Big Data Handling and Dimensionality Reduction for In-Situ Process Monitoring in Additive Manufacturing

In-situ sensing and monitoring in metal Additive Manufacturing (AM) has attracted a wide interest in the literature and a continuously increasing adoption in industry to support part and process qualification. Despite the several advantages enabled by a quick and in-line detection of errors and anomalies, various challenges still have to be faced. One of them regards the capability of making sense, through automated and real-time algorithms, of actual “big data” acquired on a layer-by-layer basis, where the term “big” refers to the volume of data, but also to their sampling frequency and variety. Indeed, high spatial and/or temporal resolution image and video image data are needed to capture the complex physical phenomena involved in the process and to characterize local and transient anomalies that may indicate the presence of defects in the part. This imposes the need for novel methods suitable to extract the relevant information content from big and fast data streams in an efficient and effective way. This study investigates the use of the Ripley’s K-function as a tool for dimensionality reduction of video image data acquired in Laser Powder Bed Fusion (L-PBF). The aim is to synthesize the spatial pattern of process by-products in L-PBF by passing from a high-speed stream of 2-D images to a data representation in the form of 1-D curves. K-functions are here proposed to describe the spatial spread of spatters generated by the laser-material interaction, and to analyze the influence of process parameters on the spatter behavior. Indeed, spatters can be used as driver of information about the process stability but also the quality of the manufactured part. A real case study is presented to demonstrate the effectiveness of the proposed approach in identifying how the spatter behavior is affected by process parameters and linked to the final quality of the part.