Federated Digital Twin — Smart Factory Predictive Maintenance Platform

Federated Digital Twin — Smart Factory Predictive Maintenance Platform

Build an intelligent smart-factory monitoring and predictive-maintenance system combining machine learning, Federated Learning, computer vision, real-time monitoring, and blockchain-based model verification.

This project demonstrates how multiple factory nodes can collaboratively train machine-learning models while keeping local training separated, with model updates aggregated through Federated Averaging (FedAvg).

The platform also combines sensor-based Remaining Useful Life (RUL) prediction with visual surface-defect classification.

---

Core Capabilities

Predictive Maintenance

The platform uses a PyTorch LSTM model to estimate Remaining Useful Life (RUL) from machine sensor data.

Supported workflow:

Sensor data → preprocessing → LSTM inference → RUL prediction → dashboard visualization

The implementation supports NASA C-MAPSS sensor datasets and also provides synthetic-data training for quick demonstrations.

Visual Defect Detection

An EfficientNet-B0 model with transfer learning is used for visual surface-defect classification.

The system supports six defect classes using the NEU Surface Defect Database.

Workflow:

Factory image → preprocessing → EfficientNet-B0 → defect classification → API response → dashboard

---

Federated Learning

The project includes a manual Federated Learning implementation using FedAvg.

Multiple simulated factory nodes independently train local models before their model updates are aggregated.

Example architecture:

Node 1 ─┐ Node 2 ─┤ Node 3 ─┼→ FedAvg Aggregation → Global Model Node 4 ─┘

This provides a practical demonstration of collaborative model training without directly centralizing each node's local training process.

The implementation does not require an external Federated Learning framework.

---

Blockchain Verification

The platform includes a simulated SHA-256 proof-of-work blockchain for recording model-related information.

Recorded information can include:

• Model hash
• Training metrics
• Timestamp
• Previous block hash

The blockchain API also provides block verification functionality.

An optional Ethereum/Ganache integration is provided through Web3.py.

Important: the default implementation is a simulated blockchain intended for demonstration and development purposes, rather than a production blockchain network.

---

Real-Time Monitoring

The frontend communicates with the Flask backend through HTTP APIs and WebSockets.

Real-time events include:

• Sensor readings
• RUL updates
• Federated Learning progress
• Node status
• Blockchain updates

This allows the dashboard to display changing factory and training information without requiring constant page refreshes.

---

Dashboard

The React frontend provides an interactive interface for monitoring:

• Factory overview
• RUL predictions
• Defect detection
• Federated Learning status
• Training metrics
• Factory node status
• Blockchain records
• Blockchain verification

Charts and visualizations are implemented using Recharts.

---

Technology Stack

Frontend

• React 18
• TypeScript
• Vite
• Tailwind CSS
• shadcn/ui
• Recharts

Backend

• Python
• Flask
• Flask-SocketIO
• REST APIs
• WebSockets

Machine Learning

• PyTorch
• LSTM
• EfficientNet-B0
• Transfer Learning

Federated Learning

• Federated Averaging (FedAvg)
• Manual implementation
• Simulated factory nodes

Blockchain

• SHA-256
• Proof of Work
• Simulated blockchain
• Web3.py
• Optional Ethereum/Ganache integration

Datasets

• NASA C-MAPSS
• NEU Surface Defect Database

---

API

System

GET /api/health

Returns system health and model status.

Predictive Maintenance

POST /api/predict/rul

Generate an RUL prediction from uploaded sensor data.

Defect Detection

POST /api/predict/defect

Classify a surface-defect image.

Federated Learning

POST /api/federated/start

Start federated training.

GET /api/federated/status

Retrieve federated-training progress.

Factory Nodes

GET /api/nodes/status

Retrieve the current status of simulated factory nodes.

Blockchain

GET /api/blockchain/logs

Retrieve blockchain transaction records.

GET /api/blockchain/verify/<hash>

Verify a blockchain record.

---

Quick Start

Backend

Install Python dependencies:

pip install -r requirements.txt

Frontend

cd frontend npm install

Train Models

The project includes synthetic-data training:

python -m training.federated_training --model both --synthetic

Run Backend

python app.py

Run Frontend

cd frontend npm run dev

The application can then be accessed through the local development server.

---

Real Dataset Training

NASA C-MAPSS

Place the C-MAPSS datasets inside:

data/sensor/

Then run:

python -m training.train_rul --data_dir data/sensor --epochs 50

Federated training is also supported.

NEU Surface Defect Database

Place the image classes inside:

data/images/

Then run:

python -m training.train_defect --data_dir data/images --epochs 20

---

Who Is This For?

This project is suitable for:

• AI/ML developers
• Industrial AI developers
• Predictive-maintenance researchers
• Computer-vision developers
• Smart manufacturing projects
• Industry 4.0 / Industry 5.0 demonstrations
• Federated Learning research and experimentation
• Academic research and demonstrations
• AI proof-of-concept development
• Developers building smart-factory prototypes

---

What You Receive

• Complete project source code
• React frontend
• Flask backend
• Machine-learning models and training code
• Federated Learning implementation
• Blockchain implementation
• REST API implementation
• WebSocket implementation
• Synthetic-data training support
• Dataset integration structure
• Setup instructions
• API documentation

---

Requirements

A basic demonstration can be run using synthetic data.

For practical model training, an NVIDIA GPU is recommended.

Real-world training performance depends on the selected dataset, batch size, model configuration, image resolution, and available hardware.

---

License

Commercial Single-Project License.

The purchased source code may be used and modified for one project.

Redistribution, resale, sublicensing, or publishing the source code as a standalone product is not permitted.

Custom development, deployment, hosting, dataset preparation, and ongoing maintenance are not included.