| Abstract Scope |
We discuss the use of AI to efficiently perform engineering design, through the automated generation of high-fidelity surrogates for the response surface of interest. We leverage online scientific machine learning to steer automated experimentation, data collection, and analysis of the next best experiment(s) that will most efficiently produce the optimal solution for an engineering design problem. Utilizing all data, uncertainty, physics, expert knowledge, and other constraining information as “coordinate” transforms enables us to restrict the search space to only produce valid solutions. This greatly simplifies the optimization problem, and enables AI to generate a high-fidelity surrogate using very sparse data. We recently utilized this approach to automate Rietveld refinement, reducing the time-to-solution for textures from months to minutes. We also have applied our framework to the engineering design of mixture models for additive manufacturing, and have seen a similar orders-of-magnitude reductions the time-to-solution without loss of fidelity. |