458
O. Ogorodnyk et al.
At the same time, Kuo, Su (2007) applied Taguchi quality method to establish
the design of experiment for nine process parameters and analysis of variance to
define the most important factors influencing the production process, while backpropagation neural network model was trained to tune the optimum conditions
received from the Taguchi’s quality method. The model was created using 180 data
samples. In (Cheng et al. 2013) a methodology that includes variable complexity
methods, constrained non-dominated sorted genetic algorithm, back-propagation
neural network and MoldFlow analysis are applied to solve the multi-objective optimization problem of injection molding parameters. The data is generated using the
MoldFlow analysis software.
In (Liu 2017) the orthogonal experiment with Taguchi method was performed, and
ANOVA analysis was carried out to define which parameters are the most influential
for the injection molding process. In total 81 samples were obtained and used to train
models with the help of the back-propagation neural network and multi-class support
vector machine methods. Finally, the multi-objective optimization was performed on
the obtained models using the nondominated sorting genetic algorithm.
Lotti et al. (2002) used DOE and ANNs to create models for prediction of
shrinkage of iPP plaques based on 30 data samples collected. The neural network
model has shown better results in comparison to those calculated with the help of
the MoldFlow™ software package. In (Gao 2018) injection process parameters optimization procedure is proposed. A combination of the response surface methodology,
artificial neural networks, radial basis function, and Kriging surrogate is used.
Nagorny, Pillet (2017) collected thermal images of 204 rectangular specimens, as
well as signals of pressure and temperature sensors situated in the mold and used the
data to train models using support vector regression, random forest, k-nearest neighbors, stochastic gradient descent, bagging decision tree, Ada boosting decision tree,
as well as convolutional neural network and long-short term memory network. Neural
networks models show better results in comparison to other regression methods for
prediction of parts quality. While in (Altan 2010) Taguchi method, design of experiments, analysis of variance were used to design the experiment and choose the most
influential process parameters. In total 27 samples of data were collected and used
in this study. ANN model was trained to predict the part’s shrinkage in injection
molding.
Simulation approaches
Another approach to optimization of the injection molding process is the use of
numerical simulations and different sorts of simulation software, such as in (Berti
and Monti 2013), for example. Here an approach based on a combination of numerical
simulations, response surface methodology, and stochastic simulations is proposed
to create a virtual prototyping environment to enable robust optimization of the IM
process. At the same time, in (Fu and Ma 2018) authors propose a method to simulate
the influence of the early part ejection on its final quality using the integration of
MoldFlow™ and Ansys™ on the contrary to the stand-alone molding simulation.
O. Ogorodnyk et al.
At the same time, Kuo, Su (2007) applied Taguchi quality method to establish
the design of experiment for nine process parameters and analysis of variance to
define the most important factors influencing the production process, while backpropagation neural network model was trained to tune the optimum conditions
received from the Taguchi’s quality method. The model was created using 180 data
samples. In (Cheng et al. 2013) a methodology that includes variable complexity
methods, constrained non-dominated sorted genetic algorithm, back-propagation
neural network and MoldFlow analysis are applied to solve the multi-objective optimization problem of injection molding parameters. The data is generated using the
MoldFlow analysis software.
In (Liu 2017) the orthogonal experiment with Taguchi method was performed, and
ANOVA analysis was carried out to define which parameters are the most influential
for the injection molding process. In total 81 samples were obtained and used to train
models with the help of the back-propagation neural network and multi-class support
vector machine methods. Finally, the multi-objective optimization was performed on
the obtained models using the nondominated sorting genetic algorithm.
Lotti et al. (2002) used DOE and ANNs to create models for prediction of
shrinkage of iPP plaques based on 30 data samples collected. The neural network
model has shown better results in comparison to those calculated with the help of
the MoldFlow™ software package. In (Gao 2018) injection process parameters optimization procedure is proposed. A combination of the response surface methodology,
artificial neural networks, radial basis function, and Kriging surrogate is used.
Nagorny, Pillet (2017) collected thermal images of 204 rectangular specimens, as
well as signals of pressure and temperature sensors situated in the mold and used the
data to train models using support vector regression, random forest, k-nearest neighbors, stochastic gradient descent, bagging decision tree, Ada boosting decision tree,
as well as convolutional neural network and long-short term memory network. Neural
networks models show better results in comparison to other regression methods for
prediction of parts quality. While in (Altan 2010) Taguchi method, design of experiments, analysis of variance were used to design the experiment and choose the most
influential process parameters. In total 27 samples of data were collected and used
in this study. ANN model was trained to predict the part’s shrinkage in injection
molding.
Simulation approaches
Another approach to optimization of the injection molding process is the use of
numerical simulations and different sorts of simulation software, such as in (Berti
and Monti 2013), for example. Here an approach based on a combination of numerical
simulations, response surface methodology, and stochastic simulations is proposed
to create a virtual prototyping environment to enable robust optimization of the IM
process. At the same time, in (Fu and Ma 2018) authors propose a method to simulate
the influence of the early part ejection on its final quality using the integration of
MoldFlow™ and Ansys™ on the contrary to the stand-alone molding simulation.