Table of Contents Author Guidelines Submit a Manuscript
Mathematical Problems in Engineering
Volume 2017 (2017), Article ID 7834621, 11 pages
https://doi.org/10.1155/2017/7834621
Research Article

Bending Angle Prediction Model Based on BPNN-Spline in Air Bending Springback Process

Department of Mechanical Engineering, Southeast University, Nanjing 211189, China

Correspondence should be addressed to Wencheng Tang; nc.ude.ues@cwgnat

Received 16 October 2016; Revised 16 January 2017; Accepted 7 February 2017; Published 27 February 2017

Academic Editor: Marek Lefik

Copyright © 2017 Zhefeng Guo and Wencheng Tang. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Abstract

In order to rapidly and accurately predict the springback bending angle in V-die air bending process, a springback bending angle prediction model on the combination of error back propagation neural network and spline function (BPNN-Spline) is presented in this study. An orthogonal experimental sample set for training BPNN-Spline is obtained by finite element simulation. Through the analysis of network structure, the BPNN-Spline black box function of bending angle prediction is established, and the advantage of BPNN-Spline is discussed in comparison with traditional BPNN. The results show a close agreement with simulated and experimental results by application examples, which means that the BPNN-Spline model in this study has higher prediction accuracy and better applicable ability. Therefore, it could be adopted in a numerical control bending machine system.

1. Introduction

The sheet metal bending is an essential metalworking process in many industries [1], wherein V-die air bending is a main topic in the stamping field. The flat sheet is bent to a curved sheet by the pressing on the bending machine. Because of the existence of elastic deformation in curved sheet, springback will occur during the unloaded process of curved sheet in the bending machine. As a result, the final bending angle of curved sheet will be obtained.

Mohammadi et al. [2] have developed a springback prediction analytical formulation for sandwich sheets based on a wrap-around assumption and primary bending theories. Numerical simulation and experiments have been carried out to verify the effect of punch radius, die opening, and punch stroke on springback. Results showed that the springback is increased with an increase in punch stroke, punch radius, and die opening. De Vin [3] has presented an air bending geometric model. This model has been found to be convenient to calculate the bending angle.

Finite element (FE) simulation is a popular method at present [4], due to low time costs and simple operation. Vorkov et al. [5] have studied the springback of different types of high-strength steels using FE model. The material properties and a correct material model for high-strength steels have been obtained and built, respectively. Fu et al. [6] have developed a material constitutive model based on Hill’s yielding criterion under plane strain conditions. The multiple step incremental air bending forming and springback processes have been simulated. Jamli et al. [7] have developed an artificial neural network based on constitutive model which can be used in the FE code through user defined material subroutine. FE analysis procedures have been carried out for the springback prediction of sheet metal based on an L-bending experiment. The FE analysis presented an improvement in the prediction accuracy in comparison with the measured data.

The approximate model can be chosen to replace the FE simulation. When the approximate model is established, the response of research parameters is obtained immediately. Approximate model is an efficient research method to reduce the time consumption. Many kinds of approximate model are able to achieve the approximation of bending springback process, such as artificial neural network [811], radial basis function [12, 13], and response surface method [14, 15]. The error backpropagation neural network (BPNN) is one of the widely used and successful networks at present. It is particularly useful in process modeling and has been used in diverse applications, such as control, robotics, pattern recognition, forecasting, and manufacturing optimization [16]. In general, the approximate model needs a lot of samples. Orthogonal experiment [1719], Latin hypercube design [20], and so forth can reduce the number of samples under the same approximate calculation accuracy.

FE simulation and theory calculation method are mainly used to research the V-die air bending springback process at present. Theory calculation method has better general property and faster prediction speed. However, there are many inevitable assumptions in calculation model which will result in low calculation accuracy. In terms of the FE simulation, at least to the authors’ knowledge, one FE model only simulates one group of bending data. Although it generates more accurate result, it is not speedy enough for bending angle prediction. In addition, the technical requirement for the user on FE software is higher. Therefore, there is a need towards an accurate, general, speedier, and more maneuverable means to predict the bending angle.

