The welding groove of ultra-narrow gap is deep and narrow

which is easy to cause the risk of arc climbing

resulting in poor fusion at the bottom side wall. Aiming at the problem that the fusion state of ultra-narrow gap welding with constrained arc by flux flake is difficult to be monitored on-line

a fusion state recognition method based on arc sound is proposed. The arc acoustic signal acquisition system for ultranarrow gap welding process is established

based on the analysis of arc acoustic generation mechanism and human ear identification behavior

the short-term energy

average amplitude

Moore loudness

Mel cepstrum coefficient and their first-order and second-order differences which effectively characterize the welding process are extracted. Based on particle swarm optimization least squares support vector machine（PSO-LSSVM）

the fusion state recognition model of ultra-narrow gap welding bottom sidewall is established. The results show that this method can realize the identification of three kinds of states:non fusion

critical fusion and good fusion

and the accuracy can reach 91.7%. It provides a new way for on-line monitoring of the fusion state of ultra-narrow gap welding.