Additive manufacturing (AM) is rapidly expanding its role in high-value industries unlocking opportunities in new segments like the electronic market. Nonetheless, the qualification of novel processes and products remains a critical hurdle, especially in the presence of challenging materials. The increasing availability of rich, in-situ sensor data offers an opportunity to address this challenge by enabling data-driven quality assurance. Harnessing heterogeneous data streams—ranging from images and videos to acoustic and thermal signals—requires new approaches capable of interpreting complex patterns in real time and flagging anomalies as they emerge.

This work investigates advanced monitoring strategies for laser powder bed fusion (L-PBF) in the production of high-performance cooling structures directly on a Direct Bonded Copper (DBC) substrate made of a ceramic core sandwiched between two copper layers.

L-PBF on such multi-material substrates introduces unprecedented challenges in terms of process stability and product integrity, due to residual stress formation, sustrate warpage, unstable heat dissipations and ceramic layer cracking.

The study present a multi-sensor architecture to monitor the process during the fabrication of every layer and automatically detect deviations and defects. During the build, the process generates a characteristic acoustic footprint that evolves across the different phases of layer deposition and can be measured with a structure-bourne acoustic emission sensor. Within this signal space, cracks produce distinctive anomalous patterns that can be detected combining dimensionality-reduction and machine learning techniques designed to disentangle normal variability from true defect signatures, thereby improving the reliability of early crack detection.

High-resolution powder bed imaging and high-speed thermal video imaging provide complementary sources of information for two different purposes. One consists of verifying the substrate positioning before the build, to meet positional accuracy requirements. The second involves the detection of unstable thermal patterns within the layer, from one layer to another and from one substrate to another. The sudden detection of underired variations in the spatio-temporal heating and cooling history of the process is indeed of paramount importance to guarante consistent and repeatable quality characteristics, and to aid the tuning and optimization of the process towards a defect-free series production.