In this study, the authors propose a springback bending angle prediction model which is the combination of BPNN and spline function (BPNN-Spline). The BPNN-Spline combined model is expressed by simple mathematical formulas. The complex numerical simulation process will be turned into a simple mathematical calculation procedure. It enables a faster and maneuverable bending angle prediction in comparison with FE simulation. Besides, the BPNN-Spline combined model is developed by the accurate FE simulation results, where the prediction accuracy is relatively high. It will offset the disadvantages of FE simulation and theory calculation method at the same time. Due to the full utilization of better generalization ability of BPNN and smooth feature of the spline function, the model outperforms traditional BPNN model. Also, through the test of application examples, the general and precise properties of BPNN-Spline are also fully demonstrated. During the bending process, input the parameters to the BPNN-Spline black box; the springback bending angle is obtained immediately. Consequently, as a computation method, the model can be applied properly in a bending machine numerical control system.

2. Introduction of Air Bending Springback Process

The flat metal sheet is bent into a curved sheet with angle on the bending machine, as shown in Figure 1(a). Due to the existence of material elasticity, when the punch is unloaded, the springback will occur, and the curved sheet will become angle (hereinafter uniformly referred to as bending angle) from , as shown in Figure 1(b), where , , , , , and are the nominal width of die notch, die corner radius, punch corner radius, sheet thickness, die angle, and punch displacement, respectively.

Figure 1: Sheet bending springback process diagram: (a) curved sheet before springback; (b) curved sheet after springback.

Bending angle is related to various process parameters, such as material, sheet thickness, and tool corner radius. When other process parameters are determined, the value of will correspond to the value of . The desired can be obtained by reasonable adjustment of . The purpose of this study is to develop a bending angle prediction model which is suitable for bending machine numerical control system. This model provides a precise mapping relation between and under diverse process parameters.

3. FE Simulation of Bending Springback Process

3.1. Research Materials and Parameter Levels

The bending angle is closely related to the material plasticity. In general, the relation between stress and strain of material plasticity could be obtained from tensile test. Then, through a series of transformation and fitting, the exponential form relation curve between true stress and true strain can be considered as follows:where , , and are the hardening coefficient, hardening exponent, and yield strain, respectively.

In this study, material parameters are determined based on tensile tests, as listed in Table 1. For the tensile test of one material, the material parameters of different sheet thickness are lightly different. The error caused by different sheet thickness is a micro amount with respect to the error caused by approximate prediction system. Therefore, in order to facilitate the operation, the variation of material parameter with different sheet thickness is not considered.

Table 1: Parameters of research materials.

As for the BPNN, the factor property should be measurable. One material has many measurable parameters such as , , and at the same time. Each material is indicated by an integrated value , as listed in Table 1, where and are Young’s modulus and yield stress, respectively.

In this study, to develop the springback bending angle prediction model, , , , , and are the input parameters, and bending angle is the output. The various parameters are listed in Table 2. and are 12 mm and 86°, respectively.

Table 2: Variable parameters with their levels.
3.2. Orthogonal Experimental Design

To further reduce the number of simulations, the orthogonal experimental design is implemented in this study. Orthogonal experimental design is an efficient design method whose features are uniform distribution and comparable order. The orthogonal experimental design sample sets are listed in Table 3.

Table 3: Experimental sample sets.

The outputs of Table 3 are the bending angles which are obtained through FE simulation. The measurement unit of is degree. Each corresponds to certain punch displacement in FE simulation. are 5 mm, 4.4 mm, 4 mm, 3 mm, and 2.6 mm, respectively. The anterior 25 samples in Table 3 are the orthogonal experiment sample sets for training the BPNN, and the extra 5 samples are selected randomly from full factorial design to test the accuracy of BPNN.

