Detection of keyhole pore formations in laser powder-bed fusion using acoustic process monitoring measurements

Abstract

In-situ process monitoring of additively manufactured parts has become a topic of increasing interest to the manufacturing community. In this work, acoustic measurements recorded during laser powder-bed fusion (L-PBF) were used to detect the onset of keyhole pores induced by the lasing process. Post-build radiography was used to identify the locations of keyhole pores in the build. The pore locations were spatially and temporally registered with the recorded time-series of laser position and acoustic pressure to identify specific partitions of the acoustic signals which correspond to pore formation. Ensemble empirical mode decomposition, traditional Fourier decomposition, and statistical measures of the time-series and corresponding frequency spectra were used to extract feature vectors which correlate to keyhole pore formation. Sequential feature selection revealed that measures associated with the acoustic spectra were most useful for identifying pore formations in L-PBF. A subset of the most informative data features was used to train a support vector machine model to predict pore formation with up to 97% accuracy.