Machine Learning in Injection Molding: An Industry 4.0 Method of Quality Prediction

Area of Science:

• Manufacturing Engineering
• Materials Science
• Computer Science

Background:

• Ensuring stable part quality in injection molding is critical but challenging due to complex, interrelated processing conditions.
• Machine learning (ML) offers a powerful approach to navigate these complexities by analyzing data in multidimensional spaces.

Purpose of the Study:

• To evaluate the effectiveness of four ML algorithms in predicting the quality of multi-cavity injection molding.
• To compare the performance of kNN, naïve Bayes, linear discriminant analysis, and decision tree algorithms for this application.

Main Methods:

• Utilized pressure-based quality indexes as input features for classification.
• Trained and compared four distinct ML algorithms: kNN, naïve Bayes, linear discriminant analysis, and decision tree.
• Assessed algorithm performance, including accuracy and computational time, with limited training data.

Main Results:

• All examined ML algorithms demonstrated adequate quality prediction capabilities, even with very little training data.
• The decision tree algorithm exhibited the highest accuracy, exceeding 90% on average.
• The decision tree algorithm achieved this high performance with a computational time of only 8-10 seconds.
• Feature selection was found to have no significant impact on the decision tree algorithm's accuracy.

Conclusions:

• Machine learning algorithms provide a viable solution for predicting and ensuring stable quality in injection molding.
• The decision tree algorithm is particularly effective, offering high accuracy and computational efficiency for quality prediction.
• The robustness of these ML models, even with limited data, highlights their practical applicability in manufacturing settings.