Abstract: The laser cladding technology is widely used in the field of material surface modification and remanufacturing due to its excellent performance. However, the rapid solidification process of laser cladding can easily lead to cracking of the cladding coating, resulting in the failure of the cladding coating. Therefore, aiming at the problem of on-line crack monitoring in the cladding process, this paper selected a non-contact small microphone to collect the acoustic signal of the laser cladding process with a cracking problem and analyzed the characteristic parameters of the cladding acoustic signal by wavelet packet decomposition and support vector machine. The results show that from the time domain, the acoustic signal of cracks in the cladding process has the characteristics of short trigger duration, high amplitude, rising relatively rapidly and exponential oscillation attenuation trend decline. From the frequency domain, it is a multi-peak burst signal. Secondly, the energy distribution of normal acoustic signal and crack acoustic signals is obviously different, in which the energy of the crack acoustic signal is mainly distributed in the（3, 1） frequency band. Simultaneously, support vector machine（SVM） classifier can identify the normal signal and crack signal, and the recognition accuracy is 100%, which can realize the effective recognition of laser cladding crack acoustic signal.