Artificial Intelligence in Soil Compaction Control #Sciencefather #Researcherawards


Introduction

The evaluation of compaction quality in earthworks and pavements traditionally relies on density-based acceptance methods derived from laboratory Proctor tests. While effective, these methods are time-intensive, costly, and spatially limited. The lightweight dynamic cone penetrometer (LDCP) offers a rapid, on-site alternative capable of producing key indices such as π‘žπ‘‘0 and π‘žπ‘‘1. However, the reliability of LDCP data often depends on site-specific calibrations, restricting its general application. To overcome these challenges, this research introduces a supervised machine learning framework designed to predict LDCP indices directly from soil descriptors, thereby optimizing the process of compaction quality control and enhancing operational efficiency in geotechnical field applications.

Methodological Framework

The study employed a supervised machine learning framework integrating multiple predictive models to estimate the LDCP indices π‘žπ‘‘0, π‘žπ‘‘1, and 𝑍𝑐. The dataset included 360 observations from various soil campaigns, incorporating key descriptors such as gradation, plasticity, moisture content, and GTR classification. Several algorithms were compared, including linear regression, support vector regression, Random Forest, XGBoost, and a compact multilayer perceptron (MLP). The training and testing processes used an 80/20 data split with 5-fold cross-validation to ensure robust performance assessment. This methodological design ensured that each model could be objectively benchmarked for prediction accuracy and computational efficiency.

Model Evaluation and Performance Metrics

To assess model reliability, the study utilized multiple performance indicators, including the coefficient of determination (𝑅²), root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE). Among all the tested models, the multilayer perceptron (MLP) exhibited the highest performance for predicting π‘žπ‘‘1, achieving an 𝑅² of 0.794 and an RMSE of 5.866, closely followed by the XGBoost model with an 𝑅² of 0.773 and an RMSE of 6.155. Statistical analysis using paired bootstrap tests and Holm correction confirmed that the differences between MLP and XGBoost performances were not significant, underscoring their comparable efficiency in compaction prediction.

Interpretability and Feature Importance

Model interpretability was a key focus, addressed using SHAP values, permutation importance, partial dependence plots (PDP), and accumulated local effects (ALE). These tools provided insight into the contribution of each soil variable toward LDCP prediction outcomes. The analysis revealed that density and moisture-related factors, particularly field dry density (𝛾𝑑,field), compaction ratio (𝑅𝐢SPC), and moisture content (W), were the most influential predictors. Gradation and plasticity parameters played a secondary yet meaningful role, fine-tuning the predictions by accounting for soil texture and composition variability. This interpretability layer enhances trust and transparency in machine learning applications within geotechnical engineering.

Calibration and Validation for QA/QC Integration

The framework further incorporated calibration diagnostics and conformal prediction intervals to align model outputs with quality assurance and control (QA/QC) protocols. These predictive intervals allowed practitioners to quantify uncertainty in model outputs, facilitating risk-aware decision-making during field compaction assessment. By integrating statistical calibration with machine learning predictions, the system bridges the gap between data-driven insights and practical field requirements. This hybrid validation approach ensures that predictions can complement traditional density-based acceptance tests while reducing the dependency on extensive LDCP calibration efforts.

Conclusion and Future Prospects

This research demonstrates that machine learning models, particularly MLP and XGBoost, can serve as powerful surrogates for LDCP indices, offering interpretable and computationally efficient tools for compaction quality assessment. The findings validate that incorporating soil characteristics into predictive modeling can significantly reduce field calibration needs and support real-time QA/QC decisions. Future research could expand this framework to incorporate larger datasets and sensor-based monitoring, further strengthening the role of artificial intelligence in advancing soil compaction control and sustainable infrastructure development.

Global Particle Physics Excellence Awards


Get Connected Here:................ Twitter: x.com/awards48084 Blogger: www.blogger.com/u/1/blog/posts/7940800766768661614?pli=1 Pinterest: in.pinterest.com/particlephysics196/_created/ Tumbler: www.tumblr.com/blog/particle196

Hashtags

#Sciencefather, #Reseacherawards, #SoilCompaction, #MachineLearning, #ArtificialIntelligence, #GeotechnicalEngineering, #QualityControl, #PavementEngineering, #Earthworks, #DynamicConePenetrometer, #DataScience, #CivilEngineering, #XGBoost, #MLP, #SoilMechanics, #Infrastructure, #PredictiveModeling, #CompactionTesting, #AIInGeotech, #SmartConstruction, #SustainableEngineering, #ResearchInnovation,

Comments

Post a Comment

Popular posts from this blog

Hunting for Dark Matter The Cosmic Mystery

Image
"Hunting for Dark Matter: The Cosmic Mystery" refers to the ongoing scientific efforts to detect and understand dark matter, a form of matter that does not emit, absorb, or reflect light, making it invisible and detectable only through its gravitational effects on visible matter. Dark matter is believed to make up about 27% of the universe's total mass and energy, yet it remains one of the biggest mysteries in modern astrophysics and cosmology. Scientists are using various methods to search for dark matter, including: Direct Detection Experiments: These experiments aim to detect dark matter particles directly by observing their interactions with ordinary matter. They use highly sensitive detectors buried deep underground to avoid interference from cosmic rays and other particles. Indirect Detection: By looking for the remnants of dark matter interactions, such as gamma rays, neutrinos, or antimatter, researchers hope to find evidence of dark matter. This approach invol...
Read more