LAUSR

Poor quality and low repeatability of additively manufactured parts are key technological obstacles for the widespread adoption of additive manufacturing (AM). In-situ monitoring and control of the AM process is vital to overcome this problem. This paper describes the combined artificial intelligence and plasma emission spectroscopy to identify the porosity of AM parts during the process. The time- and position-synchronized spectra were collected during the directed energy deposition (DED) manufacturing process of a 7075-Al alloy part. Eighteen features extracted from spectra were coupled with the deposition qualities which were characterized by the 3D X-ray Computed Tomography (CT) scan and used to train a Random Forest (RF) classifier. The well-trained RF classifier achieved up to 83% precision for the porosity recognition of depositions. The feature importance recorded by the RF classifier indicates that the intensities of spectra at the wavelength of 414.234 (Fe I) nm and 396.054 (Al I) nm, and the kurtosis of spectra at wavelength ranges of 484–490 nm and 508–518 nm, are the most effective features for porosity recognition. The physical correlations between spectra, porosity formation, and thermal accumulation during the AM process were analyzed. This study demonstrates the great potentials, as well as challenges of plasma emission spectroscopy for in-situ quality monitoring of laser AM which allows the enhancement of AM technique.