基于多层感知机的坠物对海底管道损伤预测及可靠性分析

Damage Prediction and Reliability Analysis of Submarine Pipelines Subject to Dropped Objects Based on Multi-Layer Perceptron

在海洋平台生产过程中,难免发生重物坠落事故,对海底管道造成撞击损伤,引起环境污染及经济损失。为保证管道在运行期间的安全性,有必要对其进行损伤预测和可靠性分析。本文构建了多层感知机(MLP)网络,以有限元计算结果作为训练样本,对管道损伤进行了预测。同时,基于MLP网络,提出了求解管道受坠物撞击的可靠度的Monte-Carlo法(MLP-MCM)。研究结果表明:MLP对管道损伤预测平均相对误差为3.66%,精度高且计算省时。进行管道可靠性分析时,MLP-MCM可考虑埋深及管土相互作用的影响,求得的可靠性指标较高;而经验公式法忽略上述因素的影响,偏于保守。

In the production process of offshore platforms, it is inevitable that heavy objects falling accidents occur, causing impact damage to submarine pipelines, which will lead to environmental pollution and economic losses. To ensure the safety of pipelines during operation, it is necessary to conduct damage prediction and reliability analysis. The multi-layer perceptron network(MLP network) is constructed in this paper. The results of finite element calculation are used as training samples for the network to predict the pipeline damage. Meanwhile, the Monte-Carlo method based on MLP network(MLP-MCM) is proposed to calculate the reliability of pipelines undertaking impacts from dropped objects. The results show that the average relative error of pipeline damage prediction by MLP network is 3.66%, which is accurate and time saving. In the pipeline reliability analysis, the MLP-MCM considers the influences of burial depth and pipe-soil interaction, and thus the solved reliability index is higher. While, empirical formula method ignores the above factors and the calculation result tends to be conservative.