海洋桩靴基础穿刺破坏实时预测研究

海上自升式钻井平台以其造价低廉、便于移动和安装的优势被广泛应用于海洋地质勘察、风电安装和油气开采等领域,其基础类型多为桩靴基础。为了保证平台能在恶劣的海洋环境中安全作业,桩靴基础需要贯入海床以下一定深度以获得足够的承载力。然而,当桩靴基础在上硬下软土层中贯入时可能发生穿刺破坏导致平台损坏甚至倾覆。已有的桩靴穿刺破坏分析方法基于预设的地层参数预测穿刺荷载,由于无法考虑海床中地层及土性的不确定性,其准确性有待提高。将桩靴基础贯入过程中的监测数据与穿刺破坏机理相结合,通过66组离心机模型试验结果表征土体不确定性的影响,发展贝叶斯预测模型,实现了峰值阻力和穿刺深度的实时预测。基于规范推荐的荷载扩展分析法和冲剪系数分析法,建立了适用于规范法的概率模型,采用该模型对上砂下黏土层中桩靴基础的穿刺行为进行了预测,结果表明所提方法的预测精度随着监测数据的增加而提高,预测得到的峰值阻力误差在10%以内。

Study on real-time prediction of punch-through of spudcan foundation

Offshore jack-up drilling platforms are widely used in marine geological surveys, wind power installations, oil and gas exploration and other fields due to their low cost, ease of movement and installation. Their foundation types are mostly spudcan foundations. In order to ensure that the platform can operate safely in the harsh marine environment, the spudcan foundation needs to penetrate a certain depth into the seabed to obtain sufficient bearing capacity. However, when the spudcan foundation penetrates into the stiff-over-soft soil layers, punch-through may occur resulting in damage to the platform or even overturning. The existing punch-through failure analysis method predicts the punch-through load based on the preset stratum parameters. Since the uncertainty of the stratum and soil properties in the seabed cannot be considered, its accuracy needs to be improved. Based on the monitoring data during the penetration of the spudcan foundation with the punch-through failure mechanism, the effect of the uncertainty of the soil is characterized by the results of 66 sets of centrifuge model tests. Then the Bayesian prediction model is developed to realize the peak resistance and punch-through depth real-time prediction. A probability model suitable for the code based on the load spread method and punching shear method recommended by the code is established to predict the punch-through behavior of the spudcan foundation on sand-over-clay. The results show that the prediction accuracy of the proposed method increases with the increase of monitoring data, and the peak resistance error of the prediction is within 10%.