LangGraph-predictive-maintenance-agent

# LangGraph AI Agent for Predictive Maintenance on Time-Series Sensor Data

This project implements an AI-powered diagnostic assistant for industrial time-series sensor data.

The system analyzes Turbofan engine telemetry to detect abnormal engines, identify degrading sensors, and predict which engines will fail soon.

The system combines machine learning, statistical analysis, and an LLM agent that can answer natural language questions about fleet health.

---

# Dataset

NASA Turbofan Engine Degradation Dataset

Each engine has multiple sensors recorded over time until failure.

Variables include:

- engine_id
- cycle
- 3 operational settings
- 21 sensor measurements

---

# System Pipeline

Dataset  
→ Feature Engineering  
→ Degradation Analysis  
→ Anomaly Detection  
→ RUL Prediction Model  
→ AI Diagnostic Agent  

---

# Features

### Exploratory Data Analysis

### Distribution of Engine Lifetimes

<p align="center">
  <img src="Visualizations/fig_DistributionOfEngineLifeCycles.png" width="700">
</p>

The histogram shows the distribution of engine lifetimes across the fleet.  
Most engines fail between **180–230 cycles**, while a few survive significantly longer.

This variability motivates the use of predictive maintenance models to estimate **Remaining Useful Life (RUL)** based on sensor behavior.

### Feature Engineering

- Remaining Useful Life (RUL) calculation
- Rolling window statistics
- Sensor degradation slopes

### Analytical Models

Anomaly detection using **Isolation Forest**

Predictive maintenance using **Random Forest Regressor**

### Diagnostics

The agent can answer questions such as:

```
Which engines behave abnormally?
Which engines will fail soon?
Generate a fleet diagnostic report
Which sensors degrade the most?
```

### Visualization

![Sensor degradation](Visualizations/fig_SensorDegradation.png)

Example of sensor degradation trajectory for a single engine.


![Correlation heatmap](Visualizations/fig_CorrelationHeatmap.png)

Correlation analysis between engine sensors helps identify redundant or related measurements.

Anomaly inspection  

Fleet-wide sensor trends

# Example Agent Output

```
Which engines will fail soon?

Agent Response:

The engines that are predicted to fail soon are: engine 6, engine 19, engine 96, engine 46, and engine 99.
```

```
Generate a fleet diagnostic report

The fleet diagnostic report indicates:

Anomalous engines:
[9, 61, 66, 82, 91]

Strong degradation sensors:
sensor9, sensor14, sensor4, sensor3

Engines likely to fail soon:
96, 6, 19, 76, 92
```

---

# Technologies

Python  
Pandas  
Scikit-Learn  
Matplotlib  
LangChain  
LangGraph  
Ollama (local LLM)

---

# Running the Project

Clone the repository:

```
git clone https://github.com/VaishnaviNarasimhaiahSathosh/LangGraph_predictive_maintenance.git
```

Create virtual environment:

```
python -m venv venv
source venv/bin/activate
```

Install dependencies:

```
pip install -r requirements.txt
```

Run the system:

```
python src/main.py
```

Then interact with the diagnostic agent:

```
Ask a question about the engine fleet:
```

---

# Project Highlights

• Full ML pipeline from raw sensor data to predictive maintenance insights  
• Modular code structure for experimentation  
• Interactive AI agent capable of reasoning over analytical tools  
• Visual diagnostics for fleet monitoring  

---

# License

MIT License