Abstract: In the process of grinding processing, grinding heat results in metal surface burn on parts, leading to reduced strength, increased residual stress, and other performance changes with significant security risks. Acoustic Emission (AE) technology is widely employed for monitoring grinding burn in machine tools due to its sensitivity to weak signals and ability to detect high-frequency signals. However, the accuracy of grinding burn monitoring is not ideal due to strong interference signals such as metal particle spalling and friction during the grinding process. Meanwhile, the characteristics of metal burn signal is unknown. To improve the accuracy of burn signal recognition, it is necessary to investigate the characteristics of metal burn AE signals independently. Therefore, a laser experiment was designed and conducted to simulate the grinding burn process in order to obtain pure metal burn signals. By adjusting laser energy levels, four degrees of surface metal burns were generated. It was observed that as the metal burn degree increased, the spectrum of AE signals moved to higher frequency bandwidth. Based on this phenomenon, Wavelet packet transform (WPT) was utilized to extract dominant frequency bandwidths of metal burn signals. The extracted feature vectors were then input into Principal Component Analysis (PCA) algorithm for clustering and identification metal burn signals.