Fig. 1. Schematic of experimental setup

Fig. 2. Details of each frame of acoustic signal

Fig. 3. Details of noise signal in laser welding

Fig. 4. Signal comparison before and after denoising under different welding penetration conditions. (a) Non penetration; (b) partial penetration; (c) full penetration

Fig. 5. Comparison of each frame of SPWVD in different welding penetration states

Fig. 6. Gray-level co-occurrence matrix (GLCM) features extraction from each frame of time-frequency diagram

Fig. 7. Structure of BPNN for welding condition classification

Fig. 8. Confusion matrix of classification result using BPNN model (0, 1 and 2 represent non penetration, partial penetration and full penetration states, respectively)

Fig. 9. Architecture of constructed CNN mode

Fig. 10. Variation of accuracy and loss value during training process of constructed CNN model

Fig. 11. Confusion matrix of classification results using CNN model (0, 1 and 2 represent non penetration, partial penetration and full penetration, respectively)