Predicting Linear Dimensional Accuracy of Material Extrusion Parts in Dependence of Process Parameters Using Neural...

Area of Science:

• Additive Manufacturing
• Materials Science
• Artificial Intelligence

Background:

• Material extrusion (MEX) offers economical production of complex parts but suffers from unpredictable quality due to undefined process parameters.
• This uncertainty leads to conservative parameter settings, increasing print times and material waste.

Purpose of the Study:

• To develop an accurate predictive model for the linear dimensional accuracy of MEX-manufactured parts.
• To optimize process parameters for improved quality and reduced manufacturing inefficiencies.

Main Methods:

• Utilized a neural network (NN) with hyperparameter tuning via an evolutionary algorithm.
• Trained and tested the NN model on various material extrusion process parameter combinations.

Main Results:

• Achieved a mean absolute percentage error (MAPE) below 1.3% for X and Y dimensions and 2.3% for Z dimensions.
• Demonstrated superior prediction accuracy and robustness compared to multiple linear regression, even with untrained parameter ranges.

Conclusions:

• The developed NN model accurately predicts linear dimensional accuracy in MEX, addressing a critical gap in additive manufacturing.
• This predictive capability allows for optimized parameter settings, enhancing part quality and process efficiency.