Abstract: Laser welding is accompanied with a strong acoustic signal,which contains a considerable amount of information on the welding process and indicates certain aspects of weld quality.The characteristics of weld acoustic signal during laser cladding were analyzed in details.The intensity and power spectrum components of acoustic signals were found to be a match for the weld depth.The dominant power spectra covered by the experimental set-up ranged from 2 to 10 kHz,where several obvious spectrum lines were detected.The amplitude of acoustic signals decreases and the acoustic power spectrum lines change from intensive and simple to separated and complex with the decrease of weld depth.A relation model between acoustic signal and weld depth was established by using artificial neural network,which took acoustic signals as sensing message.The built backpropagation(BP) network model provided practical accuracy of numerical detection.These results have an industrial impact on detecting depth of laser welding in real-time.