Correction: Predictive modeling on the mechanical properties of marine coral sand-clay mixtures based on machine learning algorithms and triaxial shear tests

Bo Yang, Kaiwei Xu, Yi Liu, Peng CuiFrontiers in Marine Science 2026 · DOI

Abstract

Offshore engineering in regions like the South China Sea relies heavily on marine coral sand as a fill material. However, the brittle nature of coral sand necessitates stabilization through clay mixing and geogrid reinforcement. Predicting the mechanical behavior of these Marine Coral Sand-Clay Mixtures (MCCM) is complex due to non-linear variables. This study develops a high-precision machine learning model, specifically a Backpropagation Neural Network optimized by a Logical Development Algorithm (LDA-BPNN), to forecast MCCM strength. Utilizing a dataset of 900 triaxial shear tests, the LDA-BPNN model achieved an R-value of 0.99651. Furthermore, an explicit empirical formula was derived to provide engineers with a practical tool for field applications.

Introduction

In marine environments, coral sand serves as a primary construction material for offshore foundations. Despite its availability, its inherent brittleness and susceptibility to particle breakage under pressure pose significant structural risks. To mitigate these issues, engineers frequently incorporate clay and synthetic geogrids to enhance stability.

Traditional methods for determining the mechanical properties of these mixtures involve unconsolidated undrained (UU) triaxial shear tests. While accurate, these tests are resource-intensive and time-consuming, often creating bottlenecks in the design and construction phases. There is a pressing need for digital predictive models that can rapidly and accurately estimate the strength of MCCM under varying environmental and structural conditions.

Materials and Methods

The research methodology integrated physical experimentation with advanced computational modeling.

Experimental Setup

A series of UU triaxial tests were performed on MCCM samples. To investigate the impact of reinforcement, 3D-printed triaxial geogrids with equilateral triangle apertures were embedded within the samples. The experimental matrix covered a range of clay contents (30%–70%), reinforcement layers (0, 1, or 2), confining pressures (50–200 kPa), and water contents (0%–36%).

Machine Learning Framework

The study evaluated several optimization techniques for a Backpropagation Neural Network (BPNN), including Genetic Algorithms (GA) and Particle Swarm Optimization (PSO). However, the primary innovation was the implementation of a Logical Development Algorithm (LDA) for data preprocessing. LDA functions as a supervised dimensionality reduction technique that maximizes the ratio of between-class variance to within-class variance. This "cleaning" process allows the BPNN to focus on the most discriminative features, significantly improving generalization on unseen data.

Results and Discussion

The LDA-BPNN model demonstrated superior performance across all evaluated metrics compared to standard BPNN, GA-BPNN, and PSO-BPNN models.

Model Performance

The LDA-BPNN achieved a near-perfect linear correlation with experimental data, yielding an R-value of 0.99651 on the test set. Its error rates were the lowest among the tested architectures, with a Root Mean Square Error (RMSE) of 1.24356 and a Mean Absolute Percentage Error (MAPE) of 7.15%. In contrast, the baseline BPNN model exhibited a MAPE exceeding 15%.

Sensitivity Analysis

Using Shapley Additive Explanations (SHAP), the researchers identified water content as the most critical factor influencing MCCM strength, accounting for 35.9% of the model's predictive importance. The analysis revealed a non-linear relationship: while initial water content acts as a lubricant to facilitate particle arrangement, excessive moisture weakens the cemented bonds between particles, leading to a decrease in overall strength.

Engineering Application

To bridge the gap between complex AI models and field engineering, the authors translated the LDA-BPNN model into an explicit empirical formula. This formula allows practitioners to calculate material strength using standard mathematical operations. Validation against 30 real-world cases showed an $R^2$ of 0.99798, confirming its reliability for practical use.

Conclusion

The integration of LDA-optimized machine learning provides a robust "digital twin" for predicting the mechanical properties of marine fill materials. By identifying water content as a primary driver of stability and providing a simplified empirical formula, this research offers valuable tools for ensuring the structural integrity of offshore foundations. Future work will focus on expanding the dataset to include diverse soil types and investigating material behavior under dynamic loading conditions, such as wave action and seismic events.

Summary updated

Abstract

• Please read through all the templates before choosing• Pick the most relevant text template(s) from the following page and delete all others.• Edit the text as necessary, ensuring that the original incorrect text is included for the record, please see the below.• Please do not use any extra formatting when editing the templates, and only modify the red text unless absolutely necessary• Submit to Frontiers following the instructions on this page.When the original text contained incorrect information, to preserve the scientific record, please include that text when editing the below templates. For example:There was a mistake in the Funding statement, an incorrect number was used. The correct number is "2015C03Bd051.". The publisher apologizes for this mistake.The original version of this article has been updated.In the published article, there was a mistake in the Funding statement.
BibTeX
@inproceedings{yang2026correction,
  author = {Bo Yang and Kaiwei Xu and Yi Liu and Peng Cui},
  title = {{Correction: Predictive modeling on the mechanical properties of marine coral sand-clay mixtures based on machine learning algorithms and triaxial shear tests}},
  booktitle = {Frontiers in Marine Science},
  year = {2026},
  doi = {10.3389/fmars.2026.1789998},
}
Figure