🏆 Final Year Capstone Project | Thapar Institute of Engineering & Technology
January 2025 - November 2025
Transforming reactive maintenance into predictive intelligence through IoT + AI
- The Problem
- Our Solution
- Live Demonstration
- System Architecture
- AI Detection System
- Technology Stack
- Project Journey
- Meet the Team
- Impact & Results
- Future Vision
Solar panels don't fail suddenly — they die slowly, bleeding efficiency while owners remain unaware.
| Issue | Efficiency Loss | Traditional Detection Time |
|---|---|---|
| ☀️ Dust Accumulation | 7-15% | 2-4 weeks |
| 🐦 Bird Droppings | 5-10% | 1-3 weeks |
| 🌤️ Partial Shading | 10-25% | 2-6 weeks |
| 🔥 Hotspots | Panel damage + fire risk | Often never |
| ❄️ Weather Damage | Permanent cell damage | Too late |
15-25% of solar efficiency lost annually
$200-500 revenue loss per residential installation
$50-100 cost per manual inspection
2-4 weeks average detection delay
Most users only discover problems when:
- ⚡ Energy bills spike unexpectedly
- 📉 Monthly reports show reduced output
- 💸 Significant revenue has already been lost
From Reactive to Predictive: Real-time Monitoring Meets AI Intelligence
| Traditional Approach | Solar IQ Approach |
|---|---|
| ❌ Manual monthly inspections | ✅ 24/7 automated monitoring |
| ❌ Issues found after revenue loss | ✅ Real-time issue detection |
| ❌ Expensive site visits ($50-100) | ✅ Remote AI analysis (<$1/month) |
| ❌ 2-4 week detection delay | ✅ 15-minute alert delivery |
| ❌ Human eye inspection | ✅ 89-94% AI accuracy |
| ❌ Reactive maintenance | ✅ Predictive scheduling |
🔍 Real-Time Sensor Monitoring
ESP32-powered network samples environmental data every 30 seconds
🤖 AI Visual Inspection
Custom Vision Transformer analyzes panel images with 89-94% accuracy
📱 Instant SMS Alerts
Twilio delivers actionable insights within seconds of detection
📊 Live Dashboard
Firebase-powered interface for comprehensive real-time monitoring
📈 Predictive Maintenance
Schedule cleaning before efficiency drops significantly
Complete IoT Setup: Solar Panel + ESP32 + Sensors + Wi-Fi Camera
Our compact test environment features:
- Loom Solar PV Panel - Test subject for monitoring
- ESP32 Microcontroller - Brain of the operation
- DS18B20 Sensor - Panel surface temperature tracking
- DHT11 Sensor - Ambient conditions monitoring
- Voltage Sensor - Electrical output measurement
- CP PLUS 3MP Camera - Visual condition capture
Vision Transformer Model Classifying Panel Conditions with High Confidence
Our AI model successfully detects:
- ✅ Bird Droppings - 94% confidence
- ✅ Electrical Damage - 89% confidence
- ✅ Physical Damage - 93% confidence
- ✅ Clean Panels - 85-95% confidence
Real SMS Alerts Delivered to Users' Phones
Sample Alert Messages:
✅ Status: Clean - No maintenance required
Time: 28/11/2025, 15:49:57
Confidence: 92%
🟡 Status: Dust - Cleaning required
Time: 15 Nov 2025, 11:00
Confidence: 87%
Action: Schedule cleaning within 48 hours
Web Dashboard Tracking Maintenance Actions Over Time
Dashboard Features:
- Date and time-based filtering
- Detection category breakdown
- Required action status
- Completion tracking
- Historical trend analysis
┌────────────────────────────────────────────────────────────────┐
│ EDGE LAYER │
│ ┌──────────┐ ┌──────────┐ ┌──────────┐ ┌────────────┐ │
│ │ DS18B20 │ │ DHT11 │ │ Voltage │ │ CP PLUS │ │
│ │ Temp │ │ Temp+Hum │ │ Sensor │ │ Camera │ │
│ └─────┬────┘ └─────┬────┘ └─────┬────┘ └──────┬─────┘ │
│ └──────────────┴───────────────┴────────────────┘ │
│ │ │
