Machine Learning for Predicting Mechanical Properties of 3D-Printed Polymers from Process Parameters: A Review.

Area of Science:

• Materials Science
• Mechanical Engineering
• Computer Science

Background:

• Polymer additive manufacturing (AM) is increasingly used for functional parts.
• Mechanical performance is highly sensitive to process parameters.
• Classical modeling methods offer insights but have limitations.

Purpose of the Study:

• To review recent advancements in machine learning (ML) for predicting mechanical properties of polymer AM parts.
• To assess the effectiveness of ML models in establishing process-property relationships.
• To identify current challenges and future directions in the field.

Main Methods:

• Survey of recent literature on ML applications in polymer AM.
• Analysis of ML models including artificial neural networks, tree-based ensembles, and support vector regression.
• Evaluation of prediction accuracy for mechanical properties like strength and modulus.

Main Results:

• ML models, particularly well-tuned ANNs, tree ensembles, and SVR, achieve prediction errors below 5-10% for strength and modulus.
• Data-driven surrogates demonstrate significant potential to reduce experimental trial-and-error in process optimization.
• The review highlights the growing success of ML in polymer AM.

Conclusions:

• ML techniques offer powerful tools for predicting mechanical properties in polymer AM.
• Further research is needed to address challenges like small datasets and limited coverage of non-quasi-static behaviors.
• Standardization of error metrics and expansion to fatigue and impact testing are crucial for future development.