1 Introduction
2 Background
2.1 Sheet metal forming
2.2 Machine learning applications to sheet metal forming
Authors | Strategy | Forming process | Material | Features | Outputs |
---|---|---|---|---|---|
Inamdar et al. [20] | BP-ANN | Air V-bending | Steel and aluminium alloys (4) | Material parameters + Process parameters | Springback angle + Punch displacement |
Guo and Tang [15] | BP-ANN | Air V-bending | Steel and aluminium alloys (5) | Sheet thickness + Material parameters + Process parameters | Bending springback angle |
Miranda et al. [28] | ANN+FEM | Air V-bending | Steel alloys (2) | Sheet thickness + Process parameters | Punch displacement |
Kazan et al. [22] | BP-ANN+FEM | Wipe bending | High-strength steel | Sheet thickness + Process parameters | Springback angle |
Nasrollahi and Arezoo [30] | BP-ANN+FEM | Wipe bending on perforated metal sheets | Steel alloys (2) | Hole number and geometry + Process parameters + Type of material | Springback angle |
Gisario et al. [13] | BP-ANN | Laser bending | Aluminium alloy | Starting deflection + Process parameters | Springback angle |
Ruan et al. [39] | BP/GA-ANN | Multicurvature parts | Steel and aluminium alloys (4) | Sheet thickness + Process parameters | Springback angles |
Liu et al. [25] | GA-ANN | U-bending | Not specified | Sheet thickness + Material parameters + Process parameters | Springback angle |
Sharad and Nandedkar [41] | ANN+FEM | U-bending | Steel alloys (2) | Sheet thickness + Material parameters + Process parameters | Springback angles |
Dib et al. [9] | MLP+FEM SVM+FEM DT+FEM RF+FEM NB+FEM | U-bending | Steel alloys (3) | Sheet thickness + Material parameters + Process parameters | Springback angle + Maximum thinning |
Phatak et al. [32] | BP-ANN+FEM | Axisymmetric cup deep drawing | Aluminium alloy | Material parameters | Thickness + Friction coefficient |
3 Proposed approach for building defect predictive models
3.1 Single learning classifier models
-
Multilayer perceptron (MLP)
-
Decision tree (DT)
-
Random forest (RF)
-
Naive Bayes (NB)
-
Support vector machine (SVM)
-
K-Nearest neighbours (KNN)
-
Logistic regression (LR)
3.1.1 Multilayer perceptron (MLP)
3.1.2 Decision tree (DT)
3.1.3 Random forest (RF)
3.1.4 Naive Bayes (NB)
3.1.5 Support vector machine (SVM)
3.1.6 K-nearest neighbours (KNN)
3.1.7 Logistic regression (LR)
3.2 Ensemble models
4 Experimental set-up
4.1 Simulated data sets
Materials | C (MPa) | n | \({{Y}_0}\) (MPa) | E (GPa) | \({{r}_0}\) | \({{r}_{45}}\) | \({{r}_{90}}\) | \({{t}_0}\) (mm) | |
---|---|---|---|---|---|---|---|---|---|
DC06 | µ | 565.32 | 0.259 | 157.12 | 206 | 1.790 | 1.510 | 2.270 | 0.780 |
σ | 26.85 | 0.018 | 7.16 | 3.85 | 0.051 | 0.037 | 0.121 | 0.013 | |
