Abstract: Based on acquisition and processing of audible arc sound signal in the process of MIC welding,an online test method of penetration status was proposed.Arc sound signal un-der typical penetration status in the flat butt welding process with spray transfer was acquired and analyzed in real time.11 charac-teristic parameters were extracted to characterize welding penetra-tion status by wavelet denoising and short-time windowing tech-nology.PCA parameter was synchronized based on high-dimen-sion characteristic vector,8 dimensions characteristic vector with most information of penetration status was re-synthesized and taken as input parameters,four penetratioa statuses were taken as export parameters,and network models of BP and RBF to i-dentify the penetration status were established.The application of test model shows that the two networks can realize the online recognition of penetration status.The accuracy of RBF network reaches 91.25%,which is 6.25% more than that of BP.