A. Stein
CAMINNO, Inc.,
United States
Keywords: scientific machine learning, AI, Advanced Manufacturing, closed loop optimization
Summary:
In the past 50 years simulations, design & manufacturing have been performed by subject matter experts. High-tech and deep-tech applications such as commercial nuclear fusion energy & carbon sequestration use digital engineering & manufacturing tools that face their limits. The complexity of the design and the complexity of the manufacturing process is currently treated separately. This is because our work is limited by digital tools and the processing capability of the human mind. Currently, many high-technology applications do not use advanced manufacturing (AM) processes, such as additive manufacturing, since they are not repeatable and have reliability & quality issues. These need to be addressed for AM parts in high-tech and safety applications. One reason for this is the inherent high stochasticity of AM. The spatiotemporal thermal profile of a part during AM (thermal history) influences flaw formation, microstructure evolution, and surface/geometric integrity, all of which ultimately affect the mechanical properties of the part. The AM outcome varies from machine to machine, sometimes even with two machines of the same make and model and is also dependent on a host of factors that are not completely understood. Such unreliable outcomes are a major pain point for manufacturing on-demand (MOD) of highly customized parts required in development of nuclear fusion, aerospace, and defense applications using AM. Our core innovation is CAMINNO Gen6 - a multi-objective design optimization algorithm that employs scientific ML-based process modeling - coupling microstructural models with experimental manufacturing data. This enables the creation of a manufacturing digital twin that can predict the different manufacturing outcomes in near real-time. AM surrogate models for manufacturing are necessary for fast and accurate computation - to take the data in from the manufacturing process, compute the required quality metrics, predict the probable future thermal and mechanical histories of the part, recommend the design changes required to ensure the performance criteria is satisfied by the finished component, and relay it back to the AM machine. Hybrid AI models that combine scientific ML with data-driven techniques may provide the necessary accuracy and computational speedup. A manufacturing digital twin will be capable of addressing and reducing the huge time and cost associated with current design & manufacturing workflows. Additionally, it would enable exploration of the design and manufacturing space much faster and accurately than is possible with currently prevalent workflows. Finally, it has the potential to reduce the engineering & manufacturing carbon footprint.