Table 3 shows that the bending angles on the same are different for different materials and bending process parameters. But the bending angles on the same are more close to each other in comparison with those bending angles of different . The factor is more important for the value of bending angle.

3.3. FE Model of Bending Springback Process

FE software ABAQUS is used to simulate the bending springback process. In the bending and springback process, sheet sectional deformation of width direction is almost the same. 2D elements have been used in the FE model in order to reduce the simulation time. In the FE, the surface to surface contact model is adopted where the coulomb friction law is considered. The friction coefficient is set to 0.1 as an ordinary scale of metal contact. The punch and die are considered as rigid body. Element type is chosen as the four-node bilinear plane stress quadrilateral element (CPS4R). The mesh division figure is shown in Figure 2. The remaining parameters are set to the default of the software.

Figure 2: FE mesh model of bending process.

As it is shown in Table 3, some bending angles are even less than the die angle which is equal to 86°. Since some materials in Table 1 have bigger hardening exponent value such as SUS304, these materials affect the material strengthening in bending process. The curved sheet arcs of these materials are large. If is larger, the large curved sheet arc will be bent for the second time when the low surface touches the inner wall of die notch, as shown in Figure 3. It is inevitable that the bending angle is less than die angle. However, the sheet only touches the die corner radius for most material.

Figure 3: Second bending phenomenon of larger curved arc sheet.
3.4. Verification of FE Model

Aiming at validating the FE model, the FE simulated results are compared with the experiment under the same process conditions. There are many materials and process parameters in the FE model. The experimental verification on each FE model is unnecessary. Thus, a set of material and process parameters within the variable space is chosen randomly to implement the FE simulation and verification. Due to the randomness of selection, the real precision of FE model in Table 3 will be revealed.

The sheet of steel material B340LA is selected randomly to carry out the experiment on the bending machine. is directly observed from bending machine number control equipment named Delem. The sheet is unloaded directly after bending. It will undergo springback immediately. After springback, angle measurement instrument is used to measure the final bending angle . is the mean value of three bending angles which are measured from the same . The experiment moulds are shown in Figure 4.

Figure 4: Moulds of bending experiment.

The experimental moulds and steel sheet parameters are as follows:

With the change of punch displacement, the comparisons of bending angles on FE simulation and experiment are shown in Figure 5.

Figure 5: Comparison of bending angles on FE and experiment.

In Figure 5, it can be found that the FE simulated bending angles are consistent with the bending experiment results. Most of bending angle deviations between FE and experiment are no more than 1 degree, and the largest bending angle deviation is only 1.63 degrees at the place where  mm. This indicates that the FE model has enough accuracy to simulate the bending springback process.

3.5. Effect of Length Ratio Factor on Bending Angle

As we all know, the geometrical position relationship of  mm is considered as a length ratio zooming on  mm. In this case, in terms of the bending process of  mm, its process parameters such as , , , and can be transformed by the length ratio zooming. Under this condition which transformed process parameters that belong to the range of Table 2, if the length ratio on the effect of bending angle is verified to be a small amount, the bending process of  mm is calculated by the BPNN-Spline model which is developed under the condition of  mm. This is beneficial to extend the application range of BPNN-Spline. The relationship between bending angle and length ratio factor () is discussed as follows.

A sample (the sample of number 3 is selected in this study) in Table 3 is randomly selected and its length values are zoomed by the ratio factors and , respectively. It means that the nominal widths of die notch V are 24 mm and 30 mm, respectively. , , , and are also, respectively, zoomed by these length ratio factors. The effect of length ratio factor on bending angle is shown in Figure 6.

Figure 6: Effect of length ratio factor on bending angle.

The effect of length ratio factor on bending angle is so small that it is ignored. Therefore, the bending angle is unrelated with length ratio in the process.

4. Mathematical Model

4.1. Mathematical Model of BPNN

In this study, to develop the BPNN-Spline mathematical combination model, , , , and are treated as input parameters of BPNN, and is treated as the input parameter of spline function.

