Agentic AI workflow for Predictive Maintenance

[vikalluru]

Agentic AI Workflow for Predictive Maintenance

Summary

Implementation of a multi-agent predictive maintenance system using NVIDIA AIQ Toolkit for turbofan engine RUL (Remaining Useful Life) prediction based on NASA C-MAPSS dataset.

Architecture

• React Agent: Main workflow orchestration using ReAct pattern with NIM LLM integration
• SQL Retriever Tool: Automated SQL query generation with ChromaDB vector database for schema retrieval
• RUL Prediction Tool: XGBoost regression model with StandardScaler preprocessing
• Plotting Agent: Multi-tool visualization agent supporting distribution, comparison, and time-series plots

Components Added

industries/manufacturing/predictive_maintenance_agent/
├── src/predictive_maintenance_agent/
│   ├── configs/config.yml                    # Agent configuration
│   ├── predict_rul_tool.py                  # RUL prediction implementation
│   ├── plot_*.py                           # Visualization tools
│   └── generate_sql_query_tool.py          # SQL generation with RAG
├── models/                                  # Pre-trained XGBoost and scaler
├── database/                               # SQLite database and setup
├── imgs/                                   # Test execution screenshots
└── README.md                              # Setup and usage documentation

Dependencies

• Python 3.11+
• NVIDIA AIQ Toolkit
• XGBoost, scikit-learn, pandas, numpy
• ChromaDB for vector storage
• Vanna for SQL retrieval
• Optional: Phoenix for observability

Testing

Validated with three test scenarios:

• RUL distribution analysis across engine units
• Sensor trend visualization over operational cycles
• Predictive accuracy comparison (actual vs predicted RUL)

Contributors: Vineeth Kalluru, Janaki Vamaraju, Sugandha Sharma, Ze Yang, Viraj Modak