Multi-Objective Optimization of Laser Cladding Parameters Based on BP Neural Network

Dewei Deng; Hao Jiang; Zhenhua Li; Xueguan Song; Qi Sun; Yong Zhang

doi:10.3788/LOP221821

Journals >Laser & Optoelectronics Progress >Volume 60 >Issue 17 >Page 1714001 > Article

Laser & Optoelectronics Progress
Vol. 60, Issue 17, 1714001 (2023)

Multi-Objective Optimization of Laser Cladding Parameters Based on BP Neural Network

Dewei Deng^1,3,*, Hao Jiang¹, Zhenhua Li¹, Xueguan Song²..., Qi Sun³ and Yong Zhang³|Show fewer author(s)

Author Affiliations

¹Research Center of Laser 3D Printing Equipment and Application Engineering Technology (Liaoning Province), School of Materials Science and Engineering, Dalian University of Technology, Dalian 116024, Liaoning , China

²School of Mechanical Engineering, Dalian University of Technology, Dalian 116024, Liaoning , China

³Shenyang Blower Group Corporation, Shenyang 110869, Liaoning , China

show less

DOI: 10.3788/LOP221821 Cite this Article Set citation alerts

Dewei Deng, Hao Jiang, Zhenhua Li, Xueguan Song, Qi Sun, Yong Zhang. Multi-Objective Optimization of Laser Cladding Parameters Based on BP Neural Network[J]. Laser & Optoelectronics Progress, 2023, 60(17): 1714001 Copy Citation Text

show less

Fig. 1. Morphology of cladding powder

Download full size

Fig. 2. Schematic diagram of laser cladding system

Download full size

Fig. 3. Schematic diagram of laser cladding coating

Download full size

Fig. 4. Pareto and main effects plots of dilution rates. (a) Pareto chart; (b) main effects chart

Download full size

Fig. 5. Residual and main effects plots of micro hardness. (a) Pareto diagram; (b) main effect diagram

Download full size

Fig. 6. Pareto and main effects plots of height. (a) Pareto chart; (b) main effects chart

Download full size

Fig. 7. Pareto and main effects plots of width. (a) Pareto chart; (b) main effects chart

Download full size

Fig. 8. Schematic diagram of BP neural network

Download full size

Fig. 9. GA-BP neural network flow chart

Download full size

Fig. 10. Comparison curves of response volume test and predicted values with coefficient of determination R².(a) Height; (b) width; (c) dilution rate; (d) hardness

Download full size

Fig. 11. Plot of predicted and expected value fit of response volume.(a) Width; (b) height; (c) dilution rate; (d) hardness

Download full size

Fig. 12. Relationship between the refined process parameters and gray correlation degree

Download full size

Fig. 13. Relative error of validation sample performance data for neural networks

Download full size

Material	Mass fraction /%
Material	C	Mn	P	S	Si	Cr	Ni	Fe	Mo
304 stainless steel	≤0.08	≤2.00	≤0.045	≤0.030	≤1.00	18.0-20.0	8.0-11.0	Bal.	—
Ferro 55	0.35	1.1	—	—	0.3	7	—	Bal.	2

Table 1. Chemical composition of the substrate and Ferro 55 alloy powder

Process parameter	Notation	Level
Laser power /W	LP	1000	1200	1400	1600	1800
Scanning speed /（mm·s^-1）	SS	4	5	6	7	8
Flow rate of protective gas /（L·min^-1）	FG	6	7	8	9	10

