PropulsionX

Inspiration

UAVs are often monitored using simple dashboards that show sensor values and trigger alerts after a problem has already occurred. This reactive approach can lead to false alarms, delayed maintenance, mission failure, and unsafe operating conditions. We were inspired to build PropulsionX to make UAV monitoring more intelligent and proactive. Our goal was to create a system that does not only detect an abnormal condition, but also explains its possible cause, predicts future failure, evaluates mission risk, and recommends the best action to the operator.

What it does

PropulsionX is an AI- and physics-based UAV Digital Twin for predictive maintenance and mission decision support. It:

Collects real-time or simulated UAV engine telemetry.
Validates telemetry and detects packet loss, duplicate packets, replay attempts, and sensor problems.
Uses physics-based models to estimate expected engine behavior.
Uses AI models for anomaly detection, fault identification, and Remaining Useful Life prediction.
Cross-validates AI predictions with physics results to reduce false alarms.
Calculates engine health and mission risk.
Simulates “What-If” scenarios such as changing RPM, altitude, or cooling conditions.
Uses optimization to identify safer operating conditions.
Generates operational recommendations and ATA-style maintenance work orders.
Monitors multiple UAVs through fleet management.
Provides explanations through an AI Mission Engineer.
Displays all results on a web-based Ground Control Dashboard.

How we built it

We built PropulsionX as a modular system with three main layers:

1. Data and Communication Layer Telemetry is generated by simulators or received through communication interfaces such as MQTT and CAN/J1939. The data is timestamped, validated, normalized, and sent to the backend.

2. Digital Twin and Intelligence Layer The Python backend processes telemetry through multiple analytical modules:

   Physics-based engine calculations Isolation Forest anomaly detection LSTM-based Remaining Useful Life prediction Sensor integrity monitoring Telemetry integrity monitoring AI–Physics cross-validation Engine health scoring Mission-risk calculation What-If simulation L-BFGS-B operating-point optimization Prescriptive maintenance Fleet management AI Mission Engineer responses

3. Visualization and Decision Layer The results are sent to a Next.js, React, and TypeScript dashboard. Operators can view telemetry, engine health, RUL, mission risk, What-If results, maintenance recommendations, fleet status, and AI explanations. We also created automated tests and fault-injection scenarios to verify the behavior of the system.

Challenges we ran into

Combining physics-based calculations with machine-learning predictions. Preventing false alarms from a single AI model. Designing a clear decision process for AI and physics disagreements. Simulating realistic engine degradation and sensor faults. Handling telemetry errors such as missing, duplicate, delayed, or abnormal packets. Converting technical predictions into simple operator recommendations. Designing a dashboard that displays complex analysis in an understandable way. Ensuring the optimization process respects safe operating limits. Coordinating multiple backend modules into one real-time inference pipeline. Explaining advanced concepts such as RUL, mission risk, and digital twins to non-technical users.

Accomplishments that we're proud of

Developed a complete UAV Digital Twin prototype instead of only an isolated AI model. Combined AI predictions with physics-based validation. Implemented the four-case AI–Physics cross-validation matrix. Added What-If simulation for testing alternative mission conditions. Added constrained operating-point optimization using SciPy L-BFGS-B. Converted analytics into practical maintenance recommendations. Added fleet-level UAV monitoring. Created an AI Mission Engineer for grounded operational explanations. Built a modern Ground Control Dashboard. Added telemetry and sensor-integrity monitoring. Created automated tests for important system features. Designed the system in a modular way so that new sensors, aircraft, and models can be added later.

What we learned

AI predictions become more reliable when they are checked against domain physics. A prediction is more useful when it is connected to a clear action. Data quality and sensor integrity are as important as the machine-learning model. Digital Twins require continuous synchronization between the real system and the virtual model. Mission risk is different from component health; a small fault can become serious depending on the mission conditions. Explainability is essential for operator trust, especially in aerospace applications. Simulation is valuable for testing dangerous or expensive scenarios safely. Good system design requires communication between data pipelines, analytics, backend services, and user interfaces. Beginner-friendly visualization is important because operators should not need to understand every algorithm to use the system. Automated testing and fault injection help validate whether the system behaves correctly under abnormal conditions.

What's next for PropulsionX

Our next goal is to move PropulsionX from a prototype toward a real-world UAV decision-support platform. Future improvements include:

Connecting the system to real UAV sensors and flight controllers.
Supporting additional propulsion systems, including electric and hybrid propulsion.
Deploying lightweight models directly on edge devices for low-latency decisions.
Improving model accuracy using larger real-world flight datasets.
Adding online learning and model updating as new mission data becomes available.
Expanding federated learning for privacy-preserving fleet intelligence.
Adding stronger cybersecurity and secure telemetry authentication.
Integrating digital maintenance records and spare-parts planning.
Supporting multi-UAV mission coordination.
Adding 3D visualization of the aircraft and propulsion system.
Conducting hardware-in-the-loop testing.
Validating the platform against aerospace safety and certification requirements.
Developing mobile and offline dashboard support for field operators.
Creating a complete pilot deployment with research institutions, UAV manufacturers, and defence or industrial partners.

PropulsionX aims to become an explainable, predictive, and prescriptive intelligence layer for safer and more reliable UAV operations.

Built With

• c
• can
• css3
• html5
• json
• linux
• mqtt
• next.js
• python
• react
• scikit-learn
• scipy
• sqlite
• tensorflow
• typescript
• websockets