Wednesday 19 March 2025
Scientists have made a significant breakthrough in developing more accurate and efficient methods for predicting temperature distributions during laser powder bed fusion, a crucial process in additive manufacturing.
Laser powder bed fusion is a complex process that involves melting and solidifying metal powders layer by layer to create three-dimensional objects. However, the high temperatures involved can lead to inconsistent material properties and defects, making it challenging to achieve precise control over the final product.
To address this challenge, researchers have turned to physics-informed neural networks (PINNs), a type of artificial intelligence that can learn from data and make predictions about complex systems. PINNs have been successfully applied in various fields, including fluid dynamics and heat transfer.
In this study, scientists used PINNs to predict the temperature distribution during laser powder bed fusion. They developed two approaches: a single-track model called PI-EnDeepONet and a multi-track model called sequential PINN.
The single-track model was able to accurately predict temperature distributions for simple scenarios with a single track of metal powder being melted. However, when the researchers attempted to apply this model to more complex scenarios involving multiple tracks, it struggled to maintain accuracy due to the increased complexity of the problem.
To overcome this limitation, the scientists developed the sequential PINN approach. This method involves training individual PINNs for each track and then recalling them sequentially to predict the temperature distribution across multiple tracks.
The results showed that the sequential PINN approach was able to accurately predict temperature distributions for complex scenarios involving multiple tracks. The model was also able to capture the cooling phase, which is critical for understanding material properties and defects.
The scientists demonstrated the effectiveness of their method by comparing it with traditional finite-difference solutions. They found that the sequential PINN approach was not only more accurate but also significantly faster than traditional methods.
This breakthrough has significant implications for the additive manufacturing industry. By accurately predicting temperature distributions, manufacturers can optimize process parameters to achieve consistent material properties and reduce defects. This could lead to improved product quality and reduced production costs.
The development of physics-informed neural networks for laser powder bed fusion is a major step forward in the field of additive manufacturing. The sequential PINN approach has the potential to revolutionize the way manufacturers approach temperature prediction, enabling more efficient and effective production processes.
Cite this article: “Accurate Temperature Prediction in Laser Powder Bed Fusion with Physics-Informed Neural Networks”, The Science Archive, 2025.
Laser Powder Bed Fusion, Additive Manufacturing, Temperature Prediction, Physics-Informed Neural Networks, Pinns, Single-Track Model, Multi-Track Model, Sequential Pinn, Finite-Difference Solutions, Process Optimization.