│ ┌───────▼────────┐ │
│ │ ESP32/ESP8266 │ │
│ │ Controller │ │
│ └───────┬────────┘ │
└───────────────────────────┼─────────────────────────────────────┘
│ WiFi
▼
┌────────────────────────────────────────────────────────────────┐
│ CLOUD LAYER │
│ ┌────────────────┐ │
│ │ Firebase │ │
│ │ Realtime DB │ │
│ └────────┬───────┘ │
│ │ │
│ ┌──────────────────┼──────────────────┐ │
│ ▼ ▼ ▼ │
│ ┌──────────┐ ┌──────────┐ ┌──────────┐ │
│ │ Sensors │ │ Images │ │ Alerts │ │
│ │ Data │ │ Metadata │ │ Logs │ │
│ └──────────┘ └──────────┘ └──────────┘ │
└────────────────────────────────────────────────────────────────┘
│
┌─────────────┼─────────────┐
▼ ▼ ▼
┌──────────┐ ┌──────────┐ ┌──────────┐
│ Node.js │ │ React │ │ Vision │
│ Backend │ │Dashboard │ │Transform.│
└─────┬────┘ └──────────┘ └─────┬────┘
│ │
│ ┌─────────────────┘
▼ ▼
┌─────────────────┐
│ Twilio SMS API │
│ Alert System │
└─────────────────┘
| Component | Model | Specification | Purpose |
|---|---|---|---|
| Microcontroller | ESP32 | WiFi, 32-bit dual-core | Data processing & transmission |
| Panel Temp Sensor | DS18B20 | Digital, ±0.5°C accuracy | Surface temperature monitoring |
| Environment Sensor | DHT11 | Temperature + Humidity | Ambient conditions tracking |
| Voltage Sensor | Analog | 0-25V range | Electrical output monitoring |
| Camera | CP PLUS 3MP | Wi-Fi PTZ, 3MP resolution | Visual condition capture |
| Solar Panel | Loom Solar | Polycrystalline | Test subject panel |
Data Collection:
- Sensors sample every 30 seconds
- Camera captures images every 15 minutes
- All data streams to Firebase in real-time
- ESP32 processes and validates locally
Our custom Vision Transformer (ViT) model leverages transfer learning to achieve exceptional accuracy in solar panel condition classification.
Model Specifications:
Base Model: google/vit-base-patch16-224
Framework: PyTorch + Hugging Face Transformers
Input Size: 224×224 RGB images
Output: Multi-class classification with confidence scores
Training Data: Custom annotated solar panel dataset (500+ images)
Threshold: 75% confidence for alert triggering
Accuracy: 89-94% across all condition categories
| Condition | Description | Confidence Range | Alert | Action Required |
|---|---|---|---|---|
| ✅ Normal | Clean panel, optimal operation | 85-95% | No | Continue monitoring |
| 🟡 Dust | Accumulated dust layer | 87-94% | Yes | Schedule cleaning (48hrs) |
| ⚫ Shading | Partial shadow coverage | 85-92% | Yes | Trim vegetation/adjust |
| 🔴 Bird Droppings | Localized obstruction | 89-94% | Yes | Immediate spot cleaning |
| Cracks, delamination | 91-93% | Critical | Inspect & repair | |
| ⚡ Electrical Damage | Hotspots, cell failure | 88-89% | Critical | Technical service |
Live System Performance:
- Bird drop detection: 94% confidence ✅
- Electrical damage: 89% confidence ✅
- Physical damage: 93% confidence ✅
- Clean panel recognition: 92% average ✅
Model Training Pipeline:
1. Data Collection: 500+ annotated solar panel images
2. Preprocessing: Resize to 224×224, normalize, augment
3. Transfer Learning: Fine-tune ViT-Base on our dataset
4. Training: 20 epochs with AdamW optimizer
5. Validation: Cross-validation on test set
6. Deployment: PyTorch model served via Flask APIImage Capture (15min) → ViT Analysis → Confidence Check (>75%)
↓ ↓
Firebase Storage Alert Triggered
↓ ↓
Metadata Logged Twilio SMS Sent
↓ ↓
Dashboard Update Alert Logged
✅ 30-Minute Cooldown - Prevents alert spam
✅ Confidence Scores - Transparency in detection
✅ Actionable Recommendations - Clear next steps
