Table 3 Experimental results of predicted and measured lg|Z|0.01Hz values of scratched coatings prepared under various proposed conditions.

From: Machine learning assisted discovery of high-efficiency self-healing epoxy coating for corrosion protection

Cycle

Rank

Variable parameter

MWc (g·mol–1)

r

UPy-D400 content (mol%)

ZIF-8/Ca content (wt.%)

Predicted lg(|Z|/Ω·cm2)

Measured lg(|Z|/Ω·cm2)

Initial

1

400

1.00

20

5.5

10.49 ± 0.32

10.15 ± 0.30

2

400

1.00

20

7.0

9.96 ± 0.24

10.88 ± 0.44

3

400

0.85

10

5.5

9.71 ± 0.14

10.3 ± 0.41

4

2000

0.85

15

5.5

8.93 ± 0.29

8.35 ± 0.62

5

2000

0.85

20

8.5

9.33 ± 0.20

9.05 ± 0.75

Cycle 1

1

400

0.85

20

7.0

10.14 ± 0.19

10.11 ± 0.44

2

400

1.00

20

10.0

9.88 ± 0.17

10.24 ± 0.13

3

400

1.00

15

7.0

9.52 ± 0.20

10.52 ± 0.46

4

2000

1.00

10

7.0

9.57 ± 0.18

8.23 ± 0.29

5

400

0.85

20

10.0

9.65 ± 0.17

9.52 ± 0.51

Cycle 2

1

400

0.85

15

5.5

10.27 ± 0.21

10.08 ± 0.30

 

2

400

1.00

15

8.5

10.23 ± 0.25

11.03 ± 0.38

 

3

400

0.85

15

7.0

10.03 ± 0.15

10.76 ± 0.46

 

4

400

1.00

15

10.0

9.90 ± 0.20

9.63 ± 0.64

 

5

400

0.85

15

8.5

10.31 ± 0.12

10.26 ± 0.71

Cycle 3

1

400

0.85

15

10.0

9.44 ± 0.24

10.25 ± 0.75

2

2000

0.85

15

7.0

9.12 ± 0.35

9.62 ± 0.54

3

2000

0.85

15

8.5

9.37 ± 0.36

9.94 ± 0.48

4

2000

1.00

15

5.5

8.91 ± 0.44

9.41 ± 0.38

5

2000

1.00

15

8.5

9.43 ± 0.18

9.22 ± 0.15

Cycle 4

1

2000

0.85

15

10.0

9.40 ± 0.30

9.68 ± 0.80

2

2000

0.85

20

8.5

9.30 ± 0.25

9.03 ± 0.68

3

2000

1.00

15

10.0

9.30 ± 0.20

9.84 ± 0.51

4

2000

1.00

15

7.0

9.24 ± 0.22

9.10 ± 0.74

5

2000

1.00

20

8.5

9.00 ± 0.18

9.62 ± 0.48

Cycle 5

1

2000

0.85

20

7.0

9.04 ± 0.10

9.18 ± 0.84

2

2000

0.85

20

10.0

9.28 ± 0.08

9.45 ± 0.54

3

2000

1.00

20

10.0

9.06 ± 0.15

9.21 ± 0.69

4

2000

1.00

20

5.5

8.99 ± 0.18

9.30 ± 0.50

5

2000

0.85

20

7.0

9.16 ± 0.20

9.09 ± 0.25

  1. Initial step: the top-five proposed experiments were obtained by a model trained on initial 32 samples in the range of remaining 224 untested experiments; Cycle 1: From the remaining 219 untested experiments, the another top-five proposed experiments were obtained by a model trained on 37 samples. Cycle 2 ~ 5 utilized the same method to obtain new proposed experiment and train the model.
Back to article page