Table 2. Full factor experiment level table

No.	LP /W	SS /（mm·s^-1）	FG /（L·min^-1）	W /μm	H /μm	η /%	DH /HV
1	1000	4	6	3775.811	1085.781	18.394	893.50
2	1000	4	7	3839.020	1230.670	19.766	640.44
3	1000	4	8	3653.007	1149.451	24.586	816.16
4	1000	4	9	3749.740	1101.470	26.765	631.60
5	1000	4	10	3843.952	963.0877	33.214	643.54
6	1000	5	6	3620.063	1200.113	17.523	878.22
7	1000	5	7	3589.932	1429.445	6.577	951.22
8	1000	5	8	3632.841	1128.766	17.657	847.00
9	1000	5	9	3633.135	1113.519	20.976	822.62
10	1000	5	10	3395.680	1021.660	23.473	832.60
11	1000	6	6	3378.067	1241.917	11.610	848.58
12	1000	6	7	3380.289	1159.290	13.168	833.76
13	1000	6	8	3549.465	1399.060	3.687	954.66
14	1000	6	9	3596.015	1002.324	20.108	749.00
15	1000	6	10	3431.845	986.5434	23.052	665.12
$⋮$	$⋮$	$⋮$	$⋮$	$⋮$	$⋮$	$⋮$	$⋮$
116	1800	7	6	3983.089	1296.855	28.123	705.48
117	1800	7	7	3904.384	1268.997	28.857	667.70
118	1800	7	8	3624.623	1208.514	34.520	566.60
119	1800	7	8	4292.723	1396.216	19.470	821.52
120	1800	7	10	3374.230	746.500	46.727	480.96
121	1800	8	6	3842.054	1322.708	24.122	814.26
122	1800	8	7	3834.420	1294.360	21.603	762.53
123	1800	8	8	3917.358	1266.675	26.265	733.98
124	1800	8	9	3280.512	1314.897	24.122	575.80
125	1800	8	10	3953.457	984.4884	31.864	595.90

Table 3. Morphological data and hardness data of some samples

Source	df	Adjusted SSD	Adjusted MS	F value	P value
S=4.17128 R²=87.21% Adjusted R²=86.21% Predicted R²=84.98%
Model	9	13640.0	1515.56	87.10	0.000
LP	1	451.4	451.37	25.94	0.000
SS	1	60.6	60.61	3.48	0.065
FG	1	47.6	47.59	2.74	0.101
LP×LP	1	429.8	429.80	24.70	0.000
SS×SS	1	47.5	47.50	2.73	0.101
FG×FG	1	126.3	126.29	7.26	0.008
LP×SS	1	4.7	4.67	0.27	0.605
LP×FG	1	23.8	23.83	1.37	0.244
SS×FG	1	64.0	64.01	3.68	0.058
Residual	115	2001.0	17.40
Lack of fit	114	1887.7	16.56	0.15	0.990
Pure error	1	113.3	113.25
Cor total	124	15641.0

Table 4. Analysis of variance results for dilution rates

Source	df	Adjusted SSD	Adjusted MS	F value	P value
S=94.5079 R²=63.26% Adjusted R²=60.38% Predicted R²=57.00%
Model	9	1768213	196468	22.00	0.000
LP	1	64966	64966	7.27	0.008
SS	1	35281	35281	3.95	0.049
FG	1	0	0	0.00	0.998
LP×LP	1	63104	63104	7.07	0.009
SS×SS	1	32684	32684	3.66	0.058
FG×FG	1	153	153	0.02	0.896
LP×SS	1	768	768	0.09	0.770
LP×FG	1	6929	6929	0.78	0.380
SS×FG	1	40	40	0.00	0.947
Residual	115	1027151	8932
Lack of fit	114	994659	8725	0.27	0.944
Pure error	1	32492	32492
Cor total	124	2795364

Table 5. Analysis of variance results for micro hardness

Source	df	Adjusted SSD	Adjusted MS	F value	P value
S=99.1298 R²=49.66% Adjusted R²=45.72% Predicted R²=39.58%
Model	9	1114720	123858	12.60	0.000
LP	1	12319	12319	1.25	0.265
SS	1	3196	3196	0.33	0.570
FG	1	95049	95049	9.67	0.002
LP×LP	1	8421	8421	0.86	0.357
SS×SS	1	612	612	0.06	0.803
FG×FG	1	137532	137532	14.00	0.000
LP×SS	1	273	273	0.03	0.868
LP×FG	1	10017	10017	1.02	0.315
SS×FG	1	31	31	0.00	0.955
Residual	115	1130072	9827
Lack of fit	114	1112456	9758	0.55	0.818
Pure error	1	17616	17616
Cor total	124	2244792

