Data-driven non-intrusive reduced order modelling of selective laser melting additive manufacturing process using...

Area of Science:

• Computational mechanics
• Materials science
• Machine learning

Background:

• Additive Manufacturing (AM) processes generate complex thermo-mechanical data.
• Accurate and efficient modeling of AM is crucial for quality control and process optimization.
• Traditional high-fidelity simulations are computationally expensive.

Purpose of the Study:

• To propose and compare two data-driven, non-intrusive reduced-order models (ROMs) for AM processes.
• To evaluate the accuracy and efficiency of POD-ANN and CAE-MLP models in predicting thermo-mechanical behavior.
• To validate model predictions against experimental data.

Main Methods:

• Developed a POD-ANN model combining proper orthogonal decomposition with artificial neural networks for dimensionality reduction and regression.
• Developed a CAE-MLP model using a 1D convolutional autoencoder for spatial dimension reduction and a multilayer perceptron for regression.
• Performed thermo-mechanical analysis of an AM-built part using both models.

Main Results:

• Both POD-ANN and CAE-MLP models showed strong correlation with high-fidelity simulation results.
• Model predictions exhibited good agreement with experimental data at various locations.
• The CAE-MLP model demonstrated superior prediction accuracy and performance over the POD-ANN model.

Conclusions:

• Data-driven ROMs, particularly CAE-MLP, offer a promising approach for efficient and accurate analysis of AM processes.
• Integrating reduced-order modeling with machine learning enhances the analysis of complex AM.
• The proposed models provide a robust framework for future research and applications in additive manufacturing.