Skip to content

ericyoc/drone_opt_path_gas_leak_poc

Repository files navigation

Hybrid Quantum-Classical Path Optimization for Multi-Drone Gas Leak Detection

Python 3.8+

This repository implements a Hybrid Partition-and-Routing Heuristic that combines classical machine learning with quantum-inspired optimization (QUBO) to coordinate drone fleets for industrial gas leak detection.

The framework optimizes for Time-to-Detection (TTD), significantly outperforming standard coverage-based methods in emergency response scenarios.


🛠 System Architecture

The system follows a modular hybrid pipeline:

  1. Quantum-Inspired Classifier: Uses Angle Encoding and variational circuits to identify leak severity from raw sensor data.
  2. Quantum Voronoi Partitioning: Formulates a Quadratic Unconstrained Binary Optimization (QUBO) problem to assign drone deployment sites based on the severity heatmap.
  3. Classical Path Generation: Utilizes a 2-Opt TSP heuristic to plan the exact flight path for each drone within its assigned partition.

📊 Data Sources & Methodology

This project utilizes a two-tier validation strategy to transition from theoretical optimization to industrial application.

1. Stage I: Synthetic Validation

Initial benchmarking was performed using a synthetic "Hazard Topology" in a simulated $1000m \times 1000m$ environment.

  • Purpose: To isolate the performance of the QUBO solver and verify convergence across different fleet sizes (5–20 drones).
  • Generation: Multi-modal Gaussian distributions create "ground truth" leak intensities.

2. Stage II: Real-World Proxy (GPLA-12)

Final validation uses the GPLA-12 Acoustic Signal Dataset to provide realistic hazard inputs.

  • Source: Collected by researchers at the University of Adelaide. [Reference: Li & Yao, 2021]
  • Composition: 684 samples across 12 classes simulating various pipeline pressures and leakage types.
  • Role: Acoustic signals are processed through the hybrid ensemble to generate the severity scores that drive the spatial optimization.

📈 Performance Results

The framework was benchmarked against three classical baselines. Method 3 (M3) is a high-performance, intensity-weighted K-Means algorithm used to ensure a fair comparison with the proposed model.

Comparative Analytics (7-Drone Fleet)

Metric M1 (Random) M2 (K-Means) M3 (IW K-Means) Hybrid Q-C (Ours)
Mean TTD (s) 66.2s 45.1s 43.9s 10.0s
High-Risk TTD (s) 64.8s 42.1s 40.7s 9.4s
Compute Time (s) 0.02s 1.80s 2.67s 0.21s
Speedup vs M3 - - 1.0x 12.7x

Statistical Note: Results are averaged over $n=10$ runs with fixed seeds (numpy=42, tensorflow=42). The speedup is statistically significant ($p \le 0.0002$ via Mann-Whitney U tests).


📽 Visualizations

The system provides several layers of visualization to interpret the optimization results:

  • Hazard Heatmaps: Visualizes the density of the GPLA-12 derived leak spots.
  • Voronoi Cells: Displays the QUBO-derived boundaries for drone responsibility.
  • Mission Animation: A real-time .gif showing drones traversing optimized paths.

(Note: In the provided source code, animations are saved as drone_mission_animation.gif for mission debriefing.)


🚀 Installation & Usage

1. Clone the Repository

git clone https://github.com/yourusername/hybrid-gas-leakage-detection.git
cd hybrid-gas-leakage-detection

2. Install Requirements

pip install -r requirements.txt

3. Run the Simulation

The full end-to-end pipeline (data processing → optimization → animation) is contained in the main script:

python hybrid_gas_leakage_system.py

📜 Citations & Credits

Original Dataset

If you use the GPLA-12 data, please cite:

Li, J., & Yao, L. (2021). GPLA-12: An Acoustic Signal Dataset of Gas Pipeline Leakage. arXiv preprint arXiv:2106.10277. Source


⚠️ Disclaimer

This project is an algorithmic proof-of-concept. All "Quantum" components are currently executed via classical simulation of quantum heuristics. It is intended for academic research and is not certified for deployment in safety-critical industrial infrastructure.

📄 License

This project is licensed under the MIT License. See the LICENSE file for details.