A virtual model architecture for engineering structures with Twin Extended Support Vector Regression (T-X-SVR) method

Abstract

A machine learning aided virtual model architecture framework is presented. With great computational stability and preserved convexity features, the Twin Extended Support Vector Regression (T-X-SVR) method is adopted to imitate the underpinned and sophisticated constitutive relationship between the systematic inputs and outcomes in real-world applications. Various numerical simulations can be implemented on the virtual model with greatly reduced computational costs. The multi-type information from heterogeneous sources and multifarious engineering applications are supported by the proposed framework. For the virtual model aided numerical analysis with the multi-type information, fuzzy-valued probabilistic distributional characteristics of the bounds of the concerned structural responses are estimated. The capability of the modularity feature of the virtual model improves the operational flexibility which makes the implementation of such advance technique more user friendly. To demonstrate the applicability and computational efficiency of the proposed framework, three practical engineering stimulated problems (i.e., the mechanical dynamic system, multiphysics with heat transfer and gas flow interaction, and the expansion process of a biomedical stent with both material and geometry nonlinearities), involving multi-type information (i.e., random parameters and fuzzy sets), are fully investigated through the proposed virtual model architecture framework.