Advancing metal laser powder bed fusion through in-process monitoring and X-ray computed tomography metrology

In the evolving landscape of industrial innovation, the importance of smart solutions for sustainable advanced manufacturing is increasingly evident. Manufacturing sectors that embrace innovative digital strategies have demonstrated superior resilience and adaptability. Advanced manufacturing solutions, including additive manufacturing (AM) technologies and compr ehensive quality assurance systems, are essential to meet the urgent demands for environmental and social sustainability, agile production, and seamless redesign and reconfiguration of industrial processes. Among AM technologies, laser-based powder bed fusion of metals (PBF-LB/M) is increasingly used in various industries, such as biomedical, aerospace, and automotive, to produce high-value parts. Despite advancements, achieving the stringent quality standards required in several applications remains challenging. PBF-LB/M parts often exhibit flaws like internal porosities, which can significantly affect mechanical properties and other critical characteristics, undermining product functionality or service life. Although efforts to optimize process parameters, the variability in mechanical properties and geometrical characteristics persists, primarily due to defects that are hard to predict and whose causes are not yet fully understood. In recent years, PBF-LB/M in-process monitoring has gained significant attention. In particular, various sensors and data types are exploited to detect defect formation at its early onset, thus paving the way for real-time flaw mitigation. However, current limitations affecting in-process PBF-LB/M monitoring systems emphasize the relevance of measuring defects affecting components after the fabrication. In this context, X-ray computed tomography (CT) can be used to gather post-process data on internal porosities and to establish correlations with in-process outlier events. On this basis, considering that the relationships between in-process events and subsequent flaws are still not comprehensively understood, new research efforts are needed. In this thesis, several activities aiming at enhancing metal powder bed fusion through the development of metrology and in-process monitoring solutions were investigated. Specifically, metrological X-ray computed tomography was supported by novel methodologies for obtaining accurate reference data on internal defects, which can be successfully compared to in-process layer-wise outlier events. This aspect is key to achieving accurate and robust correlations between in-process and post-process datasets. Analytical and artificial intelligence approaches, such as those based on machine learning (ML), were implemented to emphasize hidden patterns behind in-process gathered signals, in order to develop predictive models for porosity formation. The correlation performances were evaluated with reference to the size of the predicted defects, as well as to the time needed to make a prediction, which is fundamental for enabling the implementation of the approach in a real-time scenario and mitigating defect formation. Although ML-based solutions hold promise, a frequent issue relies on the difficulty of achieving generalized models, resulting in marked performance drops when dealing with input data characterized by previously unseen distributions. Novel approaches were proposed to overcome this limitation, leading to more robust modelling methodologies exhibiting comparable performances when applied to parts fabricated under various process conditions. Furthermore, recent studies have focused on predicting defect size directly from in-process data. A novel reference object was therefore developed to improve CT porosity segmentation and measurements in PBF-LB/M parts, providing accurate data for the development and validation of innovative in-process monitoring systems.