Machine learning-assisted acoustic emission monitoring for track formability prediction of laser powder bed fusion

Abstract

Laser powder bed fusion (L-PBF) additive manufacturing offers many advantages in directly fabricating geometrically complex parts. To achieve the optimal L-PBF-built part quality, the laser process parameters should be tailored before building or real-time adjusted during building if necessary. Monitoring the formation quality of L-PBF-built metal tracks during the building process is a technical prerequisite. The work studies the acoustic emission (AE) monitoring and machine learning (ML) to predict the formation quality of as-built tracks. The AE signals from the L-PBF processes under different laser parameters are collected. A method of deep neural network-based denoising is employed to identify and eliminate the AE noise-hits. Signal segments of interest signals in the track melting-solidification stage are detected and extracted. The features for constructing the ML models are extracted by combining wavelet packet transform and self-organization map network. The random forest is used for formation quality prediction of the as-built tracks. The correlation between AE signals and as-built track quality is analyzed based on the similarity matrix of self-organization map. The results indicate that the as-built track quality can be monitored by AE technology and ML model. It contributes to that the L-PBF process stability and the as-built track quality reliability can be ensured by adjusting the process parameters in time during the L-PBF process.