Current approaches for modeling structural components and assemblies with nonlinear and history-dependent behaviors rely on detailed finite element formulations of each component. Consequently, their computational cost could become prohibitive for many engineering applications. In this work, we introduce a framework to develop data-driven dimensionally-reduced surrogate models at the component level, which we call smart parts (SPs), to establish a direct relationship between the input-output parameters of the component. Our method utilizes advanced machine learning techniques to develop SPs such that all the information pertaining to history and nonlinearities is preserved. Unlike other data-driven approaches, our method is not limited to any particular type of nonlinearity and it does not impose restrictions on the type of analysis to be performed. This renders its application straightforward for a diverse set of engineering problems, as we show through multiple case studies. We demonstrate our method's ability by comparing its results with those obtained via high-fidelity finite element simulations. Our findings indicate that SPs dramatically reduce the computational cost without much loss in accuracy, thus enabling the analysis of complex assemblies in the nonlinear regime. Our approach is general and can be adopted in different fields of research such as structural health monitoring and topology optimization, to name a few.
展开▼