A methodology for the automated analysis of metal powder scanning electron microscope (SEM) images towards material characterization is developed and presented. This software-based tool takes advantage of a combination of recent artificial intelligence advances (mask R-CNN), conventional image processing techniques, and SEM characterization domain knowledge to assess metal powder quality for additive manufacturing applications. SEM is being used for characterizing metal powder alloys, specifically by quantifying the diameter and number of spherical particles, which are key characteristics for assessing the quality of the analyzed powder. Usually, SEM images are manually analyzed using third-party analysis software, which can be time-consuming and often introduces user bias into the measurements. In addition, only a few non-statistically significant samples are taken into consideration for the material characterization. Thus, a method that can overcome the above challenges utilizing state-of-the-art instance segmentation models is introduced. The final proposed model achieved a total mask average precision (mAP50) 67.2 at an intersection over union of 0.5 and with prediction confidence threshold of 0.4. Finally, the predicted instance masks are further used to provide a statistical analysis that includes important metrics such as the particle size distinction.

Process reliability and quality output are critical indicators for the upscaling potential of a fabrication process on an industrial level. Fused filament fabrication (FFF) is a versatile additive manufacturing (AM) technology that provides viable and cost-effective solutions for prototyping applications and low-volume manufacturing of high-performance functional parts, yet is defect-prone due to the inherent aspect of parametrization. A systematic yet parametric workflow for quality inspection is therefore required. The work presented describes a versatile and reliable framework for automatic defect detection during the FFF process, enabled by artificial intelligence-based computer vision. Specifically, state-of-the-art deep learning models are developed for in-line inspection of individual thermoplastic strands’ surface morphology and weld quality, thus defining acceptable limits for FFF process parameter values. We examine the capabilities of an NVIDIA Jetson Nano, a low-power, high-performance computer with an integrated graphical processing unit (GPU). The developed deep learning models used in this analysis use a pre-trained model combined with manual configurations in order to efficiently identify the thermoplastic strands’ surface morphology. The proposed methodology aims to facilitate process parameter selection and the early identification of critical defects, toward an overall improvement in process reliability with reduced operator intervention.

Trends in the sciences are indicative of data management becoming established as a feature of the mainstream research process. In this context, the European Commission introduced an Open Research Data pilot at the start of the Horizon 2020 research programme. This initiative followed the success of the Open Access pilot implemented in the prior (FP7) research programme, which thereafter became an integral component of Horizon 2020. While the Open Access phenomenon can reasonably be argued to be one of many instances of web technologies disrupting established business models (namely publication practices and workflows established over several centuries in the case of Open Access), initiatives designed to promote research data management have no established foundation on which to build. For Open Data to become a reality and, more importantly, to contribute to the scientific process, data management best practices and workflows are required. Furthermore, with the scientific community having operated to good effect in the absence of data management, there is a need to demonstrate the merits of data management. This circumstance is complicated by the lack of the necessary ICT infrastructures, especially interoperability standards, required to facilitate the seamless transfer, aggregation and analysis of research data. Any activity aiming to promote Open Data thus needs to overcome a number of cultural and technological challenges. It is in this context that this paper examines the data management activities and outcomes of a number of projects participating in the Horizon 2020 Open Research Data pilot. The result has been to identify a number of commonly encountered benefits and issues; to assess the utilisation of data management plans; and through the close examination of specific cases, to gain insights into obstacles to data management and potential solutions. Although primarily anecdotal and difficult to quantify, the experiences reported in this paper tend to favour developing data management best practices rather than doggedly pursue the Open Data mantra. While Open Data may prove valuable in certain circumstances, there is good reason to claim that managed access to scientific data of high inherent intellectual and financial value will prove more effective in driving knowledge discovery and innovation.