Deep learning system for finite element approximation and stiffness matrix generation apparatus according to reference data model

A finite element deep learning system comprising:

• a data generation unit configured to generate a finite element normalized with fixed points and variable points as training data;
• a strain computation unit configured to compute reference strain values based on the generated training data;
• a deep learning network including a plurality of layers having preset weights, respectively, and connected to one another, wherein the deep learning network is configured to generate a matching matrix for matching the reference strain values with displacements set in the training data according to arbitrary attribute information set in the training data and geometric information as position information related to the variable points;
• and
• a training control unit configured to:
• train the deep learning network with respect to a plurality of the training data generated by the data generation unit based on a cost function by which a difference between strain values computed according to the matching matrix and the reference strain values are equal to or smaller than a threshold value;
• and
• transform the matching matrix generated through the deep learning network into a matching matrix including a zero energy mode by reflecting a preset transformation matrix when the deep learning network is trained,
• wherein the trained deep learning network is independent from an analysis domain such that the trained deep learning network is applicable for finite element analysis upon a change to the analysis domain.