Deep learning fault diagnosis and gradient-based root cause analysis on nuclear reactor sensor telemetry (96 parameters, 18 accident types) achieving 0.985+ AUROC on anomaly detection and 71.2% multi-class accident classification on the NPPAD dataset.

End-to-end MV grid fault risk scoring: feature engineering, tracked training, artefact versioning, API serving, monitoring-ready outputs

A research framework for benchmarking risk-aware time-series models in aerospace PHM. It focuses on de-noising complex flight manifolds, evaluating model stability under multi-modal regimes, and ensuring prognostic generalisation through rigorous experimental auditing.

[3] The most similar motor you've seen before - already failed.

Real-time predictive maintenance system on STM32 with Zephyr RTOS. Multithreaded sensor data collection, circular buffering, and on-device anomaly detection for industrial equipment monitoring.