⚡ Orbit – Transformer Management System

Orbit is a web-based system designed to digitize and streamline routine thermal inspections of distribution transformers. This project lays the foundation for an end-to-end inspection workflow where it integrates a detection pipeline with an interactive annotation and feedback loop, allowing users to adjust detections directly through the interface. Corrected annotations are logged and later used to fine-tune the model automatically, enabling continuous performance improvement and adaptability to real-world data.

---

⚙️ Phase 3 – Interactive Annotation & Feedback Loop

We have completed phase 3 focusing on enhancing user control, data accuracy, and model improvement.

🖊️ Interactive Annotation Tools

On the anomaly detection view, inspectors can now directly interact with the AI-detected anomalies by:

• Adjusting existing anomaly markers (resize, reposition)
• Deleting incorrect detections
• Adding new anomaly markers by drawing bounding boxes

Each annotation captures:

• Annotation type (added / edited / deleted)
• Optional comments or notes
• Timestamp and user ID

This feature empowers users to refine detections and ensure high-quality inspection records.

🗃️ Metadata and Annotation Persistence

All annotation changes are automatically captured and saved in the backend.
Each record includes:

• Annotation ID
• Detection ID
• Inspection number
• Annotations type
• User
• Comment
• Timestamp

When the same image is revisited, existing annotations are automatically reloaded, maintaining a consistent inspection history.

🔁 Feedback Integration for Model Improvement

The system now facilitates the automatic collection of high-quality training data, enabling continuous model retraining (Finetuning).

• Feedback Log Generation: A service endpoint allows for the export of a structured JSON log containing:

  • The Final Human-Validated Annotations
  • Annotator metadata

• Automated Retraining Readiness: This exportable log can be formatted for direct ingestion to Roboflow, setting up the automated finetuning of the model when enough new, corrected data is accumulated.

---

📌 Phase 2 – Automated Fault Detection & Inspection Comments

In phase 2 we introduced AI-powered analysis and feedback management, enhancing the system’s intelligence and usability.

🔍 Roboflow AI Integration

Integrated Roboflow model trained on transformer thermal imagery.

On uploading a maintenance image, the backend automatically:

• Sends the image to the Roboflow API for inference
• Receives detected fault bounding boxes, labels, and confidence scores
• Stores the prediction JSON in the database
• Annotated images are displayed in the inspection page

🧠 Fault Detection Workflow

Detected anomalies (e.g., hotspots, loose connections) are classified as:

• Normal
• Potential Fault
• Fault

The results are visually rendered on the uploaded image with color-coded bounding boxes.

💬 Inspection Comment System

Inspectors can add comments and observations on each inspection. Each comment includes:

• Author name
• Comment text
• Timestamp

Comments are stored in a dedicated table linked to the inspection. Admins can view and manage all past discussions on a transformer’s inspection history.

---

🗂️ Phase 1 – Transformer and Baseline Image Management

Phase 1 laid the foundation for managing transformers and baseline images.

Admin Interface for Transformer Management

• Add new transformer records
• View and edit existing transformer records
• Delete transformer records if required

Thermal Image Upload and Tagging

• Upload thermal images linked to specific transformers

• Support for two image types:

  • Baseline: Reference images for comparisons
  • Maintenance: Images from periodic inspections

• Each image is stored with metadata:

  • Upload date/time
  • Image type (Baseline / Maintenance)
  • Uploader (admin ID or name)

Categorization by Environmental Conditions

• While uploading baseline images, users must select environmental conditions:

  • Sunny
  • Cloudy
  • Rainy

• Images are stored and searchable by these conditions.

---

🧩 API Endpoints

Phase 1 – Transformer & Inspection Management

Method	Endpoint	Description
GET	/api/v1/transformers	Get all transformers
GET	/api/v1/transformers/{transformerNumber}	Get a transformer by transformer number
GET	/api/v1/transformers/{transformerNumber}/image	Get the baseline image of a transformer
POST	/api/v1/transformers	Create a new transformer
POST	/api/v1/transformers/{transformerNumber}/image	Upload baseline image for transformer
DELETE	/api/v1/transformers/{transformerNumber}	Delete a transformer by transformer number
GET	/api/v1/inspections	Get all inspections
GET	/api/v1/inspections/{inspectionNumber}	Get an inspection by inspection number
GET	/api/v1/transformers/{transformerNumber}/inspections	Get all inspections of a specific transformer
GET	/api/v1/inspections/{inspectionNumber}/image	Get inspection image
POST	/api/v1/inspections	Create a new inspection
POST	/api/v1/inspections/{inspectionNumber}/image	Upload inspection image
DELETE	/api/v1/inspections/{inspectionNumber}	Delete an inspection

Phase 2 – AI Integration & Comment System

Method	Endpoint	Description
GET	/api/v1/inspections/{inspectionNumber}/comment	Retrieve all comments for a specific inspection
POST	/api/v1/inspections/{inspectionNumber}/comment	Add a new comment to a specific inspection
GET	/api/v1/inspections/{inspectionNumber}/analyze	Retrieve AI model predictions for an inspection
PUT	/api/v1/inspections/{inspectionNumber}/analyze	Update AI model predictions for an inspection

Phase 3 – Interactive Annotation & Feedback APIs

Method	Endpoint	Description
GET	/api/v1/inspections/{inspectionNumber}/analyze/timeline	Retrieve timeline of all annotation edits for a given inspection
GET	/api/v1/inspections/analyze/timeline/all	Retrieve timeline of all annotation edits across inspections
GET	/api/v1/inspections/analyze/all	Retrieve all detections across inspections
POST	/api/v1/inspections/{inspectionNumber}/analyze	Add a new anomaly annotation
PUT	/api/v1/inspections/analyze/{detectId}	Update or modify existing anomaly annotation
DELETE	/api/v1/inspections/analyze/{detectId}	Delete a detection or annotation

---

🛠️ Tech Stack

• Frontend: React (Vite + TypeScript + Tailwind CSS)
• Backend: Java with Spring Boot (RESTful API)
• Database: Relational DB (PostgreSQL)

---

🚀 Setup Guide

Prerequisites

Make sure the following are installed:

• Java 21 (for Spring Boot backend) - Java Downloads
• Maven 3.9+ (build tool, usually bundled with IntelliJ / Spring Boot) - Download
• Node.js (v18.x or higher is recommended) - nodejs.org
• npm (Node Package Manager, comes with Node.js)
• PostgreSQL 17 (database, use pgAdmin for management) - PostgreSQL Downloads

---

Frontend Setup (React + Vite + Tailwind)

1. Go to frontend folder:

git clone https://github.com/Orbit-Transformer-Management/Transformer-Management
cd frontend

2. Install dependencies:

npm install

3. Run development server:

npm run dev

4. Open the app in browser: http://localhost:5173

Database Setup

1. Create PostgreSQL database:

CREATE DATABASE transformer_db;

2. Spring Boot will auto-generate tables (using JPA).
3. If needed, you can insert sample data manually via pgAdmin or SQL scripts.

---

🚧 Known Limitations / Future Work

• User authentication/authorization not yet implemented (anyone can access endpoints).
• No user profile or role management is available.
• The Roboflow AI model cannot be retrained dynamically using user inputs or feedback.
• Fault detection results rely solely on the pre-trained model without continuous learning capability.

---

👩‍💻 Contributors

• Team Orbit