A tensor-based hierarchical process monitoring approach for anomaly detection in additive manufacturing

Additive manufacturing (AM) is a technology that enables the creation of complex shapes with advanced structural and functional properties. It has transformed the traditional manufacturing operations into a more flexible and efficient process, reshaping the whole value chain and allowing new levels of product customization. AM is a layer-by-layer manufacturing process, in which materials are deposited in each layer to create the object of interest. Due to the layer-wise nature of the process, anomalies and defects might occur within each layer, across several layers or throughout the whole sample. An accurate and responsive detection strategy that enables the detection of various types of anomalies is essential for ensuring the quality and integrity of the manufactured product. In this paper, a hierarchical in situ process monitoring approach, namely, a three level monitoring strategy, is proposed to detect local, layer-wise, and sample-wise anomalies using thermal videos acquired during the manufacturing process. The proposed approach integrates hierarchical low-rank tensor decomposition methods with statistical monitoring techniques to effectively detect anomalies at different levels, namely, the within-layer level, the layer level, and the sample level. Simulations are used to evaluate the performance of the method and compare with existing benchmarks. The proposed approach is also applied to thermal videos acquired during the laser powder bed fusion (L-PBF) process to illustrate its effectiveness in practice.