Fast solution of elasto-plastic problems by reduced order finite element method with manifold learning

Solution of elasto-plastic problems is often computationally intensive. The most effective way to accelerate the computation is to reduce the number of the degree of freedom of the model. Considering the fact that constitutive relations of solid materials are very diverse and usually quite complicated, the paper investigates a model order reduction technique which can be embedded into existing finite element frameworks and reuse their constitutive models. The semi-intrusive technique can thus be used to solve nontrivial problems. Two typical models, i.e., the perforated plate under tensile load and the pressure vessel head under cyclic load, are tested to evaluate the accuracy and the efficiency of the reduced order model. It is shown that a speedup of around two orders of magnitude can be achieved by the reduced order model while maintaining the same level of accuracy. Moreover, the requirements for processor and memory of the computer are also significantly reduced.