HSLA340 | µ | 673.00 | 0.131 | 365.30 | 210 | 0.820 | 1.070 | 1.040 | 0.780 |
σ | 32.30 | 0.011 | 10.67 | 7.35 | 0.033 | 0.039 | 0.061 | 0.005 | |
DP600 | µ | 1093.00 | 0.187 | 330.30 | 210 | 1.010 | 0.760 | 0.980 | 0.780 |
σ | 52.46 | 0.020 | 9.64 | 7.35 | 0.040 | 0.030 | 0.060 | 0.010 |
Material | Springback (mm) | Maximum thinning (%) | ||
---|---|---|---|---|
\(\mathrm{BHF}=4.9\) kN | \(\mathrm{BHF}=19.6\) kN | \(\mathrm{BHF}=4.9\) kN | \(\mathrm{BHF}=19.6\) kN | |
U-Channel | ||||
DC06 | 5.67 | 2.62 | 2.85 | 9.58 |
HSLA340 | 8.75 | 5.11 | 2.70 | 7.70 |
DP600 | 11.19 | 8.55 | 2.08 | 5.86 |
Maximum EPS | Maximum thinning (%) | |||
---|---|---|---|---|
\(\mathrm{BHF}=2.45\) kN | \(\mathrm{BHF}=9.8\) kN | \(\mathrm{BHF}=2.45\) kN | \(\mathrm{BHF}=9.8\) kN | |
Square Cup | ||||
DC06 | 0.92 | 0.87 | 14.01 | 20.00 |
HSLA340 | 0.92 | 0.86 | 24.22 | 47.43 |
DP600 | 1.14 | 0.85 | 28.35 | 39.18 |
4.2 Data set pre-processing
4.3 Performance measures
5 Results and discussion
5.1 Single classifiers
5.2 Ensemble classifiers
Material | Algorithms | F-score (%) |
---|---|---|
U-Channel–springback | ||
DC06 | DT, MLP, SVM | \(90.82 \pm 1.07\) |
HSLA340 | DT, MLP, SVM | \(92.90 \pm 0.93\) |
DP600 | DT, MLP, SVM | \(92.17 \pm 0.93\) |
U-Channel–maximum thinning | ||
DC06 | DT, MLP, LR | \(93.42 \pm 0.75\) |
HSLA340 | RF, MLP, LR | \(92.24 \pm 0.90\) |
DP600 | DT, MLP, LR | \(91.98 \pm 1.17\) |
Square Cup–maximum EPS | ||
DC06 | RF, DT, KNN, MLP, SVM | \(88.45 \pm 1.26\) |
HSLA340 | KNN, MLP, SVM | \(88.87 \pm 1.19\) |
DP600 | RF, DT, KNN, MLP, SVM | \(89.58 \pm 1.47\) |
Square Cup–maximum thinning | ||
DC06 | RF, DT, MLP, SVM, LR | \(87.62 \pm 1.39\) |
HSLA340 | DT, MLP, SVM | \(83.58 \pm 1.35\) |
DP600 | DT, MLP, SVM | \(83.80 \pm 1.54\) |
Material | Algorithms | F-score (%) | AUC (%) | |
---|---|---|---|---|
Base learners | Meta-classifier | |||
U-Channel–springback | ||||
DC06 | KNN, MLP, SVM | SVM | 91.42 ± 1.04 | 93.53 ± 1.12 |
HSLA340 | KNN, MLP, SVM | SVM | 93.74 ± 0.88 | 95.65 ± 0.79 |
DP600 | KNN, MLP, SVM | SVM | 92.89 ± 0.75 | 95.33 ± 0.71 |
U-Channel–maximum thinning | ||||
DC06 | KNN, NB, LR | SVM | 93.25 ± 0.98 | 95.29 ± 1.18 |
HSLA340 | RF, DT, MLP, SVM, LR | MLP | 92.09 ± 1.16 | 96.38 ± 0.68 |
DP600 | RF, DT, MLP, SVM, LR | MLP | 91.58 ± 1.13 | 96.25 ± 0.81 |
Square Cup–maximum EPS | ||||
DC06 | MLP, SVM, LR | RF | 89.74 ± 1.37 | 92.22 ± 1.48 |
HSLA340 | KNN, MLP, SVM | SVM | 90.49 ± 0.98 | 91.20 ± 1.84 |
DP600 | MLP, SVM, LR | LR | 90.01 ± 1.08 | 92.96 ± 0.96 |
Square Cup–maximum thinning | ||||
DC06 | KNN, MLP, SVM | RF | 89.33 ± 1.45 | 91.87 ± 1.26 |
HSLA340 | KNN, MLP, NB | RF | 84.47 ± 1.39 | 87.32 ± 1.49 |
DP600 | KNN, MLP, SVM | LR | 84.62 ± 1.34 | 87.22 ± 1.09 |