The BPNN is trained and tested by the MATLAB software. Training is completed when the mean squared error (MSE) reached the minimum. The MSE formula iswhere is the number of outputs, is the th predicted bending angle, and is the th actual bending angle.

When is directly treated as output of BPNN, the relationship between hidden layer node numbers and MSE of training and test are shown in Figure 7(a). Taking as the output of BPNN, the relationship between hidden layer node numbers and MSE of training and test are shown in Figure 7(b), where are 85°, 95°, 100°, 120°, and 128°, respectively.

Figure 7: Effect of hidden layer node numbers on MSE for different output formats: (a) taking as output; (b) taking as the output.

The training of BPNN in Figure 7(b) is better than Figure 7(a). When is treated as output of BPNN, five output nodes will be set at the same magnitude level. This is more conducive to train the network. Thus, is treated as the output of BPNN in this study. It can be seen that the BPNN will have a better comprehensive error for training and test when hidden layer node number is six, as shown in Figure 7(b). As a result, the BPNN consisted of four input nodes, six hidden layer nodes, and five output nodes. The BPNN-Spline combined model is shown in Figure 8. The symmetrical sigmoid and purelin function are selected as the activation function on hidden and output layer, respectively. After the training and test, the MSE of training and test are 0.4947 and 1.7354, respectively.

Figure 8: Diagram of BPNN-Spline combined model.

A series of relation formulas between process parameters and bending angle can be obtained as follows:where , , , , and are the input, connection weight between input and hidden layer, connection weight between hidden layer and output, and threshold values of hidden and output layer, respectively.

The BPNN model could be considered as a black box function which is the mapping between bending process parameters and bending angles:

4.2. Combined Model of BPNN-Spline

Five as the outputs of BPNN are, respectively, corresponding to in Table 3; this one-to-one corresponding relation is equivalent to five points such as , , , , and . The five points are treated as interpolation points; a cubic spline function will be fitted, as shown in Figure 9 and (6). and are, respectively, treated as the independent variable and dependent variable.

Figure 9: Fitted spline curve diagram.

The combination of (5) and (6) derives the following:

Equation (7) is obtained on the precondition of  mm. Because the length ratio factor was defined as , when the bending process parameters are , , , and , the bending angle is predictable based on the conversion of process parameters as the formula , , , and . Treating as an input parameter, (7) can be written as

Equation (8) is a general prediction model on bending springback process.

4.3. Comparison with Traditional BPNN

In order to verify the advantage of BPNN-Spline, its prediction accuracy is compared with traditional BPNN based on the FE simulation results of Section 3.4, as shown in Figure 10.

Figure 10: Comparison of prediction accuracy between BPNN-Spline and traditional BPNN.

Traditional BPNN is a network by taking as the output, , , , , and as the inputs, and eight as the number of hidden layer nodes. The anterior 25 samples in Table 3 are normalized to train the BPNN (where is the unique output, the training sample number becomes 125). The MSE of training and test are 11.2 and 6.8, respectively.

In Figure 10, the bending angle deviation between traditional BPNN and FE is larger. The biggest deviation is 6.5 degrees, which does not meet the accuracy requirement. However, the bending angle deviation between BPNN-Spline and FE is so small that the biggest deviation is only 1.78 degrees. Thus, BPNN-Spline is more precise in comparison with the traditional BPNN method.

4.4. Discussion

To predict the bending angle, the BPNN-Spline has great advantages in comparison with the traditional BPNN model. It mainly displays two aspects: The higher training accuracy and monotonic decrease relationship between and are easier to ensure. The explanation is as follows.

