AI for nuclear safety: Predicting component remaining useful life
summary
Predicts remaining useful life of the sodium coolant pump in fast breeder reactors using multivariate time series from eight sensors (temp, pressure, flow, level, current, voltage, speed, vibration) with redundant and diversified sensing. Combines virtual twin data + real pump calibration to generate supervised examples, applies resampling, anomaly scoring, exponential smoothing and windowed statistics, then trains three tree‑based models fused by regression and a 2‑out‑of‑3 deployment voting scheme on redundant hardware. Reported model performance reaches about 93% accuracy and supports automated alerts, isolation and maintenance scheduling; LLMs for RCA remain exploratory.
AI-written summary · about
chapters (13)
- 0:00Introduction and speaker intro
- 1:14Industry 4.0 and predictive maintenance
- 3:55Nuclear reactor safety challenges
- 6:37IGCAR and fast breeder reactors
- 8:40Coolant pump and failure modes
- 11:48Data generation and virtual modeling
- 13:25Data insights and preprocessing
- 14:56Feature engineering and model architecture
- 16:34Decision engine and deployment
- 18:32Future work and conclusion
- 19:23Q&A on AI in aerospace vs nuclear
- 20:10AI model constraints across domains
- 22:03LLM use in reactor monitoring
related talks
21mApplication of AI technologies in nuclear power plants
20mAI Driven Vectorized Motor Health Monitoring
18mPredictive Network: How AI Forecasts System Failures
40mPredictive Maintenance
30mAI for Waste Water Treatment
14mTransforming Manufacturing: AI-Driven Predictive Maintenance
15mFrom Alarms to Action — AI for Conveyor Reliability
31mFinding the Needle in the Haystack: Predicting Storage Device Failures in...