Inspiration

Formula 1 races are lost not just on the track, but in the garage when unexpected mechanical failures strike. After witnessing Max Verstappen's tire blowout in Baku 2021 and countless DNFs costing teams millions in championship points, we realized that traditional fixed-schedule maintenance was insufficient. We wanted to create an AI system that could predict component failures before they happen, giving teams precious time to prevent disasters and keep cars racing at peak performance.

What it does

PitPredict is an intelligent predictive maintenance system that monitors 9 critical F1 component sensors in real-time including vibration, temperature, pressure, and RPM. Using a trained Random Forest machine learning model with 95%+ accuracy, it calculates failure probability (0-100%) for specific components like engines, gearboxes, and brake systems. The system provides a three-tier alert system (Healthy/Warning/Critical) and features a premium dashboard where teams can select specific cars, drivers, and components to monitor, with support for analyzing real F1 race telemetry through FastF1 API integration.

How we built it

We built PitPredict using Python 3.8+ with scikit-learn's Random Forest classifier as our core ML engine, training on 10,000 synthetic sensor readings that simulate realistic component failure patterns. The frontend uses Streamlit for the interactive dashboard with Plotly for dynamic visualizations and real-time gauge charts. We integrated FastF1 API to pull actual race telemetry from 2018-2024 F1 seasons, converting it to our sensor format for real-world validation. Feature engineering was critical - we created 15+ derived features including vibration-temperature ratios and composite risk scores to improve prediction accuracy by 15% over raw sensor data alone.

Challenges we ran into

Our biggest challenge was creating realistic synthetic data without access to confidential F1 telemetry, requiring deep research into mechanical failure modes and sensor correlations. Feature engineering proved complex as raw sensor readings alone weren't predictive enough - we had to develop interaction features that captured relationships between sensors like vibration normalized by temperature. Integrating FastF1 API presented issues with data format inconsistencies (boolean vs numeric brake values) and long initial load times that we solved with caching and progress indicators. Balancing model sensitivity was tricky - too sensitive caused alert fatigue from false positives, too conservative missed critical failures, so we implemented configurable three-tier thresholds.

Accomplishments that we're proud of

We achieved 95%+ model accuracy with 0.96 ROC-AUC score, making predictions trustworthy for real-world deployment where false alarms are costly. Our premium dashboard design rivals professional F1 garage monitoring systems with team-specific color themes, real car images, and smooth animations. Successfully integrating FastF1 API allows analysis of actual race data from all 10 F1 teams across 24 Grand Prix races spanning multiple seasons. The system includes automated analysis history that saves and displays past predictions in beautiful expandable cards, and we built complete modularity with separate dashboards for simulated data, real F1 telemetry, and premium team-specific monitoring.

What we learned

We gained expertise in time-series sensor data analysis and learned that domain knowledge is as important as ML algorithms - understanding how vibration patterns indicate bearing wear was crucial for effective feature engineering. Working with FastF1 taught us real-world API integration challenges including handling missing data, caching strategies, and format inconsistencies. Most importantly, we learned that model accuracy alone doesn't determine success - visualization, clear communication of uncertainty, and actionable recommendations are equally critical for AI adoption. The project showed us that building trust through transparent explanations and intuitive interfaces is essential when deploying ML systems in high-stakes environments like professional racing.

What's next for PitPredict – AI-Powered Predictive Maintenance System

Next, we'll implement LSTM neural networks for time-series forecasting to predict failures hours or days in advance rather than just current risk. We're developing a mobile companion app for iOS and Android so pit crew and strategists can receive instant alerts on-the-go during race weekends. Long-term, we envision expanding beyond F1 into aerospace predictive maintenance for aircraft engines, manufacturing equipment monitoring in production lines, and medical device reliability in hospitals - any high-stakes environment where unexpected failures have serious consequences.