In the Arctic shallow waters, marine pipelines are threatened by traveling icebergs where the seabed may be gouged by these moving masses during warmer months. Estimation of the subgouge soil response is considered as a serious design factor for the subsea infrastructures since minimizing the required burial depth to physical protection is quite crucial for the project budget. In this paper, the capability of the Multi- Layer Perceptron Neural Network (MLPNN) is utilized to simulate the ice-induced sand deformations. By conducting a sensitivity analysis, the best MLPNN models and the most significant input parameters are identified.


Arctic area contains a huge amount of hydrocarbon deposits such as crude oil and natural gases that increased energy demand is the cause of exploration in the Arctic regions. Subsea pipelines are widely used to transfer the hydrocarbon and other exploration and production-related contents between the onshore and offshore facilities (Alba 2015). Marine pipelines are threatened by the ice gouging, and pressure ridges attack crossing the pipeline route in the Arctic shallow waters. Subsea trenching and backfilling are commonly used to bury the pipeline for physical protection against the ice scour. The schematic layout of an ice-seabed interaction process is illustrated in Figure 1. As shown, a hyperbolic curve is produced just beneath the ice keel bottom where the maximum soil displacement is occurred at the soil surface.

Identifying the maximum deformations for safe and cost-effective protection of the pipeline is a challenging problem. Costly experimental and long-running numerical simulation is mandatory for accurate modeling of the subgouge soil deformation and consequently, the pipe response. For instance, a joint industry and government-sponsored research program entitled the Pressure Ridge Ice Scour Experiment (PRISE) was conducted at Centre for Cold Ocean Resources Engineering (C-CORE) to understand the requirement for the safe and cost-effective design of subsea pipelines against the ice scour. The study showed the importance of “dead wedge” underneath the ice keel with respect to variation in scour loads and subscour soil displacements (C-CORE 1995). Hynes (1996) carried out a centrifuge ice gouging study in sand observed a linear relationship between the scour loads and depth. The author showed the sand deformation affected simple direct shear due to the stress-strain behavior of soil. Eventually, it was suggested that numerical studies should be performed to simulate the centrifuge ice gouging modeling. Yang (2009) conducted a centrifuge study to measure deformations of ice-scoured sand. It was shown that the maximum gouge force was a function of the gauge geometry and keel attack angle and value of the frontal berm height. Arnau and Ivanovic (2019) carried out 1g floor tests on cohesion-less seabed scour due to the ice attack. Ultimately, drifting velocity on the scouring loads was identified as an important parameter. Nematzadeh and Shiri (2019a) developed a CEL model for free-field ice gouging analysis in sand using ABAQUS/Explicit. The authors incorporated the non-linear strain rate and softening effects through a user-defined subroutine. The study resulted in an improved prediction of the subgouge soil deformation and the keel reaction forces obtained from published experimental studies. Additionally, Nematzadeh and Shiri (2019b) simulated the ice-seabed interaction process using a self-correcting soil model in order to update the shear strength parameters during the pre-peak hardening and the postpeak softening of the sand. The authors showed that the subgouge soil deformation might be overestimated by the conventional decoupled approaches. Additionally, Nematzadeh and Shiri (2020) modeled the effect of non-linear stress-strain behavior of dense sand in an ice gouging problem by using a modified Mohr-Coulomb (MMC) model. The authors concluded that the size of side berms and the frontal mound were affected by the magnitude of attack angle, where greater subgouge deformations and reaction forces were observed for models with the shallower attack angles.

This content is only available via PDF.
You can access this article if you purchase or spend a download.