High efficient load paths analysis with U* index generated by deep learning

Abstract A novel load transfer path analysis in mechanical structures is realized by using deep learning (DL) approach with high efficiency and accuracy. The load transfer paths on plate structures considering the light-weight and reinforcement designs are obtained through a mathematical index of the load transfer calculation, U * , which is generated by a proposed DL model. Finite element models (FEM) of different sample plates are firstly built to generate the corresponding U * data used as training data set. The DL model is then trained by a small amount of U * data set obtained from FEM with an optimized network structure by random search for hyper-parameters. U * index and load transfer paths on plates with any random hole openings and stiffeners are then obtained from the trained DL model and compared with the ground true results from FEM to prove the efficiency and accuracy of the DL model after training. Feasibility of the DL application to the load transfer path derivation on varying design of mechanical structure is proven for efficient structural analysis and design improvement without re-modeling and numerical calculations.