AI-Assisted Thermal Response Classification of PCB Materials for Energy-Efficient Femtosecond Laser Processing: A Simulation-Based Study

Özel Y., Vall M. M., Yıldırım M. A., Balcı H. Ş., Balcı F., Ilgın H. A.

Gazi University Journal of Science Part A: Engineering and Innovation, cilt.13, ss.1-19, 2026 (TRDizin)

Özet

The thermal behavior of printed circuit board (PCB) materials under femtosecond laser irradiation varies significantly due to material-dependent thermal and optical properties. This variation may lead to unnecessary energy delivery or insufficient processing effects when a fixed fluence is applied. In this study, the femtosecond laser-induced thermal responses of Aluminum, Copper, FR-4, Graphene, PTFE, and Titanium were modeled using ANSYS-based simulations, and the resulting temperature distributions were converted into two-dimensional thermal images. For thermal image classification, a Cross-Attention Fusion-based DualViT-CNN architecture was proposed, integrating the local feature extraction capability of CNNs with the global context learning ability of Vision Transformers. An ArcFace-based angular margin classifier was employed to enhance class separability among materials exhibiting similar thermal patterns. The proposed model achieved an accuracy of 95.95% and an F1-score of 95.94% on the original dataset, and an accuracy of 99.18% and an F1-score of 99.17% on the augmented dataset. Class prediction-based fluence selection demonstrated an approximately 47% theoretical energy reduction potential compared with the fixed reference fluence approach of 0.40 J/cm². The findings indicate that the proposed simulation-assisted artificial intelligence framework provides a feasible preliminary decision-support approach for thermal response classification and energy-efficient femtosecond laser processing of PCB materials.