🛠️ GARNET: AI-Driven P&ID Symbol Detection and Analysis

GCME AI-Recognition Network for Engineering Technology
Precision in Every Connection

GARNET is an AI-powered tool designed to automate symbol detection, classification, and connectivity analysis in Piping and Instrumentation Diagrams (P&IDs). Built for engineers and maintenance teams, it combines state-of-the-art object detection (YOLOv8) with graph-based analytics to transform P&ID workflows.

---

🚀 Features

• Symbol Detection: Identify valves (gate, globe, check), pumps, tanks, and more using YOLOv8.
• Automated Counting: Generate counts for each symbol type in a P&ID.
• Graph-Based Analysis: Model P&IDs as networks to analyze connectivity, critical paths, and system dependencies.
• Export Results: Export detection results and connectivity graphs to CSV, PDF, or JSON.
• User-Friendly Interface: Simple CLI and scriptable API for integration into existing workflows.

---

📦 Installation

1. Clone the Repository:

   git clone https://github.com/your-username/GARNET.git
   cd GARNET

2. Install Dependencies:

   pip install -r requirements.txt

---

🖥️ Usage

1. Web Application (Interactive Inference)

Run the web interface to upload P&IDs, select models, and view results in real time:

1. Start the Web Server:

   uvicorn main:app --reload

   Access the app at http://localhost:8000.

2. Using the Web Interface:

   • Upload a P&ID: Select an image file (JPG/PNG).
   • Model Configuration: Choose a model type (e.g., YOLOv5, YOLOv8) and upload custom weights (.pt file).
   • Run Inference: Click "Submit" to detect symbols and display results.
   • Results: View annotated images, symbol counts, and download reports (CSV/JSON).

---

2. Batch Inference Script

Run inference on multiple P&IDs in a folder using predict_images.py:

1. Command-Line Arguments:

   python predict_images.py \
       --image_path path/to/pids_folder \
       --model_type yolov8 \
       --model_path path/to/model_weights.pt \
       --output_path results/

2. Output:

   • Annotated images (saved in output_path).
   • CSV file with symbol counts (output_path/symbol_counts.csv).

Example code snippet:

from garnet.inference import predict_images

predict_images(
    image_path="path/to/pids_folder",
    model_type="yolov8",
    model_path="path/to/model_weights.pt",
    output_path="results/"
)

---

3. Pipeline: End-to-End P&ID Digitization and Connectivity Analysis

Based on our reference methodology, the following pipeline will be implemented:

1. Preprocessing

   • Load and normalize the P&ID image.
   • Optional: Denoising, binarization, and removal of scanning artifacts.

2. Object Detection

   • Detect P&ID symbols using YOLOv11/YOLOv8.
   • Classes include valves, pumps, tanks, instruments, and custom symbols.
   • Export detections as bounding boxes (YOLO or COCO format).

3. Text Recognition

   • Use EasyOCR or PaddleOCR for text extraction (supports horizontal and vertical text).
   • Merge OCR results from multiple rotations.
   • Output text annotations (JSON).

4. Line Extraction

   • Remove detected symbols and text areas from the image to isolate pipelines.
   • Apply morphological operations to extract lines (handles 90° turns and intersections).
   • Merge fragmented line segments.

5. Line-Symbol Connection

   • Detect intersection points between line segments and symbol bounding boxes.
   • Assign connectivity relationships between symbols and lines.

6. Graph Construction

   • Convert connected symbols and pipelines into a NetworkX graph.
   • Nodes = symbols (with attributes like tag, type).
   • Edges = pipelines (with attributes like length, type).

7. Graph Analysis

   • Perform connectivity analysis: critical paths, loops, flow direction inference.
   • Detect isolated components or redundant loops.

8. DEXPI Export

   • Convert the annotated P&ID into DEXPI-compliant XML.
   • Include geometry, symbol metadata, and connectivity.

9. Visualization and Reporting

   • Overlay detected objects, text, and pipelines on the original P&ID.
   • Export results as PNG/PDF/JSON.
   • Provide interactive graph visualization (future feature).

Note: This pipeline is designed to be modular, so each step can be run independently or as part of the full digitization workflow.

Reference: Adapted from "End-to-End Digitization of Image Format Piping and Instrumentation Diagrams" (2024).

Note: A staged, runnable pipeline with CLI is available in garnet/pid_extractor.py. See garnet/README_pid_pipeline.md for usage, stages, tuning knobs, and output artifacts.

---

4. Model Training (Optional)

To train custom YOLO models for P&ID symbols:

yolo train \
    data=data.yaml \
    model=yolov8n.pt \
    epochs=100 \
    imgsz=640 \
    batch=16

📂 Dataset

GARNET uses the YOLOv8 dataset format. Example structure:

dataset/
├── train/
│   ├── images/  # P&ID images (.jpg, .png)
│   └── labels/  # YOLO-format labels (.txt)
├── val/
│   ├── images/
│   └── labels/
└── data.yaml     # Dataset config (class names, paths)

Example data.yaml:

train: dataset/train/images
val: dataset/val/images

nc: 6 # Number of classes
names: ["valve", "gate_valve", "globe_valve", "check_valve", "pump", "tank"]

---

📊 Results

Detection	Graph Analysis

Example output: Symbol counts and connectivity graph for a P&ID.

---

📈 Future Outcomes

Additional planned outcomes from the GARNET project include:

• MTO for Valves: Automated generation of material take-off lists for all detected valve types, including specifications and quantities.
• Line List: Extraction and tabulation of pipeline data, including line tags, sizes, service, and connected equipment.

---

🤝 Contributing

Contributions are welcome! Please fork the repository and submit a pull request.
For major changes, open an issue first to discuss your ideas.

---

📜 License

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

---

📧 Contact

For questions or collaborations, contact:

• Your Name - [email protected]
• GCME (GC Maintenance and Engineering Co., Ltd.) - www.gcme.com