Table 9 Comparison of RUL prediction by CNNs with different number of cells.

From: Light convolutional neural network by neural architecture search and model pruning for bearing fault diagnosis and remaining useful life prediction

Cell count

Test condition

Test loss

R-square

Predicted RUL (\(\times\)10 s)

Actual RUL (\(\times\)10 s)

RUL error percentage

2

Condition 1_4

0.863

0.965

1281

1126

13.7%

Condition 1_5

2.704

0.892

2500

2295

8.9%

Condition 1_6

1.299

0.948

2453

2295

6.8%

3

Condition 1_4

1.574

0.937

1117

1126

0.8%

Condition 1_5

6.620

0.735

2002

2295

12.7%

Condition 1_6

3.023

0.879

2161

2295

5.8%

4

Condition 1_4

1.449

0.942

1083

1126

3.8%

Condition 1_5

5.899

0.764

2066

2295

10.0%

Condition 1_6

2.673

0.893

2257

2295

1.7%

5

Condition 1_4

1.489

0.940

1159

1126

2.9%

Condition 1_5

4.414

0.823

2131

2295

7.1%

Condition 1_6

2.402

0.904

2455

2295

6.9%

6

Condition 1_4

1.228

0.951

1130

1126

0.4%

Condition 1_5

4.715

0.811

2222

2295

3.2%

Condition 1_6

2.324

0.907

2292

2295

0.1%

Back to article page