The bending angle is related to , , , , and . When building a traditional BPNN to predict the bending angle, , , , , and need to be set as the input parameters, and needs to be set as the unique output parameter. is roughly distributed from 90 to 130 degrees, and this large range requires the traditional BPNN to have enough precision. A precise training of the traditional BPNN is very difficult to achieve. However, the BPNN in BPNN-Spline can be easily trained to a precise degree in comparison with the traditional BPNN. When the BPNN in the BPNN-Spline is built, , , , and are set as inputs, and five bending angles corresponding to each certain punch displacement are set as outputs, respectively. Four variable inputs of the BPNN in BPNN-Spline are fewer, and the range of bending angle is also smaller.

As far as we know, the monotonic decrease relationship between and is inevitable in bending process when other parameters are determined. This monotonic relationship is very important for prediction accuracy. Due to the characteristics of network structure, traditional BPNN is not able to ensure the monotonic relationship between and . But the smooth spline function in BPNN-Spline is easier to ensure this monotonic relationship in comparison with the traditional BPNN, as shown in Figure 10. As a consequence, the spline function in BPNN-Spline is more likely to obtain the accurate prediction results in bending process.

5. Application Examples and Discussion

The BPNN-Spline model is suitable for bending angle prediction in the case:(a)If and , the applicable variable space is , , , and .(b)If and , the length ratio factor is . The applicable variable space is , , , and .

Six application examples under diverse material, process parameter, and length ratio factor are randomly selected to determine the prediction accuracy of BPNN-Spline. The predicted bending angles are compared with FE simulation results, as shown in Figures 11(a), 11(b), 11(c), 11(d), 11(e), and 11(f).

Figure 11: Comparison of FE and BPNN-Spline: (a) B340LA with  mm,  mm,  mm,  mm; (b) SUS304 with  mm,  mm,  mm,  mm; (c) DP600 with  mm,  mm,  mm,  mm; (d) DC06 with  mm,  mm,  mm,  mm; (e) B340LA with  mm,  mm,  mm,  mm; and (f) 6111-T4 with  mm,  mm,  mm,  mm.

In Figure 11, it can be observed that the predicted bending angle of BPNN-Spline is in good agreement with FE simulation results for diverse material, nominal width of die notch, die corner radius, punch corner radius, sheet thickness, and punch displacement. The bending angle deviations between BPNN-Spline and FE are so small that the biggest one is no more than 1.2 degrees, which is at the place where  mm in Figure 11(c). Therefore, it can be concluded that the BPNN-Spline model has enough accuracy to predict the bending angle under diverse bending condition.

In order to determine the prediction accuracy of BPNN-Spline in actual bending experiment aspect, two kinds of sheet are randomly selected to carry out the bending experiment. The results are compared with BPNN-Spline predicted results under the same process conditions. The sheets of steel material B340LA with 1.4 mm thickness and DP600 with 1.8 mm thickness are, respectively, used to carry out the experiment on the bending machine. The experimental mould and other process parameters are the same as the experiments of Section 3.4.

The comparisons between experiential and BPNN-Spline predicted bending angles are shown in Figures 12(a) and 12(b), respectively.

Figure 12: Comparison of BPNN-Spline and experiment: (a) B340LA with 1.4 mm thickness; (b) DP600 with 1.8 mm thickness.

As can be seen from Figure 12, predicted bending angles by BPNN-Spline are consistent with the experiment results. The largest bending angle deviation between BPNN-Spline and experiment for B340LA is no more than 1.4 degrees, which is at the place where  mm. The largest bending angle deviation for DP600 is no more than 1 degree, which is at the place where  mm. The prediction precision of BPNN-Spline is verified by actual bending experiments. In conclusion, the BPNN-Spline model can accurately predict the bending angle in actual production process. The technical requirement on BPNN-Spline model is lower compared to FE simulation. The user only needs to take the model as a black box and input the material and bending process parameters into the model; the corresponding bending angle will be deduced out immediately. It is very convenient and efficient.

The bending machine numerical control system should accurately and rapidly deduce out the bending angle for diverse process parameters. Therefore, BPNN-Spline combined model could be integrated properly into the system to control the bending process.

