A comprehensive machine learning and data analysis framework for Microtunneling Boring Machine (MTBM) operations. This project analyzes 23+ key operational parameters to optimize tunneling performance, predict maintenance needs, and ensure precise tunnel alignment.
Predicts tunnel deviations using Random Forest regression based on cylinder pressures and operational parameters.
Key Findings:
- R-squared: 0.89 (89% accuracy)
- Top predictor: Cylinder Pressure Differential (28% importance)
- MAE: 1.8mm (well within ±10-25mm tolerance)
Ensemble ML model for penetration time and drive duration prediction based on geological conditions.
Key Findings:
- Ensemble model achieves R² = 0.91
- SPT N-Value is most important geological feature (25%)
- Learning curve shows model generalizes well with 800+ samples
Supports all AVN protocols (800, 1200, 2400, 3000) with standardized feature engineering.
Key Findings:
- K-means successfully classifies soil types from operational data
- Feature engineering expands 8 raw inputs to 45 predictive features
- Cross-validation confirms consistent performance across data splits
Calculates optimal slurry flow rates, bentonite injection, and pumping requirements.
Key Findings:
- Optimal slurry density: 1.15 g/cm³
- Flow rate scales with diameter² (double diameter = 4× flow)
- Gravel requires 4× more bentonite than clay (60 vs 15 L/m)
Simulates steering corrections with different strategies and visualizes 3D tunnel paths.
Key Findings:
- Gradual correction (SF=0.6) provides best balance of speed and stability
- Small deviations (0-5mm) have 98% correction success in 3 strokes
- Large deviations (>20mm) drop to 60% success requiring 22 strokes
Implementation of academic research for soil penetration modeling and labor performance analysis.
Key Findings:
- Hegab soil-specific models (R²=0.9369) outperform generic ML (R²=0.9188)
- T×√L transformation is most predictive feature (60% importance)
- Hard soil takes 2.4× longer than soft soil (57 vs 24 min/m)
Log-Logistic probability model for crew productivity estimation.
| Crew Performance | Prep Time per Pipe |
|---|---|
| High (Q1) | ≤ 42 min |
| Typical (Median) | ≤ 53 min |
| Low (Q3) | ≤ 67 min |
# Clone the repository
git clone https://github.com/abdinzaghi5601/MTBM-Machine-Learning.git
# Navigate to project
cd MTBM-Machine-Learning
# Install dependencies
pip install -r ml_requirements.txt
# Generate all visualizations
python generate_all_visualizations.py
# Run Hegab comparison
python hegab_comparison_ml.py| Document | Description |
|---|---|
| ML Visualizations Gallery | Detailed explanation of all 36 graphs with methodology |
| Hegab Model Results | Complete results from Hegab paper implementation |
| Vertical Alignment Guide | Causes and mitigation of alignment deviations |
ML for Tunneling/
├── Core ML Files
│ ├── steering_accuracy_ml.py # Steering prediction model
│ ├── avn3000_predictive_planning_ml.py # Drive time prediction
│ ├── unified_mtbm_ml_framework.py # Multi-protocol framework
│ ├── hegab_comparison_ml.py # Academic paper implementation
│ └── flow_rate_calculator.py # Slurry flow optimization
│
├── Visualization
│ ├── generate_all_visualizations.py # Regenerate all graphs
│ ├── viz_*.png # Generated visualizations
│ └── ML_VISUALIZATIONS_GALLERY.md # Graph explanations
│
├── Steering Tools
│ ├── steering_calculator.py # Steering calculations
│ ├── steering_cli.py # Command-line interface
│ └── steering_correction_simulator.py # Correction simulation
│
├── Documentation
│ ├── HEGAB_MODEL_RESULTS.md
│ ├── VERTICAL_ALIGNMENT_DEVIATION_GUIDE.md
│ └── Various protocol PDFs
│
└── Data & Outputs
├── data/raw/ # Original data files
├── data/processed/ # Cleaned datasets
└── outputs/ # Generated reports
| Model | Purpose | Accuracy |
|---|---|---|
| Random Forest | Steering deviation prediction | R² = 0.89 |
| Ensemble (RF + GB + Ridge) | Drive time estimation | R² = 0.91 |
| K-Means Clustering | Automatic soil classification | 87% |
| Hegab Regression | Soil-specific penetration time | R² = 0.94 |
Survey Position
- Tunnel length, horizontal/vertical deviation, total deviation
Survey Orientation
- Yaw, pitch, reel angle, temperature (ELS/MWD)
Steering Control
- 4 hydraulic cylinder positions, total steering force
Operational
- Advance speed, interjack force, cutter wheel pressure/RPM
| Protocol | Diameter Range | Features |
|---|---|---|
| AVN 800 | 600-900mm | Basic monitoring |
| AVN 1200 | 1000-1400mm | Enhanced sensors |
| AVN 2400 | 1800-2600mm | Advanced analytics |
| AVN 3000 | 2400-3200mm | Full ML integration |
| Category | Deviation | Action |
|---|---|---|
| Excellent | ≤ 25mm | Continue operation |
| Good | 26-50mm | Monitor closely |
| Acceptable | 51-75mm | Begin correction |
| Poor | > 75mm | Stop and assess |
Note: Higher time = Slower progress
| Soil Type | Time (min/m) | Speed (m/hr) |
|---|---|---|
| Soft (A) | 24 | 2.5 (fastest) |
| Medium (B) | 35 | 1.7 |
| Hard (C) | 57 | 1.1 (slowest) |
- Hegab, M. Y., & Smith, G. R. (2006). "Soil Penetration Modeling in Microtunneling Projects"
- Hegab, M. Y., & Smith, G. R. (2009). "Labor Performance Analysis for Microtunneling Projects"
This project is for tunneling operation analysis and optimization.
For questions about interpreting results or operational recommendations, consult with tunnel engineering specialists.
Last Updated: January 2026 Framework Version: 2.0 ML Models: 6 specialized modules Data Parameters: 23+ operational metrics