Back to Projects
Digital Twin

Digital Twin of Satellite

View Source Code

Project Overview

A high-fidelity physics and machine learning digital twin perfectly synchronized with live satellite telemetry, enabling simulation of thermal, power, and communication systems in real-time.

The Problem

Testing software updates or simulating harsh space weather on live orbital assets is incredibly dangerous and expensive. Ground operators lacked a safe, accurate sandbox to test maneuvers and predict systemic degradation over time.

Our Solution

We developed a real-time digital twin utilizing Neural Ordinary Differential Equations (Neural ODEs) to model physical systems and reinforcement learning to simulate state changes. The twin perfectly mirrors its physical counterpart, synchronized via low-latency telemetry ingestion.

System Architecture

Built on Unreal Engine 5 for high-fidelity visual rendering, paired with a massive Python simulation backend running on AWS. The data sync is handled via Apache Kafka, ensuring the twin's state is never more than 50ms behind the physical satellite.

Impact & Results

Enabled operators to test 4 critical software patches in a simulated environment before successful deployment. Identified a fatal thermal flaw during a simulated solar flare scenario, allowing engineers to upload a preventive cooling protocol, saving a $200M asset.

AI Models

Neural ODEsDeep Reinforcement LearningPhysics-Informed Neural Networks (PINNs)

Technology Stack

Unreal Engine 5PythonAWSApache KafkaReact