.23	94.14
E	E1								8			1.60	98.34
	E2								10			1.68	94.12
	E3								4			1.45	97.76
F	F1									0.01		1.53	92.72
	F2									0.3		1.53	96.22
G	G1										0.3	1.53	97.32
	G2										0.7	1.53	98.67
	G3										1	1.53	98.46