Combined ANN-FEM approach for spatial-temporal structural response prediction: Method and experimental validation

The prediction of the mechanical behavior and possible failure mechanisms of engineering systems is essential in the design process. It is a challenging task in view of numerous influencing parameters, various unavoidable uncertainties and time-consuming numerical analyses based on complex finite element models. Therefore, a combined hierarchical ANN-FEM approach is proposed in this contribution. By means of this approach, the advantages of both simulation methods are used to predict uncertain spatial-temporal structural responses with less effort and comparable accuracy. A non-linear stability analysis of a three-dimensional carbon fiber reinforced plastic (CFRP) structure is conducted to illustrate the efficiency of the combined ANN-FEM approach. The individual components space, time and uncertainty are investigated in detail regarding their influence on the ANN architecture and complexity. The numerical predictions of the buckling behavior and the material failure have been compared then to experimental values measured in a symmetric three-point bending test. A sensitivity analysis is finally conducted which clarifies the strong dependence of the outcomes for the ultimate limit state on the fiber volume content, the structural thicknesses and the stiffness in fiber direction.