A D V A N C E D M A T E R I A L S & P R O C E S S E S | J U L Y / A U G U S T 2 0 2 1 1 7 of additive components. Currently, the qualification and certification of a production part is based on expensive and part-specific point-qualification datasets. As analternative approach, EWI, GE Research, Raytheon, and Youngstown State University showed proof of concept for feature-based qualification in an America Makes program entitled “Feature-Based Qualification Method for Directed Energy Deposition AM.” In this approach, feature specific datasets are used to train machine learning methods to predict microstructure and mechanical performance. This abstraction from part-specific to feature-specific qualification drastically extends the predictive power of a given dataset. The combination of techniques like these and the reusable nature of FAIR data will provide a paradigm change in qualification/certification. NIST & Data Hub: Figure 4 illustrates a fast AM qualification framework based on federated FAIR AM data. Five key components are identified to enable the fast AM qualification process including: a collaborative and federated AM data hub, advanced data analytics to leverage on heterogeneous datasets, integrated computational material engineering (ICME), a hybrid method to combine data-driven models with first principle-based models for better predictive accuracy, and an adaptive sampling mechanism for material test planning. A collaborative AM data management system combines community efforts and leverages on legacy FAIR data to turn the small data sets from individual material tests into vast amounts of data necessary for statistically sound process qualification. With advanced data analytics, such as transfer learning, heterogeneous data sets with a multitude of geometries, material types, and processes can be mined to develop empirical correlations between material properties, microstructure, part geometry and process parameters. Methods like transfer learning make machine learning systems more efficient and able to work with less data. In addition, combining physics-basedmodeling and toolsets like ICME can provide additional data sets based on simulations. AM qualification can also benefit from better design of experiments using physics-based models. Such advancements lead to less effort and time for material and process testing and promise to speed up AM process/material/part qualification. It is critical to develop and validate efficient and effective advanced data analytics, complemented by physics-based models, capable of undertaking rapid explorations of AM process-structure-property relationships with limited and diverse data sets. ASM International Ecosystem: ASM International has recognized the need to enable the digital capabilities of its membership and has embarked on an ambitious initiative to bring these capabilities to fruition. This initiative is tentatively labeled the ASM Data Ecosystem and is currently a “proof of concept” digital materials analytical environment. Ultimately, this proof of concept will be scaled into a fully featured digital “store” where members can access materials data, simulation tools, and use computational infrastructure with much lower barriers to entry. If targeted and executed properly, many of the FAIR data principle imperatives will be advanced, and some of the PEST challenges will be mitigated. The Data Ecosystem that ASM International is currently building will have many similarities with the data commons previously described. In addition to the digital environment, ASMwill join high-quality and pragmatic data management education to the core “table stake” of high-quality, useful materials data. The schematic of this initiative is provided in Fig. 5. Fig. 4 — A FAIR additive manufacturing framework for rapid qualification.
RkJQdWJsaXNoZXIy MTYyMzk3NQ==