Prediction of Cladding Layer Morphology Based on BP Neural Network Optimized by Regression Analysis and Genetic Algorithm

Sirui Yang; Haiqing Bai; Jun Bao; Li Ren; Chaofan Li

doi:10.3788/LOP202259.2114002

Laser & Optoelectronics Progress, Volume. 59, Issue 21, 2114002(2022)

Prediction of Cladding Layer Morphology Based on BP Neural Network Optimized by Regression Analysis and Genetic Algorithm

Sirui Yang¹, Haiqing Bai^1,2、*, Jun Bao¹, Li Ren¹, and Chaofan Li¹

Author Affiliations

¹School of Mechanical Engineering, Shaanxi University of Technology, Hanzhong 723001, Shaanxi, China

²Shaanxi Key Laboratory of Industrial Automation, Hanzhong 723001, Shaanxi, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(18)

Fig. 1. 3 kW fiber laser cladding machine

Download full size

Fig. 2. Ultra depth of field microscope

Download full size

Fig. 3. Surface morphology of cladding layer with different process parameters

Download full size

Fig. 4. Measurement position of cladding layer width and height

Download full size

Fig. 5. Algorithm flow chart of GA-BP neural network

Download full size

Fig. 6. BP neural network structure diagram

Download full size

Fig. 7. Comparison between test values and predicted values. (a) Width; (b) height

Download full size

Fig. 8. Iteration curves of fitness function. (a) Width; (b) height

Download full size

Fig. 9. Fitting diagrams of cladding layer morphology predicted and expected value. (a) Width; (b) height

Download full size

Fig. 10. Morphology of 5 groups of test cladding layers

Download full size

Fig. 11. Comparison between test value and predicted value of cladding layer width

Download full size

Fig. 12. Comparison of prediction errors of cladding width with different prediction models

Download full size

Table 1. Chemical composition of test materials
View table
Table 1. Chemical composition of test materials
Item Mass fraction /%
C Si Mn Ni Cr Fe
45 steel 0.45 0.27 0.65 0.25 0.12 Bal.
Fe45 powder 0.75 3.00 3.00 13.00 16.00 Bal.

Table 2. Test factor level table
View table
Table 2. Test factor level table
Process parameter Level1 Level2 Level3
Level
4
Level5
Laser power /kW 1.80 1.95 2.10 2.25 2.40
Scanning speed /（mm·s^-1） 4 6 8 10 12
Powder feed speed /（r·min^-1） 1.8 2.0 2.2 2.4 2.6

Table 3. Test scheme and results

View table

Table 3. Test scheme and results

Number	Laser power /kW	Scanning speed /（mm·s^-1）	Powder feed speed /（r·min^-1）	Height /mm	Width /mm	Number	Laser power /kW	Scanning speed /（mm·s^-1）	Powder feed speed /（r·min^-1）	Height /mm	Width /mm
1	1.80	4	1.8	0.663	2.163	14	2.10	10	1.8	0.378	2.173
2	1.80	6	2.0	0.537	2.228	15	2.10	12	2.0	0.310	2.124
3	1.80	8	2.2	0.430	1.975	16	2.25	4	2.4	1.258	2.460
4	1.80	10	2.4	0.381	2.170	17	2.25	6	2.6	1.032	2.425
5	1.80	12	2.6	0.483	2.032	18	2.25	8	1.8	0.465	2.338
6	1.95	4	2.0	0.840	2.264	19	2.25	10	2.0	0.428	2.120
7	1.95	6	2.2	0.710	2.244	20	2.25	12	2.2	0.510	2.157
8	1.95	8	2.4	0.632	2.180	21	2.40	4	2.6	1.512	2.622
9	1.95	10	2.6	0.560	2.080	22	2.40	6	1.8	0.804	2.440
10	1.95	12	1.8	0.358	2.123	23	2.40	8	2.0	0.590	2.487
11	2.10	4	2.2	1.050	2.570	24	2.40	10	2.2	0.493	2.324
12	2.10	6	2.4	0.663	2.335	25	2.40	12	2.4	0.403	2.390
13	2.10	8	2.6	0.650	2.254

Table 4. Parameters of GA-BP prediction algorithm

View table

Table 4. Parameters of GA-BP prediction algorithm

Parameter settings	Parameter value	Parameter settings	Parameter value
Population size	40	Variables get binary bits	10
Maximum number of iterations	80	Number of training	1000
Crossover probability	0.71	Training goal	0.01
Mutation probability	0.01	Learning rate	0.1

Table 5. Error analysis of the test value and predicted value of cladding layer morphology

View table

Table 5. Error analysis of the test value and predicted value of cladding layer morphology

Number	Width			Height
Number	Test value / $m m$	Predicted value / $m m$	Error /%	Test value / $m m$	Predicted value / $m m$	Error /%
3	1.975	2.05	3.59	0.430	0.44	3.50
9	2.080	2.17	4.46	0.560	0.54	-2.32
11	2.570	2.59	1.11	1.050	1.07	2.69
18	2.338	2.28	-2.39	0.465	0.45	-3.44
24	2.324	2.47	3.22	0.493	0.51	4.91

Table 6. Comparison between test value and predicted value of width of cladding layer

View table

Table 6. Comparison between test value and predicted value of width of cladding layer

No．	Laser power /kW	Scanning speed /（mm·s^-1）	Powder feed speed /（r·min^-1）	Test result /mm	GA-BP /mm	BP /mm	Regression analysis /mm
1	1.80	8	1.8	2.463	2.558	2.643	2.082
2	1.80	12	2.0	2.182	2.240	2.064	1.963
3	1.95	8	2.0	2.384	2.421	2.482	2.174
4	1.95	12	2.2	2.280	2.233	2.138	2.050
5	2.10	8	2.2	2.413	2.460	2.491	2.265

Tools

Get Citation

Copy Citation Text

Sirui Yang, Haiqing Bai, Jun Bao, Li Ren, Chaofan Li. Prediction of Cladding Layer Morphology Based on BP Neural Network Optimized by Regression Analysis and Genetic Algorithm[J]. Laser & Optoelectronics Progress, 2022, 59(21): 2114002

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites