Misc. - ROBOFIX

ROBOFIX — AI-Powered Predictive Maintenance & Wear Optimization for Industrial Robots

CLEAR INDICATION: Claude and Cursor used together to develop this project. AI has been used in this project. If demo link is to be used, contact Misc. to ensure backend is running as it is on a free tier.

Inspiration

Industrial robots operate under constant mechanical stress. Joint wear leads to unplanned downtime costing manufacturers thousands per hour. Current approaches rely on scheduled maintenance—replacing parts on a calendar, not based on actual condition. The question that inspired us was: can we predict which joint will fail next, and what material would make it last longer?

What it does

ROBOFIX is a full-stack AI system that takes any industrial robot's sensor data, detects which joints are degrading, predicts how quickly they'll wear out, and recommends the best replacement materials—all visualized on an interactive 3D robot model. It transforms reactive robot maintenance into a proactive, data-driven process, telling you which joint is degrading, how fast, and what material to use—before it fails.

How we built it

The system was built iteratively across 27 commits, using a modern tech stack:

The core pipeline is:

1. Dataset-Agnostic Ingestion: Auto-detects the schema of any uploaded CSV (accelerometers, gyroscopes, torque, etc.) using pattern matching and heuristics.
2. Modular Feature Engineering: Dynamically generates statistical, spectral, and vibration-specific features based on available sensors.
3. Anomaly Detection: A model registry supports four algorithms (Isolation Forest, LOF, One-Class SVM, Autoencoder) which users can compare.
4. Physics-Based Wear Estimation: Inspired by Archard's wear law, it calculates a normalized wear severity index (0-1) for each joint.
5. Material Recommendation: A database of 29 materials is ranked by a composite practicality score that balances wear reduction against density and cost.
6. 3D Visualization & Joint Mapping: Joints are color-coded on an interactive 3D model. Users can upload a robot photo for auto-detection or manual placement of joints.

Challenges we ran into

A major technical challenge was making the ML pipeline truly dataset-agnostic. We had to move from hardcoded assumptions to a system that could infer schema, dynamically generate relevant features, and adapt its models to any number of joints and sensor modalities. We overcame this through a complete ML pipeline refactor, implementing robust schema inference and a modular feature engine. Team coordination was managed through iterative development, with progress tracked across 27+ commits.

Accomplishments that we're proud of

We are incredibly proud of building a system with key differentiators:

• Dataset-agnostic: Works with any robotics sensor CSV.
• User-configurable pipeline: Allows feature deselection, downsampling, and model switching.
• Physics-grounded: Wear model is based on Archard's law, not just ML output.
• Cost-aware material ranking: A practicality score penalizes impractical materials.
• Full ML diagnostics: Offers threshold analysis, feature importance, and model comparison.
• Interactive joint mapping: Upload a photo to auto-detect joint positions.

UN Sustainable Development Goals (SDGs) Supported:

• SDG 9: Industry, Innovation, and Infrastructure: ROBOFIX promotes sustainable and resilient infrastructure through efficient resource use and predictive maintenance, reducing industrial downtime.
• SDG 12: Responsible Consumption and Production: By optimizing part replacement and suggesting longer-lasting materials, we directly contribute to reducing waste and promoting sustainable manufacturing practices.

What we learned

This hackathon was a massive learning experience. We gained deep insights into building scalable ML pipelines, mastering 3D visualizations with Three.js, and the importance of creating a truly flexible, user-centric architecture. We also learned how to ground machine learning outputs in real-world physics to create industrially viable solutions.

What's next

The future for ROBOFIX is bright. Our next steps include:

• Expanding the sensor fusion capabilities to integrate vision data for crack detection.
• Developing a time-series forecasting module to predict the exact Remaining Useful Life (RUL) of a joint.
• Integrating the system with real-time data streams from actual robot controllers for live monitoring.
• Building a more extensive material database with even more advanced composites and coatings.

Built With

• fastapi
• jspdf
• next.js
• numpy
• opencv
• pandas
• python
• react
• recharts
• scikit-learn
• scipy
• tailwindcss
• three.js
• typescript
• uvicorn