Machine Learning-Based Partition Method for Cyclic Development Mode of Submarine Soil Martials from Offshore Wind Farms

Ben He, Mingbao Lin, Zhishuai Zhang, Bo Han, Xinran YuJournal of Marine Science and Engineering 2025 · DOI

TL;DR

This paper introduces a machine learning-based approach to classify the cyclic behavior of submarine soils, crucial for the stability of offshore wind turbine foundations. By analyzing hysteresis loops from dynamic triaxial tests, they developed models that can distinguish between "stable" and "destructive" soil responses, offering better predictive capabilities than traditional methods.

What Problem Does This Paper Address?

Offshore wind turbines are subjected to constant cyclic loading from wind, waves, and currents. This puts immense stress on the surrounding seabed soils, which can lead to failure and compromise the safety and operational lifespan of these structures. While research has explored aspects like strain accumulation and liquefaction, a nuanced understanding of the cyclic development modes of these soils, particularly as reflected in their hysteresis loop behavior, has been lacking. This gap hinders accurate foundation design and risk assessment. The paper aims to address this by developing a robust method to classify these cyclic behaviors, providing critical insights for engineers.

Methods

The core of this research lies in applying machine learning to classify the hysteresis loop shapes of submarine soils. The authors conducted dynamic triaxial tests on 78 soil samples from four offshore wind farms, varying confining pressures and cyclic stress ratios (CSR).

The key methodological contributions are:

  • Identification of Two Development Modes: Through experimental analysis, they identified two distinct modes of soil behavior under cyclic loading: a "stable development mode" under low cyclic loads and a "destructive development mode" under high cyclic loads. These modes are characterized by distinct hysteresis loop shapes and effective stress paths.
  • Machine Learning-Based Partitioning: They employed two prominent machine learning algorithms for classification:
  • Random Forest (RF): An ensemble learning method that builds multiple decision trees and aggregates their predictions. It's known for its robustness and ability to handle complex relationships.
  • Gradient Boosting Decision Tree (GBDT): Another ensemble method that builds trees sequentially, with each new tree correcting the errors of the previous ones.
  • Hyperparameter Optimization: To ensure optimal performance, the hyperparameters of both RF and GBDT models were tuned using a grid search method.
  • Model Evaluation: The models were evaluated using standard classification metrics, including accuracy, precision, recall, F1-score, and Receiver Operating Characteristic (ROC) curves with Area Under the Curve (AUC).

Datasets

The study utilized a dataset of 78 submarine soil samples collected from four offshore wind farm locations in China: the Bohai Sea, Yellow Sea, and East China Sea. These samples represented five distinct soil types: mucky clay (MC), silty clay (SC), silty soil (SS), weathered residual soil (WRS), and completely weathered granite (CWG). The experimental data generated from dynamic triaxial tests on these samples formed the basis for training and evaluating the machine learning models. The data was split into a 7:3 ratio for training and testing.

Key Insights and Ideas

  • Hysteresis Loop as a Signature: The hysteresis loop, often overlooked in favor of strain or liquefaction metrics, is a powerful indicator of a soil's cyclic mechanical properties and its development mode under stress.
  • Two Distinct Paths to Failure (or Stability): The clear distinction between "stable" and "destructive" development modes highlights that soil failure under cyclic loads isn't a single phenomenon but can manifest in different ways, each with unique stress-strain characteristics.
  • Machine Learning's Predictive Power: The study demonstrates that machine learning, particularly Random Forest, can effectively learn these complex patterns from experimental data, offering a more automated and potentially more accurate way to classify soil behavior than manual interpretation.
  • Critical CSR as a Threshold: The identification of a "critical CSR" for each soil type and confining pressure provides a quantifiable threshold for predicting when a soil might transition from a stable to a destructive state, directly informing engineering design.

Key Results

  • Model Performance: The Random Forest (RF) model outperformed the Gradient Boosting Decision Tree (GBDT) model in discriminating hysteresis loop shapes.
  • Accuracy Metrics (Test Set):
    • RF Model: Achieved an accuracy of 0.9583, precision of 0.9474, recall of 0.9730, F1-score of 0.9737, and AUC of 0.9737.
  • GBDT Model: Achieved an accuracy of 0.9167, precision of 0.9474, recall of 0.9474, F1-score of 0.9474, and AUC of 0.8737.
  • Generalization: The RF model showed superior generalization ability, with its performance on the test set being significantly higher than GBDT, indicating its robustness to unseen data.
  • Critical CSR Values: The paper provides critical CSR values for different soil types and confining pressures, serving as a direct engineering parameter. For instance, completely weathered granite under 100 kPa confining pressure has a critical CSR of 0.35.

Limitations and Future Work

  • Laboratory vs. Field Conditions: The authors acknowledge that laboratory conditions are simplified and may not fully capture the complexity of the actual marine environment, which includes combined effects of waves, tides, and currents.
  • Single-Factor Focus: Much of the existing research, including this study, tends to focus on single factors or simplified conditions, lacking in-depth analysis of multi-factor coupling effects.
  • Model Refinement: While the machine learning models are effective, there is still room for improvement in their refinement and optimization.
  • Future Work: The authors suggest future research should focus on combining field-measured data to verify and correct the laboratory-based models, thereby enhancing their applicability in practical engineering. Investigating the coupling effects of various environmental factors on soil behavior is also a crucial next step.

Summary updated

Abstract

Offshore wind turbines are subjected to long-term cyclic loads, and the seabed materials surrounding the foundation are susceptible to failure, which affects the safe construction and normal operation of offshore wind turbines. The existing studies of the cyclic mechanical properties of submarine soils focus on the accumulation strain and liquefaction, and few targeted studies are conducted on the hysteresis loop under cyclic loads. Therefore, 78 representative submarine soil samples from four offshore wind farms are tested in the study, and the cyclic behaviors under different confining pressures and CSR are investigated. The experiments reveal two unique development modes and specify the critical CSR of five submarine soil martials under different testing conductions. Based on the dynamic triaxial test results, the machine learning-based partition models for cyclic development mode were established, and the discrimination accuracy of the hysteresis loop were discussed. This study found that the RF model has a better generalization ability and higher accuracy than the GBDT model in discriminating the hysteresis loop of submarine soil, the RF model has achieved a prediction accuracy of 0.96 and a recall of 0.95 on the test dataset, which provides an important theoretical basis and technical support for the design and construction of offshore wind turbines.
BibTeX
@article{he2025machine,
  author = {Ben He and Mingbao Lin and Zhishuai Zhang and Bo Han and Xinran Yu},
  title = {{Machine Learning-Based Partition Method for Cyclic Development Mode of Submarine Soil Martials from Offshore Wind Farms}},
  journal = {Journal of Marine Science and Engineering},
  year = {2025},
  doi = {10.3390/jmse13030533},
}
Figure