Edge App - Fault Detection System with Node-RED and TensorFlow Lite

This project demonstrates an edge computing setup that simulates sensor data, performs fault detection using a TensorFlow Lite model, and visualizes the results using Node-RED. Designed for deployment on the IXON SecureEdge Pro, this solution showcases how to integrate machine learning and visualization tools at the edge for real-time diagnostics.

Key components of this project include:

1. A training environment to generate synthetic data and train a TensorFlow model for fault detection.
2. A lightweight TensorFlow Lite inference server for efficient fault prediction at the edge.
3. A Node-RED-based dashboard to simulate sensor data and display predictions in real-time.

This project provides a robust foundation for implementing advanced edge analytics and proactive maintenance strategies in industrial IoT environments. Users can adapt and extend the solution for their specific needs.

Prerequisites

Ensure your environment is properly set up by following this guide: Running custom Docker applications on the SecureEdge Pro.

Steps to Deploy the Fault Detection System

1. Train the Model

On Unix-based systems:

Run the train_model.sh script from the project root directory:

./train_model.sh

On Windows:

Run the train_model.cmd script from the project root directory:

train_model.cmd

These scripts will:

• Build the training container.
• Generate synthetic data.
• Train a TensorFlow model and convert it to TensorFlow Lite format.
• Copy the generated .tflite model to the tensor-flow-lite directory for deployment.

2. Build and Push Containers

Run the build_and_push_containers.sh script to build and push all project containers:

For Unix-based systems:

./build_and_push_containers.sh

For Windows:

build_and_push_containers.cmd

3. Deploy the Containers on SecureEdge Pro

TensorFlow Lite Inference Container

• Access the local web interface of the SecureEdge Pro.
• Create a tensor-flow-lite container using the tensor-flow-lite image with the following settings:
  • Port Mapping: 5050:5050

Refer to the screenshot for configuration details:

Node-RED Container

• Create a node-red container using the node-red image with the following settings:
  • Port Mapping: 1880:1880
  • Volume Mapping: node-red-data -> /data

Refer to the screenshot for configuration details:

4. Start the Containers

• Wait for the containers to be created and start them via the SecureEdge Pro interface.

5. Access Node-RED Dashboard

• In IXON Cloud, create an HTTP Web Server to access the Node-RED dashboard on the SecureEdge Pro.

Refer to the screenshot for configuration details:

• Open the Node-RED dashboard to configure the flows and simulate sensor data.

6. Access Node-RED Virtual HMI

• In IXON Cloud, create an HTTP Web Server to access the Node-RED Virtual HMI on the SecureEdge Pro.

Refer to the screenshot for configuration details:

• Open the Node-RED Virtual HMI to monitor the system.

7. Access TensorFlow Lite Inference API

• The TensorFlow Lite Flask API will be available at http://<gateway-ip>:5050/predict. Node-RED sends simulated sensor data to this endpoint for inference.

8. Troubleshooting

• Ensure containers are running and accessible.
• Check the Node-RED debug nodes and logs for errors.
• Verify that the TensorFlow Lite model is loaded and accessible in the inference container.

Customization

• Model Threshold/Probability: Adjust the fault probability in generate_dataset.py to better simulate real-world scenarios.
• Model Architecture: Modify the model in train.py for improved accuracy or performance.
• Node-RED Flows: Customize the flows.json file to integrate additional logic or sensor inputs.

Conclusion

This project provides a complete edge computing solution for real-time fault detection, serving as a starting point for implementing advanced IoT diagnostics and analytics. With easy deployment on IXON's SecureEdge Pro, it enables users to leverage edge-level AI and visualization tools for enhanced operational efficiency.