AI-Driven Fault Classification System for F1 Power Units: Design and Evaluation of Multiple Diagnostic Models

This thesis presents the design and implementation of an AI-based anomaly classification system for Formula 1 Power Units (PU), developed within the current regulatory context. The system leverages high-frequency telemetry data both for offline root cause analysis and for real- time detection and classification of critical failure modes. A multi-stage pipeline integrates expert-driven thresholding, robust feature engineering, and supervised learning models including decision trees, random forests, neural networks, and ensemble architectures. Special emphasis is placed on hybrid classifiers that improve the recognition of healthy states, reducing false positives. The system is evaluated through cross- validation and deployed both in an offline analysis dashboard and a real-time telemetry environment. The results demonstrate strong performance across a wide range of failure types, offering a scalable and interpretable data-driven tool for PU diagnostics. Designed with future compatibility in mind, the solution serves as a modular foundation for next-generation reliability monitoring systems under upcoming technical regulations.