✅ Priority Levels - Critical vs routine issues
✅ Historical Tracking - Complete alert logs
Clean Status:
✅ Solar IQ Status Update
Panel: Rooftop-01
Status: Clean - No maintenance required
Confidence: 92%
Time: 28/11/2025, 15:49:57
Next check: Automatic monitoring continues
Dust Detection:
🟡 Solar IQ Maintenance Alert
Panel: Rooftop-01
Issue: Dust Accumulation Detected
Confidence: 87%
Impact: Estimated 10-15% efficiency loss
Action: Schedule cleaning within 48 hours
Time: 15 Nov 2025, 11:00
Dashboard: Check for detailed analytics
Critical Issue:
🔴 Solar IQ CRITICAL Alert
Panel: Rooftop-01
Issue: Physical Damage Detected
Confidence: 93%
Impact: Panel integrity compromised
Action: IMMEDIATE inspection required
Time: 10 Nov 2025, 14:30
Contact: Call maintenance service now
Microcontroller: ESP32/ESP8266 (WiFi-enabled)
Sensors: DS18B20 (Temp), DHT11 (Temp+Humidity), Voltage Sensor
Camera: CP PLUS ezyLiv 3MP Wi-Fi PTZ
Panel: Loom Solar Polycrystalline Module
Power: 5V DC supply (with solar backup option)
Backend:
Runtime: Node.js v16+
Framework: Express.js
Database: Firebase Realtime Database
Authentication: Firebase Auth
APIs: Twilio SMS, Firebase Admin SDK
Language: JavaScript/TypeScript
Frontend:
Framework: React.js v18
State Management: Context API + Hooks
Styling: Tailwind CSS
Charts: Recharts
Real-time: Firebase SDK
Language: JavaScript/JSX
Machine Learning:
Framework: PyTorch 2.0+
Library: Hugging Face Transformers
Model: Vision Transformer (ViT-Base)
Training: Transfer Learning
Deployment: Flask API
Language: Python 3.8+
IoT/Embedded:
IDE: Arduino IDE
Language: C/C++
Libraries: WiFi.h, FirebaseESP32.h, OneWire, DallasTemperature, DHT
Protocol: MQTT (optional), HTTP/HTTPS
Version Control: Git, GitHub
IDE: VS Code, Arduino IDE, PyCharm
Testing: Jest (Frontend), Pytest (ML)
Deployment: Firebase Hosting, Heroku
Monitoring: Firebase Analytics
Documentation: Markdown, JSDoc
| Phase | Duration | Key Milestones |
|---|---|---|
| Planning & Research | Jan 2025 | ✅ Requirements, feasibility study, team formation |
| Hardware Development | Feb-Mar 2025 | ✅ ESP32 setup, sensor integration, circuit testing |
| Dataset Creation | Mar-Apr 2025 | ✅ Collected 500+ images, annotation completed |
| AI Model Training | May-Jul 2025 | ✅ ViT training, achieved 89-94% accuracy |
| Backend Development | Aug-Sep 2025 | ✅ Node.js API, Firebase integration, Twilio setup |
| Frontend Development | Sep-Oct 2025 | ✅ React dashboard, real-time features |
| Integration & Testing | Oct-Nov 2025 | ✅ End-to-end testing, bug fixes, optimization |
| 🏆 Final Showcase | Nov 28, 2025 | ✅ Successfully Presented & Demonstrated |
Total Project Duration: 11 months (January - November 2025)
✨ Demonstrated live working system
✨ Achieved 89-94% AI detection accuracy
✨ Real-time alerts delivered in <15 minutes
✨ Complete end-to-end IoT+AI integration
✨ Scalable architecture for future expansion
Five Engineering Students, One Vision: Smarter Solar Energy
|
Divyansh Sharma System Architecture & Integration Lead 🔧 End-to-end system design 🔗 Component integration 📊 Project management |
Karan Veer Singh IoT Hardware & ESP32 Development ⚡ Sensor integration 🔌 Circuit design 📡 WiFi communication |
Drishti Agarwal Machine Learning & AI Development 🤖 ViT model training 📈 Dataset creation 🎯 Accuracy optimization |
Aditya Sachdeva Frontend Development & UI/UX Design 💻 React dashboard 🎨 User interface 📱 Responsive design |
Abhilasha Tiwari Backend Development & Cloud Integration 🔙 Node.js API ☁️ Firebase setup 📧 Twilio integration |
