CASE STUDY
Slope Design Curves with Machine Learning
Date:
2024
Industry:
Mining, Open Pit
Client:
Confidential
ITASCA Office:
Minneapolis
FLAC2D
Project Background
Open pit mining demands precise and efficient slope stability analysis to ensure safety while remaining cost effective. Traditional methods are reliable but time consuming and computationally intensive, which is challenging in rapidly changing mining environments. In this project, ITASCA bridges that gap by leveraging machine learning (ML) to streamline slope stability assessment.
FLAC2D modeling capabilities were integrated with cutting-edge machine learning algorithms to create a robust, user-friendly tool that provides engineers with quick, reliable, and accessible slope safety predictions.
Objectives
- Streamline and automate slope stability analysis to enhance decision-making speed.
- Improve accuracy and reliability of Factor of Safety (FoS) predictions through machine learning.
- Provide engineers with a user-friendly, real-time tool accessible on mobile devices.
- Reduce operational risks and improve safety in open-pit mining.
- Optimize the overall design process for slope stability, saving time and resources.
Challenges
The execution of surrogate models is unbelievably fast, but there are some limitations to this technique:
- The models are only valid for the range of inputs over which the training was done.
- The models are only as good as the numerical model that created the data they are trained from.
- The models may give rare, but large error outliers.
- Unlike a full numerical model, only a limited range of outputs can be predicted by a surrogate model.
- So many synthetic datapoints may be required or the individual models could take so long to run that the methodology could be impractical for some applications.
ITASCA’s Role
A combination of specialized consulting services and advanced software tools were employed to deliver a comprehensive solution that streamlined slope stability analysis, providing clients with rapid, accurate, and user-friendly assessments.
- Geotechnical Analysis: ITASCA conducted comprehensive assessments of slope stability, leveraging our extensive experience in geomechanics.
- Machine Learning Integration: The team developed and implemented ML models tailored to predict the Factor of Safety (FoS) for various slope conditions.
- Data Generation and Analysis: Utilizing FLAC2D with Python and cloud computing, ITASCA generated a robust dataset to train the machine learning models, ensuring accurate and reliable predictions.
The simulation and data generation process involved using FLAC2D to simulate slope stability scenarios. Over 120,000 simulations were run to create a comprehensive training dataset, covering a range of slope geometries, material properties, and groundwater conditions. The input parameters for the simulations included rock mass properties (cohesion, friction angle, and unit weight), slope geometry (height and angle of the slope), and water table conditions (depth and influence of groundwater).
For the ML model, a multi-layer perceptron (MLP) architecture was used, optimized for regression tasks to predict the Factor of Safety (FoS). The model incorporated input features like rock mass parameters, slope geometry, and water table conditions. The dataset was divided into training and validation sets to ensure robust model performance. Additionally, hyperparameter tuning was carried out to improve prediction accuracy and generalization.
Methodology
ITASCA’s system to define a geotechnical problem along with the range of values for the parameters of interest first involves writing a parameterized datafile for FLAC2D that can work with any combination of input values. Next, a cloud-based system is used to coordinate running all the cases and gathering the data. The data is gathered on a local computer and the ML training is done. Finally, a React.js based web application is custom written to provide an intuitive user interface.
Results & Impact
ITASCA’s expertise and software enabled the client to achieve their goals of automating slope stability assessments. The project also demonstrated successful integration of advanced technologies like machine learning into traditional engineering workflows, setting a precedent for future innovation. Benefits include:
- Efficiency Gains: The ML powered tool reduces the time required for slope stability analysis from hours or days to real-time predictions, streamlining the decision-making process.
- Cost Reduction: Automating over 120,000 FLAC2D simulations enables comprehensive data generation at a lower cost than manual analysis, reducing overall project expenses.
- Accessibility: The mobile-compatible web application ensures slope safety assessments can be performed on site, eliminating the need for specialized software or hardware.
- Enhanced Safety: By providing fast, accurate Factor of Safety (FoS) predictions, ITASCA’s solutions minimize the likelihood of slope failures, reducing risks to personnel, equipment, and operations.
- Improved Accuracy: The neural network model trained on high-fidelity FLAC2D data delivers reliable, consistent predictions, avoiding human errors common in manual calculations.