Femtosecond Laser Micro-Nano Structuring and Prediction of Aluminum Bronze Qal9-4 Driven by Transformer

Abstract

To address multi-scale morphology manipulation challenges in aluminum bronze QAL9-4 femtosecond laser processing, we develop an architecture-optimized Transformer neural network model. Systematic multi-pulse experiments demonstrate energy accumulation-induced ablation threshold attenuation while quantitatively establishing the modulation laws of laser power and scanning speed on surface morphology. By strategically removing decoder modules and optimizing encoder architecture, our modified Transformer demonstrates higher prediction accuracy compared to conventional sequence models. Validation results demonstrate the architecture's superior performance in groove depth and width prediction tasks for laser micromachining. Our analysis quantifies laser power's predominant control over feature width (Spearman's ρ=0.73) and uncovers nonlinear parameter coupling mechanisms, establishing an optimization framework for QAL9-4 laser processing. The developed process parameter-geometry mapping framework enables controllable surface texturing for aerospace and microelectronics applications while expanding Transformer-based multi-physics modeling of laser-material interactions.

Keywords: femtosecond laser, Aluminum bronze QAL9-4, Micro-nano fabrication, Transformer model, Surface morphology prediction