|
Dr. Sandeep Mandia Technical Advisor & Project Guide Department of Electronics and Communication Engineering Thapar Institute of Engineering and Technology Patiala, India |
Dr. Amanpreet Kaur Project Mentor & Domain Expert Department of Electronics and Communication Engineering Thapar Institute of Engineering and Technology Patiala, India |
Thapar Institute of Engineering and Technology
Patiala, Punjab, India
Department of Electronics and Communication Engineering
| Traditional Challenge | Solar IQ Solution |
|---|---|
| 15-25% efficiency loss undetected | Real-time detection within 15 minutes |
| $50-100 per manual inspection | <$1/month automated monitoring |
| 2-4 weeks detection delay | Instant SMS alerts |
| Human error in inspection | 89-94% AI accuracy |
| Reactive maintenance only | Predictive scheduling enabled |
| No historical data tracking | Complete analytics dashboard |
✅ Real-time Detection - Issues identified within 15 minutes
✅ Cost Reduction - 80% cheaper than manual inspection
✅ Proactive Maintenance - Prevents 70% of efficiency loss
✅ Scalable Design - Works for 1-100+ panel installations
✅ High Accuracy - 89-94% detection confidence
✅ User-Friendly - SMS alerts + web dashboard
Our system is designed for:
🏠 Residential Installations - 1-10 panels, homeowner-friendly
🏢 Commercial Rooftops - 10-100 panels, business optimization
🏭 Solar Farms - 100+ panels, scalable zone monitoring
🎓 Educational Institutions - Campus solar systems
🌍 Remote Monitoring - Off-grid or difficult-to-access locations
Cost Breakdown:
Hardware Cost (One-time):
├── ESP32 Module: $8
├── Sensors (DS18B20+DHT11): $5
├── Voltage Sensor: $3
├── Camera (CP PLUS): $45
├── Miscellaneous: $9
└── Total Hardware: ~$70
Operational Cost (Monthly):
├── Firebase: Free tier
├── Twilio SMS: ~$0.50
├── Cloud hosting: ~$0
└── Total Monthly: ~$0.50
Traditional Manual Inspection:
└── Cost per visit: $50-100
└── Frequency: Monthly
└── Annual cost: $600-1200
Solar IQ Annual Cost: ~$76 (hardware amortized)
Savings: $524-1124 per year
ROI Period: <3 months
Phase 1: Enhanced Monitoring (Q1 2026)
- Multi-panel zone monitoring with GPS mapping
- Weather API integration for contextual analysis
- Advanced analytics with ML-powered predictions
- Mobile app for iOS and Android
Phase 2: Automation (Q2-Q3 2026)
- Automated cleaning system integration
- Solar inverter direct data connection
- Energy output correlation dashboard
- Predictive efficiency forecasting
Phase 3: Scale (Q4 2026+)
- Enterprise solar farm deployment
- Drone-based aerial inspection
- Multi-site management platform
- Commercial product launch
🔬 Ongoing Research Areas:
-
Advanced ML Models
- Transformer architectures (Swin, ConvNeXt)
- Real-time edge inference on ESP32-CAM
- Federated learning for privacy-preserving training
-
Enhanced Sensor Fusion
- Thermal imaging integration
- Vibration sensors for structural health
- Irradiance sensors for precise efficiency calculation
-
Predictive Analytics
- Time-series forecasting for maintenance scheduling
- Anomaly detection using LSTM networks
- Energy yield optimization algorithms
-
Scalability Solutions
- Distributed sensor networks
- Edge computing for large installations
- Cloud-based centralized management
| Metric | Value | Target | Status |
|---|---|---|---|
| Sensor Update Frequency | 30 seconds | 30 seconds | ✅ Met |
| Image Capture Interval | 15 minutes | 15 minutes | ✅ Met |
| AI Inference Time | 3-5 seconds | <10 seconds | ✅ Exceeded |
| Alert Delivery Time | 5-10 seconds | <30 seconds | ✅ Exceeded |
