Using Artificial Neural Networks to Predict Influences of Heterogeneity on Rock Strength at Different Strain Rates

Area of Science:

• Geotechnical Engineering
• Materials Science
• Computational Mechanics

Background:

• Natural rock masses exhibit significant heterogeneity due to pre-existing cracks and infill materials, altering their mechanical properties.
• Existing literature lacks efficient and accurate methods to predict the impact of heterogeneity on rock strength, especially at varying strain rates.

Purpose of the Study:

• To bridge the literature gap by employing artificial neural network (ANN) approaches for predicting the influence of heterogeneity on the strength of 3D-printed rocks.
• To quantitatively define rock heterogeneity using parameters such as crack number, orientation, and offset distance.
• To investigate the effect of different strain rates on the strength of heterogeneous rocks.

Main Methods:

• Development and training of an artificial neural network model using data from 42 quasi-static and 42 dynamic Brazilian disc experimental tests.
• Quantitative definition of rock heterogeneity through crack number, initial crack orientation relative to the loading axis, crack tip distance, and crack offset distance.
• Optimization of the ANN architecture, including hidden layer count and transfer functions, via parametric studies.

Main Results:

• The trained ANN model demonstrated excellent prediction accuracy for the influences of high-dimensional heterogeneous parameters and strain rate on rock strength.
• Sensitivity analysis revealed that strain rate is the most critical factor influencing the strength of heterogeneous rock.
• The study establishes a robust relationship between rock strength and defined heterogeneous parameters across various strain rates.

Conclusions:

• Artificial neural networks provide an efficient and accurate tool for predicting the mechanical behavior of heterogeneous rocks.
• Strain rate is a paramount factor governing the strength of heterogeneous rock masses, underscoring the importance of dynamic testing and analysis.
• The methodology offers a pathway for understanding and predicting the complex interplay between material heterogeneity and mechanical response in engineered and natural rock.