Predicting the Optimal Input Parameters for the Desired Print Quality Using Machine Learning.

Area of Science:

• Additive Manufacturing
• Machine Learning Applications
• Materials Science

Background:

• 3D printing offers precision and speed but suffers from common errors like stringing and overheating.
• Existing error monitoring methods often fall short in addressing these 3D printing challenges.
• Optimizing process parameters is crucial for mitigating defects and improving print quality.

Purpose of the Study:

• To leverage machine learning for identifying optimal 3D printing process parameters.
• To develop an algorithm capable of predicting ideal settings and detecting print errors.
• To minimize material waste caused by printing defects in the manufacturing industry.

Main Methods:

• Investigated optimal parameters including infill structure/density, materials (ABS, PLA, Nylon, PVA, PETG), wall/layer thickness, count, and temperature.
• Trained machine learning algorithms using four network architectures: CNN, Resnet152, MobileNet, and Inception V3.
• Implemented an error detection system designed to pause printing immediately upon identifying a defect.

Main Results:

• The machine learning algorithm successfully predicted optimal 3D printing parameters for specific requirements.
• The algorithm demonstrated capability in detecting various print errors.
• The Inception V3 network architecture achieved the highest accuracy at 97% in error detection and parameter prediction.
• The system effectively paused prints when errors were detected, preventing further material waste.

Conclusions:

• Machine learning, particularly with the Inception V3 architecture, offers a highly accurate solution for optimizing 3D printing processes.
• The developed algorithm can predict optimal parameters and detect errors, significantly reducing material waste.
• This approach has substantial applications in the manufacturing industry for improving efficiency and sustainability in 3D printing.