Machine Learning Uncertainty Methods for Turbine Temperature Prediction
A new arXiv paper (2605.30585) benchmarks five machine learning uncertainty quantification methods for predicting turbine gas temperature degradation in modern engines. The methods—Delta method, Bayesian Monte Carlo Dropout, Bootstrap, Lower-Upper Bound Estimation, and Mean-Variance Estimation—are evaluated within a unified framework using cross-validation, repeated train-test splits, and multiple metrics including Coverage Probability, Normalized Mean Prediction Interval Width, and Coverage Width-based Criterion. The study aims to improve prognostics and health management by ensuring reliable and safe predictions.
Key facts
- Paper arXiv:2605.30585 benchmarks five uncertainty quantification methods for turbine gas temperature prediction.
- Methods: Delta method, Bayesian Monte Carlo Dropout, Bootstrap, Lower-Upper Bound Estimation, Mean-Variance Estimation.
- Unified framework uses cross-validation for hyperparameter selection.
- Repeated train-test splits assess performance robustness.
- Metrics: Coverage Probability, Normalized Mean Prediction Interval Width, Coverage Width-based Criterion.
- Goal: improve prognostics and health management of modern engines.
- Focus on accuracy and tightness of prediction intervals.
- Neural network predictions of turbine gas temperature are used.
Entities
Institutions
- arXiv