Neural Network Modeling Of The Nonlinear Dynamic Structural Offshore System With Hysteresis
Free (open access)
13 - 24
D. M. Rocha, N. F. F. Ebecken, L. P. Calôba & D. L. Kaiser
This paper proposes an empirical modeling of the system formed by the riserplatform connection, in deep water. This connection has the objective of minimizing the acting bending moment, possesses high complexity and highcriticity due to economic and environmental consequences from its fault. The main element in the joint is made of elastomeric material, which reveals nonlinear hysteresis. In addition, this whole connection system presents nonlinearities due to the action of dynamic loading and large motions. TDNN and Recurrent Neural Networks (RNN) are being investigated since they possess the ability to model nonlinear hysteretic behaviors and also dynamic systems. Simulation results have confirmed that RNN is the one that presents the best representation of the system studied. Emphasis shall be given to the additional difficulties, which arise from the utilization of real data in the modeling process for this system. Keywords: Recurrent Neural Network, NARX, hysteresis, Flexjoint SCR. 1 Introduction The specific system under study is that which makes the connection between the riser and the platform, through a connector called Flexjoint. The Flexjoint (see Figure 1) is installed on the SCR (Steel Catenary Riser) top, and through an elastomeric joint it allows the SCR to bend when submitted to dynamic environmental loading of wave, wind and current. The elastomer plays an important and complex role and any deterioration of the elastomer can affect the
Recurrent Neural Network, NARX, hysteresis, Flexjoint SCR.