Toward Reusable Surrogate Models: Graph-Based Transfer Learning on Trusses

Abstract Surrogate models have several uses in engineering design, including speeding up design optimization, noise reduction, test measurement interpolation, gradient estimation, portability, and protection of intellectual property. Traditionally, surrogate require that all training data conform to the same parametrization (e.g., variables), limiting freedom prohibiting reuse historical data. In response, this article proposes graph-based (GSMs) for trusses. The GSM can accurately predict displacement fields from static loads given structure’s geometry as input, enabling across multiple parametrizations. GSMs build upon recent advancements geometric deep learning, which led ability learn on undirected graphs: a natural representation To further promote flexible models, explores transfer learning within context demonstrates positive knowledge sets different topologies, complexities, loads, applications, resulting more data-efficient