Hybrid Machine-Learning-Based Prediction Model for the Peak Dilation Angle of Rock Discontinuities

Area of Science:

• Geotechnical Engineering
• Rock Mechanics
• Computational Intelligence

Background:

• The peak dilation angle is crucial for evaluating rock mass mechanical behavior.
• Traditional methods for determining the shear dilation angle lack sufficient accuracy and efficiency.
• Machine learning offers a promising approach for predicting outcomes based on multiple factors.

Purpose of the Study:

• To develop and validate a novel hybrid machine learning model for predicting the peak dilation angle.
• To enhance prediction accuracy and optimize model hyperparameters for rock discontinuity analysis.
• To compare the performance of the proposed model against existing methods.

Main Methods:

• A hybrid machine learning model combining Support Vector Regression (SVR) with a grid search optimization algorithm was developed.
• The model was trained and tested on a dataset of eighty-nine rock discontinuity samples.
• Input variables included morphology and mechanical property parameters, with a focus on normal stress.

Main Results:

• The proposed hybrid SVR model achieved a coefficient of determination (R²) of 0.917 and a mean absolute percentage error (MAPE) of 4.5%.
• The model demonstrated superior performance compared to the original SVR model and traditional analytical models.
• Normal stress was identified as the most influential mechanical parameter affecting the peak dilation angle.

Conclusions:

• The developed hybrid machine learning model is effective for predicting the peak dilation angle of rock discontinuities.
• The integration of grid search optimization significantly improves SVR performance.
• This approach provides a more accurate and efficient tool for rock mechanics assessments.