| Dashboard Latency | <1 second | <2 seconds | ✅ Exceeded |
| System Uptime | 99.2% | >99% | ✅ Met |
| Detection Accuracy | 89-94% | >85% | ✅ Exceeded |
| False Positive Rate | <5% | <10% | ✅ Exceeded |
Vision Transformer Accuracy by Category:
Normal (Clean): 92% ████████████████████░░
Dust: 87% ██████████████████░░░░
Shading: 85% ████████████████░░░░░░
Bird Droppings: 94% ████████████████████░
Physical Damage: 93% ███████████████████░░
Electrical Damage: 89% ██████████████████░░░
Overall Average: 90% ██████████████████░░░
This intensive 11-month capstone project provided comprehensive hands-on experience across multiple domains:
Hardware & IoT:
- ✅ ESP32/ESP8266 microcontroller programming
- ✅ Sensor integration and calibration
- ✅ Circuit design and troubleshooting
- ✅ WiFi communication protocols
- ✅ Real-time embedded systems
Machine Learning & AI:
- ✅ Deep learning model training (Vision Transformers)
- ✅ Transfer learning techniques
- ✅ Dataset creation and annotation
- ✅ Model optimization and deployment
- ✅ Computer vision applications
Full-Stack Development:
- ✅ Node.js backend development
- ✅ React.js frontend development
- ✅ Firebase real-time database
- ✅ RESTful API design
- ✅ Real-time data synchronization
System Integration:
- ✅ End-to-end system architecture
- ✅ Cloud services integration (Firebase, Twilio)
- ✅ IoT + AI pipeline design
- ✅ Data flow optimization
- ✅ System testing and validation
Project Management:
- 📊 11-month timeline management
- 📋 Task delegation and coordination
- 📈 Progress tracking and reporting
- 🎯 Milestone achievement
Team Collaboration:
- 🤝 5-member team coordination
- 💬 Regular communication and updates
- 🔄 Agile development methodology
- 🎭 Role-based responsibility management
Problem Solving:
- 🔍 Research and analysis
- 💡 Creative solution design
- 🐛 Debugging and troubleshooting
- ⚡ Real-time issue resolution
Communication:
- 📝 Technical documentation writing
- 🎤 Presentation and demonstration
- 🗣️ Stakeholder communication
- 📄 Report preparation
This project is licensed under the MIT License.
✅ Commercial Use - Use this project in commercial applications
✅ Modification - Modify and adapt the code for your needs
✅ Distribution - Share and distribute this project
✅ Private Use - Use privately for personal projects
✅ Sublicensing - Include in projects with different licenses
Conditions:
- Include original license and copyright notice
- State significant changes made to the code
Limitations:
- No warranty or liability provided
- Use at your own risk
See the LICENSE file for complete details.
Faculty Mentors:
- Dr. Sandeep Mandia - For invaluable technical guidance, mentorship, and continuous support throughout the project
- Dr. Amanpreet Kaur - For domain expertise, feedback, and encouragement at every stage
Institution:
- Thapar Institute of Engineering and Technology - For providing excellent facilities, resources, and the opportunity to pursue this innovative project
Open Source Community:
- PyTorch and Hugging Face teams for exceptional ML frameworks
- Firebase team for reliable cloud infrastructure
- ESP32 community for comprehensive documentation
- All open-source contributors whose libraries made this possible
Our Families:
- For unwavering support, patience, and encouragement during the intensive 11-month journey
- Conceived: January 2025
- Developed: January - November 2025 (11 months)
- Demonstrated: November 28, 2025
- Status: Successfully Completed ✅
Divyansh Sharma
System Architect & Integration Lead
