Artificial Neural Networks Applied to Flexible Pipes Stochastic Non Linear Dynamic Analysis
Victor Viana Chaves1; Luis Volnei Sudati Sagrilo1; Vinícius Ribeiro Machado da Silva1
1 Federal University of Rio de Janeiro /COPPE
doi:10.20906/CPS/CILAMCE2015-0510
Resumo
In order to refine and optimize flexible pipes design, due the continuous increasingly challenging deep water environments, a lot of attention is put into setting up more realistic and sophisticated models for global dynamic numerical analyses. Recently, different hybrid methods combining Finite Element Analyses (FEA) and Artificial Neural Networks (ANN) have presented promising results in terms computational costs reduction when applied to the simulation of mooring lines. The hybrid ANN-FEA methodology presented in this work aim to reduce the computational costs involved in flexible pipes stochastic nonlinear global dynamic analyses, predicting flexible pipe tension and curvatures in the bend stiffener region. A typical 6"free-hanging flexible pipe was numerically simulated with the application of 88 different fatigue sea states coming from 8 directions. For each sea state, the hybrid approach is applied in this way: firstly using short FEA simulations an ANN is trained and then using only the ANN and the prescribed floater motions the rest of the response time histories are obtained. With the predicted tension and curvatures a local analysis is used to calculate stresses in tensile armour wires and the corresponding fatigue lives. To evaluate the optimal ANN configuration a sensitive analysis was developed regarding two ANN main parameters: training time series length and neurons in the hidden layer.
Palavras-chave: artificial neural networks; flexible pipes; stochastic dynamic analysis