A Non-Contact method for plasma state perception in laser shock peening using Acoustic-Visual signal fusion

Laser shock peening (LSP) is a typical multi-physics coupling manufacturing process. The use of multimodal information allows for a more comprehensive understanding of plasma state evolution. The spatiotemporal characteristics of the plasma during LSP of γ-TiAl alloy were investigated by simultaneously collecting acoustic signals and high-speed image data. The effects of different laser energies, probe angles, and distances on the acoustic signal were analyzed. The results show that the acoustic signal exhibits a typical ‘N-shaped wave’ characteristic and demonstrates spherical wave propagation. Additionally, the flight time decreases as the laser energy increases. After absorbing the laser energy, the plasma expands rapidly and then decays quickly. The decay rate near the water surface is faster, and the overall plasma lifetime is approximately 80 microseconds. Based on these data, a 2D-3D CNN deep learning model, integrating both acoustic and visual signals, was developed for plasma state perception. The decision-level fusion model achieved an accuracy of 94.6%, while the feature-level fusion model reached 91.9%, significantly outperforming the single-modal models (visual: 89.7%; acoustic: 85.0%). These results demonstrate that the fusion of acoustic and visual signals is highly effective for plasma state perception, offering a new approach for intelligent monitoring and quality control in LSP.