Deep Learning Predicts Tritium Source Stability for KATRIN Neutrino Experiment
The Karlsruhe Tritium Neutrino Experiment (KATRIN) has utilized advanced deep learning time-series forecasting techniques to anticipate stability following disturbances in its windowless gaseous tritium source. KATRIN's goal is to achieve an unprecedented level of sensitivity in measuring the absolute mass of neutrinos, necessitating meticulous observation of the tritium beta decay source. Although beta-induced X-ray spectroscopy allows for real-time diagnostics, traditional methods for detecting drift struggle with rare transient instability occurrences. This study integrates cutting-edge forecasting models, such as LSTM, N-BEATS, TFT, NHITS, DLinear, NLinear, TSMixer, and Chronos-LLM, with actual experimental data, surpassing conventional benchmarking on static datasets. The results indicate enhanced prediction capabilities for the time required to reach stability, thereby improving monitoring of the neutrino mass measurement source.
Key facts
- KATRIN aims to measure absolute neutrino mass with unprecedented sensitivity.
- The experiment uses a windowless gaseous tritium source for beta decay.
- Beta-induced X-ray spectroscopy provides real-time source diagnostics.
- Traditional drift detection struggles with infrequent transient instability events.
- The study applies deep learning forecasting models to predict time to stability.
- Models used include LSTM, N-BEATS, TFT, NHITS, DLinear, NLinear, TSMixer, and Chronos-LLM.
- The approach bridges forecasting models with real-world experimental data.
- The work is published on arXiv with ID 2605.08140.
Entities
Institutions
- Karlsruhe Tritium Neutrino Experiment (KATRIN)
- arXiv
Locations
- Karlsruhe
- Germany