A statistical learning method for image-based monitoring of the plume signature in laser powder bed fusion

The industrial breakthrough of metal additive manufacturing processes mainly involves highly regulated sectors, e.g., aerospace and healthcare, where both part and process qualification are of paramount importance. Because of this, there is an increasing interest for in-situ monitoring tools able to detect process defects and unstable states since their onset stage during the process itself. In-situ measured quantities can be regarded as “signatures” of the process behaviour and proxies of the final part quality. This study relies on the idea that the by-products of laser powder bed fusion (LPBF) can be used as process signatures to design and implement statistical monitoring methods. In particular, this paper proposes a methodology to monitor the LPBF process via in-situ infrared (IR) video imaging of the plume formed by material evaporation and heating of the surrounding gas. The aspect of the plume naturally changes from one frame to another following the natural dynamics of the process: this yields a multimodal pattern of the plume descriptors that limits the effectiveness of traditional statistical monitoring techniques. To cope with this, a nonparametric control charting scheme is proposed, called K-chart, which allows adapting the alarm threshold to the dynamically varying patterns of the monitored data. A real case study in LPBF of zinc powder is presented to demonstrate the capability of detecting the onset of unstable conditions in the presence of a material that, despite being particularly interesting for biomedical applications, imposes quality challenges in LPBF because of its low melting and boiling points. A comparison analysis is presented to highlight the benefits provided by the proposed approach against competitor methods.