Transferability of Temperature Evolution of Dissimilar Wire-Arc Additively Manufactured Components by Machine...

Area of Science:

• Materials Science and Engineering
• Manufacturing Processes
• Computational Modeling

Background:

• Wire-arc additive manufacturing (WAAM) offers significant industrial potential but requires optimization to mitigate temperature gradient-induced stresses and defects.
• Real-time process control is crucial for consistent WAAM outcomes, necessitating data-driven predictive methods.

Purpose of the Study:

• To develop and evaluate machine learning (ML) models for predicting temperature history in WAAM-produced aluminum components.
• To assess the transferability of ML models across varying geometries and process parameters, including bar length, deposition layers, and heat source speed.

Main Methods:

• Utilized finite element (FE) simulations to generate training and testing datasets for ML models.
• Developed multilayer perceptron (MLP) models to predict temperature evolution during the WAAM process.
• Evaluated model performance using mean absolute percentage error (MAPE) across different simulation scenarios.

Main Results:

• Baseline MLP models achieved high accuracy (MAPE < 0.7%) for training and testing data.
• Model performance showed good transferability to increased bar length or layer count (MAPE < 3.22%), though with increased variability.
• Predicting temperature history for varied scanning speeds resulted in lower accuracy, with some models exhibiting higher MAPE (up to 14.91%).

Conclusions:

• Simple MLP models demonstrate effective temperature history prediction for WAAM, highlighting the potential for data-driven process optimization.
• The study confirms the transferability of ML-based temperature prediction in WAAM, offering a pathway to control residual stresses and microstructural defects.