Abstract: Aiming at CO₂ arc welding process of short-circuit trans-fer,an on-line adaptive control system of welding current waveform is introduced.In this system,a neural networks model is set up for optimizing welding parameters and waveform control parameters,by which shot-cir-cuiting frequency and arc sound energy that indirectly indicates welding process stability and spatter number are made use of characteristic parameters of network import.At the same time,a double-closed circle wave formeontrnl algorithm is put forward,which abstains PID parameters by means of a neural network self-adjusted model.Experimental result shows,this control system is available to matching and optimizing welding parameter sand waveform contrnl parameters,and that,it can remarkably make the welding spatter decreased and metal transfer stability improved.