32 Prediction of Width and Thickness of Injection Molded …
457
which became a problem for the whole world. Keeping process parameters under
control and beforehand prediction of quality of the plastic products can decrease
amounts of produced plastic scrap and energy consumption, contributing to the more
environmentally conscious use of the plastic materials.
This paper uses data from 160 machine runs during which 47 machine and process
parameters were logged while producing dogbone specimens type A defined by
ISO 527–2 (ISO 2012). REPTree (decision tree), random forest, k-nearest neighbors
(kNN) and multilayered perceptron (MLP) machine learning algorithms were applied
to create models for prediction of width and thickness of the focus parts. The results
are discussed in terms of the ability to accurately predict the part’s quality and
interpretability of the chosen methods.
32.2 Literature Review
In recent years new methods for quality prediction and optimization of the injection
molding process have been proposed and applied. Some of the approaches include
the application of machine learning methods (Chen 2008; Manjunath and Krishna
2012; Kuo et al. 2007; Lotti et al. 2002), the use of simulation approaches (Berti and
Monti 2013; Fu and Ma 2018) and the development of particular hardware solutions
(Panchal and Kazmer 2010; Johnston 2015).
Machine learning methods
Machine learning methods (ANN, genetic algorithm, self-organizing maps, etc.)
have been used for the development of prediction models of parts shrinkage, general
parts quality, to find solutions for multi-objective optimization problem of the injection molding process parameters, etc. Different researchers have used numbers of
samples ranging from 27 to 1000, some of them were collected through the simulation
software, others during laboratory experiments.
For example, in (Chen 2008) the authors used a combination of a self-organizing
map and a back-propagation neural network to create a dynamic quality predictor for
the injection molding process. Nine process parameters were included in the model
to predict the weight of the final part. To enhance the performance of the neural
network, Taguchi’s parameter design was also utilized. The dataset included 160
samples of experimental data. Manjunath and Krishna (Manjunath and Krishna 2012)
applied forward and reverse mapping ANN to predict dimensional shrinkage of the
produced part and appropriate set of process parameters to reach the required dimensional shrinkage correspondingly. The networks were trained using 1000 samples
generated in the simulation software using equations reported by other researchers.
In (Ogorodnyk et al. 2018) the multi-layered perceptron artificial neural network
model and J48 decision trees algorithm is used to create models for prediction of
injection molded parts quality. Data from 160 machine runs is utilized to train the
models.
457
which became a problem for the whole world. Keeping process parameters under
control and beforehand prediction of quality of the plastic products can decrease
amounts of produced plastic scrap and energy consumption, contributing to the more
environmentally conscious use of the plastic materials.
This paper uses data from 160 machine runs during which 47 machine and process
parameters were logged while producing dogbone specimens type A defined by
ISO 527–2 (ISO 2012). REPTree (decision tree), random forest, k-nearest neighbors
(kNN) and multilayered perceptron (MLP) machine learning algorithms were applied
to create models for prediction of width and thickness of the focus parts. The results
are discussed in terms of the ability to accurately predict the part’s quality and
interpretability of the chosen methods.
32.2 Literature Review
In recent years new methods for quality prediction and optimization of the injection
molding process have been proposed and applied. Some of the approaches include
the application of machine learning methods (Chen 2008; Manjunath and Krishna
2012; Kuo et al. 2007; Lotti et al. 2002), the use of simulation approaches (Berti and
Monti 2013; Fu and Ma 2018) and the development of particular hardware solutions
(Panchal and Kazmer 2010; Johnston 2015).
Machine learning methods
Machine learning methods (ANN, genetic algorithm, self-organizing maps, etc.)
have been used for the development of prediction models of parts shrinkage, general
parts quality, to find solutions for multi-objective optimization problem of the injection molding process parameters, etc. Different researchers have used numbers of
samples ranging from 27 to 1000, some of them were collected through the simulation
software, others during laboratory experiments.
For example, in (Chen 2008) the authors used a combination of a self-organizing
map and a back-propagation neural network to create a dynamic quality predictor for
the injection molding process. Nine process parameters were included in the model
to predict the weight of the final part. To enhance the performance of the neural
network, Taguchi’s parameter design was also utilized. The dataset included 160
samples of experimental data. Manjunath and Krishna (Manjunath and Krishna 2012)
applied forward and reverse mapping ANN to predict dimensional shrinkage of the
produced part and appropriate set of process parameters to reach the required dimensional shrinkage correspondingly. The networks were trained using 1000 samples
generated in the simulation software using equations reported by other researchers.
In (Ogorodnyk et al. 2018) the multi-layered perceptron artificial neural network
model and J48 decision trees algorithm is used to create models for prediction of
injection molded parts quality. Data from 160 machine runs is utilized to train the
models.