Physics-informed and data-driven thermal modeling and temperature field prediction for thermal field-assisted direct ink writing with PDMS
Thermal field assistance enables in situ heating, improving the curing kinetics and printability of thermosetting polymers. However, it alters the temperature distribution during forming, affecting printing quality. Accurate temperature field prediction is essential for process optimization. Althoug...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-08-01
|
| Series: | Materials & Design |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S026412752500704X |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Thermal field assistance enables in situ heating, improving the curing kinetics and printability of thermosetting polymers. However, it alters the temperature distribution during forming, affecting printing quality. Accurate temperature field prediction is essential for process optimization. Although numerical simulations can capture temperature evolution in additive manufacturing (AM), they are time-consuming and storage-intensive, limiting their efficiency for real-time applications. To address this limitation, developing a rapid temperature prediction surrogate model based on deep learning becomes essential. In this study, a thermal simulation framework was developed specifically for thermal field-assisted AM (TFAM). Based on this framework, a surrogate model integrating physics-based and data-driven approaches was constructed and trained using multi-condition simulation datasets. To validate the simulation accuracy and evaluate the surrogate model, an experimental platform for thermosetting polymers was also established. The results showed that the average relative error between simulation and experiments was below 3 %. The proposed surrogate model outperformed other data-driven methods, achieving an average R2 above 0.99, with maximum RMSE and MAE of 0.3314 °C and 0.3174 °C. Notably, it maintained RMSE below 1 °C with only 10 % of the training data. Additionally, it reduced prediction time to seconds and storage to megabytes, significantly exceeding the efficiency of traditional simulations. |
|---|---|
| ISSN: | 0264-1275 |