Application of KNN and ANN Metamodeling for RTM Filling Process Prediction

Area of Science:

• Composite Materials Engineering
• Computational Fluid Dynamics
• Artificial Intelligence in Manufacturing

Background:

• Resin transfer molding (RTM) process simulation is crucial for optimization but computationally intensive for large datasets.
• Existing simulation methods struggle with multi-physical and multi-scale complexities, limiting their application in data-rich scenarios.

Purpose of the Study:

• To develop predictive surrogate models for RTM process optimization.
• To establish input-output correlations for mold design using machine learning metamodels.
• To assess the feasibility of metamodeling for data-intensive composite manufacturing applications.

Main Methods:

• Implementation of K-nearest neighbors (KNN) metamodels.
• Development of artificial neural network (ANN) metamodels.
• Training and validation of metamodels using resin injection location and viscosity as inputs.
• Investigating the prediction of required vents and maximum injection pressure as outputs.

Main Results:

• Both KNN and ANN metamodels achieved high prediction accuracies for RTM process parameters.
• KNN metamodels exhibited prediction errors ranging from 5.0% to 15.7%.
• ANN metamodels demonstrated prediction errors between 6.7% and 17.5%.

Conclusions:

• Metamodeling presents a computationally efficient and accurate approach for RTM process optimization.
• The developed surrogate models effectively relate process inputs to outputs, aiding mold design.
• Metamodeling shows significant promise for data-intensive applications, including process digital twinning in composite manufacturing.