Modified Kernel Marginal Fisher Analysis for Feature Extraction and Its Application to Bearing Fault Diagnosis

Table 1

Description of the data sets.

Data set	Training sample size	Testing sample size	Defect size (inches)	Load (hp)	Fault type	Classification label
A	10	90	0.021	1	I	1
	10	90	0.021	2	B	2
	10	90	0.021	3	O	3
	10	90	0	0	N	4
B	Varied from 10 to 90 each class	The remaining samples	0.021	1	I	1
			0.021	2	B	2
			0.021	3	O	3
			0	0	N	4
C	50	50	0.007	1	I	1
	50	50	0.007	1	B	2
	50	50	0.007	1	O	3
	50	50	0.014	2	I	4
	50	50	0.014	2	B	5
	50	50	0.014	2	O	6
	50	50	0.021	3	I	7
	50	50	0.021	3	B	8
	50	50	0.021	3	O	9
	50	50	0	0	N	10

I: inner race fault, B: ball fault, O: outer race fault, and N: normal.