Abstract: Laser process parameters in selective laser melting (SLM) significantly affect the quality of formed parts. Most of the current process optimization paths are "blind touch" process tests based on the empirical formula of laser energy density, which is difficult to fully reflect the many environmental factors that affect the quality of the formed parts, and in the SLM process, the instability of the molten pool and the discontinuous laser irradiation occur from time to time. Therefore, the online dynamic control of process parameters and quality monitoring technology in the SLM process needs to be developed urgently to ensure the reliability of the manufacturing process and the stability of the quality of the formed parts. Through signal processing and neural network algorithm, the classification of acoustic emission signals in the SLM forming process of the melt channel with different laser process parameters is realized. In the noise reduction module, an intuitive evaluation method based on the correlation between sequences and the inverse hyperbolic tangent function is proposed to integrate the two algorithms of empirical mode decomposition and neural network. In the classification prediction module, the performance of the neural network designed based on empirical mode decomposition and wavelet packet transform for two acoustic signal feature extraction methods is compared. The feasibility of the neural network model to classify and predict the acoustic emission signals generated by different laser parameters has been verified, which can directly guide the online control optimization or optimization of the SLM process parameters, and lay the theoretical method foundation for the online monitoring system of the forming quality of the SLM process.

Key words: acoustic emission monitoring, machine learning, selective laser melting, wavelet packet transform, empirical mode decomposition