A machine learning approach to detect pores in laser powder bed fusion additive manufacturing

Real-time detection of pores in the Laser Powder Bed Fusion (LPBF) metal Additive Manufacturing (AM) process is proposed in this study and can be utilized for in-situ process monitoring and quality control. The average light emission data from the process captured by an optical tomography camera can be integrated into a defect detection module to characterize defects after the deposition of a layer. The light emission contains information on the process zone which could be extracted with the appropriate data techniques. In this paper, we proposed a machine-learning approach that utilizes the mean light intensity data from the melt-pool monitoring unit of a commercial LPBF system (EOS M290) to detect anomalies in the process. In the reported results we achieved an accuracy of 87 % for defect detection of implanted voids of varying sizes from 25 µm to 300 µm in Inconel 718 samples. The processing time of the machine learning models has been found to reach real-time processing time and could be used for the rapid qualification of AM components.