Multi-dimensional crashworthiness performance prediction and constrained optimization of the HFC energy absorbing structures for railway vehicles driven by deep learning frameworks

Abstract

The honeycomb-filled structures have better energy absorption performances but cross-effected by the complex interactions between filled honeycomb and metallic tube. Traditional surrogate model-based optimization methods can only ensure that the crashworthiness indicators meet the design expectations. However, unstable buckling and drastic load fluctuations cannot be avoided. In view of this, a multi-dimensional crashworthiness performance prediction and constrained optimization of a novel kind of honeycomb-filled composite structure are investigated. By utilizing the deep learning technique, the deformation images, crashworthiness indicators and load curves of the structure are predicted and introduced as constraints to the multi-objective optimization. Compared to the regular optimization results, the range of the Pareto front is significantly reduced after the introduction of extra constraints. Furthermore, the best solution obtained from the constrained optimization not only satisfy the conventional indicators constraint, but also performs well in terms of the deformation mode and load history. By applying the proposed method, the reliability of the optimization is dramatically improved. It is well proved that the proposed methodology can provide a feasible reference for similar problems of crashworthiness optimization of energy-absorbing structures.