Concepts de base
Our research contributes to achieving real-time probabilistic boundary condition estimation in heat transfer problems using an Ensemble-based Simultaneous Input and State Filtering approach with Radial Basis Functions. The method efficiently handles noisy measurements and errors, crucial for continuous casting machinery operation.
Résumé
The study focuses on solving the inverse heat transfer problem to estimate the transient heat flux between a mold and molten steel. By incorporating Bayesian methods with RBFs, the research provides insights into efficient real-time monitoring and control of critical industrial processes like continuous casting machinery. The methodology addresses key hyperparameters not documented in existing literature, enhancing accuracy and computational efficiency.
The content discusses the application of EnSISF-wDF with RBFs to predict temperature distribution and estimate unknown boundary conditions in heat transfer problems. It explores sensitivity analysis of hyperparameters such as ensemble size, shape parameter, prior weight shifting, time step, observation span, and covariance matrix scaling factor. Results show that Multiquadric kernels outperform Gaussian kernels in accuracy and computational cost efficiency.
Key points:
- Formulation of stochastic inverse heat transfer problem using Ensemble-based Simultaneous Input and State Filtering.
- Incorporation of Radial Basis Functions to reduce unknown inputs and computational burden.
- Importance of accurate real-time prediction for smooth operation of Continuous Casting machinery.
- Investigation of hyperparameters' impact on estimation accuracy in Bayesian framework.
- Comparison between Gaussian and Multiquadric kernels for HF estimation efficiency.
Stats
A spatiotemporal relative error metric is utilized to distinguish the impact of hyperparameter variations on the accuracy of the proposed method.
Optimal parameters include ensemble size (Sn), shape parameter (η), prior weight scaling factor (κ), prior weight shifting, time step (∆t), observation span, and resulting error rates.
Citations
"Our research contributes to achieving real-time probabilistic boundary condition estimation in heat transfer problems."
"The Multiquadric kernel outperforms Gaussian kernels in accuracy and computational cost efficiency."