An efficient Monte Carlo strategy for elasto-plastic structures based on recurrent neural networks

The Monte Carlo method is computationally expensive. Especially for reliable response statistics in the tail region, a large number of samples has to be computed. In this paper, we propose a strategy that speeds up the crude Monte Carlo method using artificial intelligence. We introduce a recurrent neural network that replaces the iterative calculation procedure to evaluate nonlinear, hysteretic behavior of a structure subjected to a randomized load history. The new strategy is demonstrated on an academic example, i.e., an elasto-plastic frame, which is loaded by a time-variant force. The trained recurrent neural network reliably approximates the structures mechanical response and considerably reduces the computation time per sample.