6. Conclusions

A bending springback angle prediction model based on the combination of BPNN and spline is presented in this study. Firstly, through FE and orthogonal experimental design, the training sample set is obtained. After that, the mathematical model of BPNN-Spline is established, and its advantage for bending angle prediction is discussed in comparison with traditional BPNN. Finally, through application examples, the BPNN-Spline model is used to predict the bending angle and shows higher prediction accuracy. The following conclusions are summarized:(1)The bending angle is unrelated with length ratio factor .(2)BPNN-Spline combined model is better than traditional BPNN on bending angle prediction. This is because the BPNN-Spline combined model has special network structure and smooth spline function, which will bring about a great advantage for the bending angle prediction.(3)The BPNN-Spline combined model shows better prediction accuracy on different material, thickness, nominal width of die notch, and other process parameters by the test of application examples; the general and precise characteristics are fully revealed.(4)The BPNN-Spline combined model not only has better applicable ability, but also has higher prediction accuracy.

Although the BPNN-Spline model has a good performance under the case of , it still cannot be applied in the case of . Future study should develop a model which can predict the bending angle accurately with respect to the case of .

Competing Interests

The authors declare that they have no competing interests.

Acknowledgments

The research is continuously funded by the National Science and Technology Major Project of China (2012ZX0401031).

