Abstract: Aiming at the problem of back-side penetration in aluminum alloy pulsed MIG welding, a method for modeling and control of back-side penetration was proposed based on the experience of human welders. A random variable waveform of welding speed was designed to make the weld pool geometry change. A passive visual sensing system was used to detect the of back-side image of molten pool. The welder observed the molten pool and adjusted the welding current to keep the molten pool width consistent. At the same time, the molten pool width was extracted by developing image processing algorithms. A nonlinear Hammerstein model between adjusted current and back-side width was established by a least square system identification method. Current and welding speed disturbance experiments were carried out. The results showed that the nonlinear Hammerstein model based on human welder experience can control back-side penetration effectively, and it has a strong anti-interference ability.