Control of weld penetration is currently one of the most important and crucial research issues in the area of welding. The weld pool can provide accurate and instantaneous information about the weld penetration, however, the establishment and confirmation of the correlation between weld pool and weld penetration require numerous accurate measurements and suitable geometrical modeling of weld pool. A normalized model is proposed to characterize the weld pool two-dimensionally. More than 6,000 weld pools are measured from experiments using a developed real-time weld pool sensing system. A data analysis shows that the weld penetration is correlated with the weld pool which is specified by the three characteristic parameters proposed in the study. However, the correlation is nonlinear. To approximate the complicated nonlinearity, neural networks are used. Comparative modeling trails show that the weld penetration can be more accurately calculated if the adjacent weld pools are also used. This implies that the correlation between the weld penetration and weld pool is dynamic. Hence, an on-line nonlinear dynamic estimation system is developed to estimate the weld penetration.
Automatic penetration bead welding technology in horizontal position using weld pool image recognition