References

  1. D.-K. Leu, “Position deviation and springback in V-die bending process with asymmetric dies,” International Journal of Advanced Manufacturing Technology, vol. 79, no. 5–8, pp. 1095–1108, 2015. View at Publisher · View at Google Scholar · View at Scopus
  2. S. V. Mohammadi, M. Parsa, and M. Parsa, “Simplified springback prediction in Al/PP/Al sandwich air bending,” Journal of Sandwich Structures and Materials, vol. 17, no. 3, pp. 217–237, 2015. View at Publisher · View at Google Scholar · View at Scopus
  3. L. J. De Vin, “Expecting the unexpected, a must for accurate brakeforming,” Journal of Materials Processing Technology, vol. 117, no. 1-2, pp. 244–248, 2001. View at Publisher · View at Google Scholar · View at Scopus
  4. Y. Zong, P. Liu, B. Guo, and D. Shan, “Springback evaluation in hot v-bending of Ti-6Al-4V alloy sheets,” International Journal of Advanced Manufacturing Technology, vol. 76, no. 1–4, pp. 577–585, 2015. View at Publisher · View at Google Scholar · View at Scopus
  5. V. Vorkov, R. Aerens, D. Vandepitte, and J. R. Duflou, “Springback prediction of high-strength steels in large radius air bending using finite element modeling approach,” in Proceedings of the 11th International Conference on Technology of Plasticity (ICTP '14), pp. 1005–1010, Nagoya, Japan, October 2014. View at Publisher · View at Google Scholar · View at Scopus
  6. Z. M. Fu, W. Chen, X. L. Tian, and B. K. Hu, “Modeling and simulation for multiple-step incremental air-bending forming of sheet metal,” International Journal of Advanced Manufacturing Technology, vol. 72, no. 5–8, pp. 561–570, 2014. View at Publisher · View at Google Scholar · View at Scopus
  7. M. R. Jamli, A. K. Ariffin, and D. A. Wahab, “Incorporating feedforward neural network within finite element analysis for L-bending springback prediction,” Expert Systems with Applications, vol. 42, no. 5, pp. 2604–2614, 2015. View at Publisher · View at Google Scholar · View at Scopus
  8. M. Haddadzadeh, M. R. Razfar, and M. R. M. Mamaghani, “Novel approach to initial blank design in deep drawing using artificial neural network,” Proceedings of the Institution of Mechanical Engineers, Part B: Journal of Engineering Manufacture, vol. 223, no. 10, pp. 1323–1330, 2009. View at Publisher · View at Google Scholar · View at Scopus
  9. H. Hasanzadehshooiili, A. Lakirouhani, and A. Šapalas, “Neural network prediction of buckling load of steel arch-shells,” Archives of Civil and Mechanical Engineering, vol. 12, no. 4, pp. 477–484, 2012. View at Publisher · View at Google Scholar · View at Scopus
  10. K. Hiramoto, “Control system design of mechanical systems with time varying mechanical and control parameters,” Journal of Advanced Mechanical Design, Systems, and Manufacturing, vol. 8, no. 3, pp. 1–14, 2014. View at Publisher · View at Google Scholar
  11. J. Xu, K. Yamada, K. Seikiya, R. Tanaka, and Y. Yamane, “Comparison of applying static and dynamic features for drill wear prediction,” Journal of Advanced Mechanical Design, Systems and Manufacturing, vol. 8, no. 4, p. JAMDSM0056, 2014. View at Publisher · View at Google Scholar · View at Scopus
  12. S. Kitayama, S. Huang, and K. Yamazaki, “Optimization of variable blank holder force trajectory for springback reduction via sequential approximate optimization with radial basis function network,” Structural and Multidisciplinary Optimization, vol. 47, no. 2, pp. 289–300, 2013. View at Publisher · View at Google Scholar · View at Scopus
  13. J. Srirat, S. Kitayama, and K. Yamazaki, “Simultaneous optimization of variable blank holder force trajectory and tools motion in deep drawing via sequential approximate optimization,” Journal of Advanced Mechanical Design, Systems and Manufacturing, vol. 6, no. 7, pp. 1081–1092, 2012. View at Publisher · View at Google Scholar · View at Scopus
  14. J.-H. Song, H. Huh, and S.-H. Kim, “Stress-based springback reduction of a channel shaped auto-body part with high-strength steel using response surface methodology,” Journal of Engineering Materials and Technology, Transactions of the ASME, vol. 129, no. 3, pp. 397–406, 2007. View at Publisher · View at Google Scholar · View at Scopus
  15. X.-Y. Guo, D. Li, Z. Wu, and Q.-H. Tian, “Application of response surface methodology in optimizaing the sulfation-roasting-leaching process of nickel laterite,” International Journal of Minerals, Metallurgy and Materials, vol. 19, no. 3, pp. 199–204, 2012. View at Publisher · View at Google Scholar · View at Scopus
  16. C. Shen, L. Wang, and Q. Li, “Optimization of injection molding process parameters using combination of artificial neural network and genetic algorithm method,” Journal of Materials Processing Technology, vol. 183, no. 2-3, pp. 412–418, 2007. View at Publisher · View at Google Scholar · View at Scopus
  17. Y. Rostamiyan, A. Seidanloo, H. Sohrabpoor, and R. Teimouri, “Experimental studies on ultrasonically assisted friction stir spot welding of AA6061,” Archives of Civil and Mechanical Engineering, vol. 15, no. 2, pp. 335–346, 2015. View at Publisher · View at Google Scholar · View at Scopus
  18. M. Winnicki, A. Małachowska, and A. Ambroziak, “Taguchi optimization of the thickness of a coating deposited by LPCS,” Archives of Civil and Mechanical Engineering, vol. 14, no. 4, pp. 561–568, 2014. View at Publisher · View at Google Scholar · View at Scopus
  19. Y. Tzeng, F. Chen, and C. Chen, “A hybrid approach for multi-objective design optimization of ball grid array gold wire bonding process,” Journal of Advanced Mechanical Design, Systems, and Manufacturing, vol. 9, no. 2, p. JAMDSM0021, 2015. View at Publisher · View at Google Scholar
  20. M. Piffl and E. Stadlober, “The depth-design: an efficient generation of high dimensional computer experiments,” Journal of Statistical Planning and Inference, vol. 164, pp. 10–26, 2015. View at Publisher · View at Google Scholar · View at Zentralblatt MATH · View at MathSciNet · View at Scopus