Table 6. Analysis of variance results for height

Source	df	Adjusted SSD	Adjusted MS	F value	P value
S=171.605 R²=78.09% Adjusted R²=76.38% Predicted R²=74.06%
Model	9	12073380	1341487	45.55	0.000
LP	1	218954	218954	7.44	0.007
SS	1	123178	123178	4.18	0.043
FG	1	19186	19186	0.65	0.421
LP×LP	1	33293	33293	1.13	0.290
SS×SS	1	154	154	0.01	0.943
FG×FG	1	6926	6926	0.24	0.629
LP×SS	1	28494	28494	0.97	0.327
LP×FG	1	200091	200091	6.79	0.010
SS×FG	1	43861	43861	1.49	0.225
Residual	115	3386568	29448
Lack of fit	114	3163389	27749	0.12	0.995
Pure error	1	223179	223179
Cor total	124	15459948

Table 7. Analysis of variance results for width

No.	LP /W	SS /（mm·s^-1）	FG /（L·min^-1）	GRC
1	1000	4	6	0.5455
2	1000	4	7	0.4814
3	1000	4	8	0.5468
4	1000	4	9	0.5495
5	1000	4	10	0.5891
6	1000	5	6	0.5075
7	1000	5	7	0.4698
8	1000	5	8	0.5105
9	1000	5	9	0.5254
10	1000	5	10	0.5506
11	1000	6	6	0.4564
12	1000	6	7	0.4691
13	1000	6	8	0.4647
14	1000	6	9	0.5239
15	1000	6	10	0.5242
$⋮$	$⋮$	$⋮$	$⋮$	$⋮$
25	1000	8	10	0.3947
$⋮$	$⋮$	$⋮$	$⋮$	$⋮$
106	1800	4	6	0.6250
$⋮$	$⋮$	$⋮$	$⋮$	$⋮$
116	1800	7	6	0.5705
117	1800	7	7	0.5704
118	1800	7	8	0.5028
119	1800	7	8	0.5482
120	1800	7	10	0.5502
121	1800	8	6	0.5344
122	1800	8	7	0.5060
123	1800	8	8	0.5505
124	1800	8	9	0.4607
125	1800	8	10	0.6033

Table 8. Gray correlation degree of some samples

Process parameter	Parameter range	Upward gradient	Number of levels
LP /W	1000-1800	10	81
SS /（mm·s^-1）	4-8	0.1	41
FG /（L·min^-1）	6-10	0.5	9

Table 9. Parameter level design

Sample number	LP /W	SS /（mm·s^-1）	FG /（L·min^-1）
T₁	1370	5.0	7.5
T₂	1410	6.5	8.0
T₃	1060	6.2	6.0
T₄	1700	7.5	9.0
T₅	1500	4.0	7.0
H₁	1090	4.4	10.0
H₂	1090	4.4	10.0
H₃	1090	4.4	10.0

Table 10. Sample parameters for validation

Sample	W /μm	H /μm	η /%	DH /HV
T₁	3882.1	1287.5	24.069	715.30
T₁′	3993.5	1234.3	30.000	602.05
T₂	3885.3	1301.9	21.143	754.24
T₂′	3723.9	1198.0	25.800	707.74
T₃	3405.7	1299.4	12.671	772.44
T₃′	3504.1	1277.9	12.732	905.74
T₄	3907.8	1259.2	27.893	660.16
T₄′	3771.5	1200.0	32.221	610.27
T₅	4129.7	1191.8	32.907	529.98
T₅′	4225.3	1209.1	38.730	535.17