Smart Manufacturing Multi-Agent System (MAS)

An intelligent Multi-Agent System for predictive maintenance and optimization in smart manufacturing. This system uses LLM-powered orchestration to automatically load, preprocess, analyze, and generate prescriptive recommendations from manufacturing data.

🚀 Features

• LLM-Powered Orchestration: Intelligent workflow planning using Google Gemini or local LLMs (Ollama)
• Adaptive Intelligence: Automatic model selection and performance optimization
• Intelligent Preprocessing: Automatic feature analysis, encoding, and data quality handling
• Multi-Model Analysis: Supports classification, regression, and anomaly detection
• Prescriptive Recommendations: Actionable maintenance suggestions with priority ranking
• Human-in-the-Loop: Interactive approval workflow for critical decisions
• Comprehensive Logging: Detailed audit trails and performance metrics

✨ Recent Improvements

Bug Fixes & Enhancements (Latest Release)

1. Fixed One-Hot Encoding Issue: Protected identifier columns (e.g., Machine_ID) are now properly handled as pass-through features instead of being one-hot encoded, reducing feature explosion and improving performance.

2. Fixed ID Column Handling: All models now automatically drop ID columns during training while preserving them for recommendations and reporting.

3. Enhanced Contributing Factors: All priority levels (Critical, Medium, Low) now show detailed contributing factors based on actual feature values instead of generic messages.

4. Improved Workflow Completion: Fixed LLM validation logic to properly handle workflow completion signals.

5. Better Data Flow Logic: Improved tracking of data shapes through the entire pipeline for better debugging and transparency.

📋 Prerequisites

• Python: 3.8 or higher
• API Key: Google Gemini API key (optional if using local LLMs)
• Ollama: Optional, for local LLM support

🛠️ Installation

1. Clone the Repository

git clone <repository-url>
cd smart_manufacturing_mas_code

2. Create Virtual Environment

# Create virtual environment
python3 -m venv mas_venv

# Activate virtual environment
# On macOS/Linux:
source mas_venv/bin/activate
# On Windows:
mas_venv\Scripts\activate

3. Install Dependencies

pip install -r requirements.txt

4. Set Up Environment Variables (Optional)

If using Google Gemini, create a .env file:

echo "GEMINI_API_KEY=your_gemini_api_key_here" > .env

5. Install Ollama (Optional - for Local LLMs)

# Install Ollama
curl -fsSL https://ollama.ai/install.sh | sh

# Pull a model (example with Qwen3)
ollama pull qwen3:4b

🎯 Quick Start

Interactive Mode (Recommended for First-Time Users)

python3 main_llm.py

This will guide you through:

• Dataset selection
• Feature and target selection
• Problem type identification
• Approval of critical decisions

Automated Mode

python3 main_llm.py --auto --dataset "data/Smart Manufacturing Maintenance Dataset/smart_maintenance_dataset.csv"

With Local LLM

python3 main_llm.py --decision-llm ollama --decision-model qwen3:4b

📖 Usage Examples

Basic Analysis

# Run with default Gemini LLM
python3 main_llm.py

# Run with auto schema discovery
python3 main_llm.py --auto --dataset "path/to/dataset.csv"

# Run with local LLM
python3 main_llm.py --decision-llm ollama --decision-model qwen3:4b

# Batch process all datasets
python3 main_llm.py --batch

Advanced Configuration

# Custom LLM configuration
python3 main_llm.py \
    --planner-llm gemini \
    --decision-llm ollama \
    --decision-model llama3:8b \
    --auto

# Specific dataset with auto mode
python3 main_llm.py \
    --auto \
    --dataset "data/Intelligent Manufacturing Dataset/manufacturing_6G_dataset.csv"

🧠 Architecture

Agents

1. LLM Planner Agent: Orchestrates workflow using LLM reasoning
2. Data Loader Agent: Loads and inspects datasets
3. Preprocessing Agent: Intelligent feature analysis and data preparation
4. Dynamic Analysis Agent: Multi-model analysis with automatic selection
5. Optimization Agent: Generates prescriptive recommendations

Key Components

• Schema Discovery: Automatic dataset understanding
• Tool Decider: Intelligent preprocessing/model selection
• Adaptive Intelligence: Performance-based model switching
• Intelligent Summarization: Clean output with full logging

📊 Supported Problem Types

• Classification: Predict categorical outcomes (e.g., maintenance priority)
• Regression: Predict continuous values (e.g., failure probability)
• Anomaly Detection: Identify unusual patterns (no target needed)

🔧 Configuration

Command Line Options

Option	Description
--planner-llm	LLM backend for planner (gemini, ollama, mock)
--planner-model	Model name for planner
--decision-llm	LLM backend for decisions (ollama, mock, or None)
--decision-model	Model name for decisions (e.g., qwen3:4b)
--dataset	Path to CSV dataset
--auto	Enable auto mode with schema discovery
--batch	Process all datasets in data/ directory
--interface	HITL interface (cli or web)

📁 Project Structure

smart_manufacturing_mas_code/
├── agents/                    # Core agent implementations
│   ├── llm_planner_agent.py   # LLM orchestration
│   ├── data_loader_agent.py   # Data loading
│   ├── preprocessing_agent.py # Preprocessing
│   ├── dynamic_analysis_agent.py # Analysis
│   └── optimization_agent.py  # Recommendations
├── utils/                     # Utility modules
│   ├── schema_discovery.py    # Auto schema detection
│   ├── tool_decider.py        # Model/tool selection
│   ├── hitl_interface.py      # Human-in-loop UI
│   └── reporting.py           # Logging and reporting
├── data/                      # Sample datasets
│   ├── Smart Manufacturing Maintenance Dataset/
│   └── Intelligent Manufacturing Dataset/
├── documentation/             # Documentation
│   ├── usage_guide.md         # Detailed usage guide
│   ├── architecture_and_workflow.md
│   └── adaptive_intelligence_system.md
├── logs/                      # Generated logs
├── main_llm.py               # Main entry point
└── requirements.txt          # Dependencies

📈 Output

Console Output

The system provides real-time progress updates and a final intelligent summary.

Log Files

All runs generate:

• logs/workflow_report_*.json: Detailed workflow report
• logs/detailed_results_*.json: Complete structured data
• logs/hitl_audit.json: Human approval audit trail

🐛 Troubleshooting

Common Issues

1. Import Errors

# Make sure virtual environment is activated
source mas_venv/bin/activate

# Verify installation
pip list | grep scikit-learn

2. API Key Issues

# Check .env file exists
cat .env

# Verify API key is valid
python -c "import google.generativeai as genai; genai.configure(api_key='your_key')"

3. Ollama Connection Issues

# Start Ollama service
ollama serve

# Test model availability
ollama list

4. Dataset Issues

# Verify dataset path
ls -la data/

# Check CSV format
head -5 data/Smart\ Manufacturing\ Maintenance\ Dataset/smart_maintenance_dataset.csv

Debug Mode

Enable detailed logging:

export LOG_LEVEL=DEBUG
python3 main_llm.py --auto

🎓 Learn More

• Detailed Usage Guide
• Architecture and Workflow
• Adaptive Intelligence System

🤝 Contributing

Contributions are welcome! Please follow these guidelines:

1. Create a feature branch
2. Add tests for new functionality
3. Update documentation
4. Submit a pull request

📝 License

[Add your license information here]

🙏 Acknowledgments

This project is developed for research in intelligent manufacturing and predictive maintenance systems.

🔗 Contact

[Add contact information]

---

Ready to get started? Run python3 main_llm.py to begin your first analysis!