Physics-guided analytical model validation

Inventors: Hany S. Abdel-Khalik (Oak Ridge, TN); Ugur Mertyurek (Oak Ridge, TN)

Assignee: UT-Battelle, LLC

G06N7/04

View Patent ↗

Loading inventors, assignments & file history…

Monitor This Case

Get email alerts when status or documents change.

Order Certified Copies

Most orders are placed with the USPTO same day — all within 24 business hours.

Order via The Patent Place →

Pre-filled with this patent's details

Quick Facts

Patent No.: US 12,367,408
App. No.: 17/470,171
Granted: Jul 22, 2025
Kind: B2

External Links

USPTO Patent Center↗ Full Patent PDF↗

Abstract

This invention relates to a parameter or response assist filter that ensures that the predictions of a post-validation calibrated physics system simulator will remain within boundaries of a predetermined model validation domain. Embodiments utilize one or more filters to ensure calibrated model parameters {acute over (P)} and/or calibrated responses {tilde over (ϕ)} cause physics simulator model predictions to remain within the boundaries of the model validation domain MVD for a target application. The filters can be constructed prior to use or automatically inferred, or otherwise determined, from available measurements and other renditions of the physics system simulator during operation.

Claims (52)

1. A method for post-validation adjustment of a physics system simulator configured to simulate predicted behavior and/or state of a physical system based on an application model (M A ) and multiple physical or adhoc parameters (P), denoted as model parameters, and their corresponding known parameter variations (ΔP), wherein the application model (M A ) is related to one or more scaled-down experimental models (M E 1 ,M E 2 , . . . ), each scaled-down experimental model (M E j ) being associated with a respective set of experimental measurements (φ E j ), where j=1, 2, . . . , wherein the physics system simulator is validated for a target application model (M A ), as described by a model validation domain (MVD) the boundaries of which are evaluated mathematically based on deterministic or stochastic multi-variate functions of the target application model's responses, a set of scaled-down experimental-models' responses, the corresponding sets of experimental measurements (φ E 1 ,φ E 2 , . . . ), derivatives thereof, and the model parameters (P), the corresponding parameter variations (ΔP), and an uncertainty estimator, the method comprising:

predicting, by a first implementation ( 810 ) of the physics system simulator, first experimental responses (Φ E 1 , Φ E 2 . . . ) of the physical system by modeling the physical system using the scaled-down experimental models (M E 1 , M E 2 , . . . ) based on the model parameters (P) and their corresponding parameter variations (ΔP);

filtering, by a validation assist parameter filter ( 820 ) having an MVD boundary filter operator, the parameter variations (ΔP) corresponding to variations in responses for experimental models (ΔΦ E ) that cause the predictions of a second physics system simulator, denoted by post-validation calibrated (PVC) physics system simulator, for the target application responses to fall outside of the MVD of the first physics system simulator;

updating, by the first implementation ( 810 ) of the physics system simulator, the first scaled-down experimental responses (Φ E 1 , Φ E 2 , . . . ) of the physical system by modeling the physical system using the scaled-down experimental models (M E 1 , M E 2 , . . . ) based on the physical parameters (P) and their corresponding filtered parameter variations (fΔP);

adjusting, with a parameter calibration module ( 830 ), the physical parameters (P) based on the updated first scaled-down experimental responses (Φ E 1 *, Φ E 2 *, . . . ), the corresponding sets of experimental measurements ((φ E 1 ,φ E 2 , . . . ), and the filtered parameter variations (fΔP); and

predicting, by the first implementation ( 810 ) of the physics system simulator, a posteriori application response ({tilde over (Φ)} A ) of the physical system by modeling the physical system using the application model (M A ) based on the adjusted physical parameters ({tilde over (P)}).

2. The method of claim 1 , wherein the filtering comprises

predicting, by a second implementation ( 821 ) of the physics system simulator different from the first implementation ( 810 ) of the physics system simulator, second scaled-down experimental responses (Φ E′ 1 *, Φ E′ 2 *, . . . ) of the physical system by modeling the physical system using the same or second scaled-down experimental models (M (E′ 1 ), M(E′ 2 ), . . . ) based on the same model parameters (P) and their corresponding parameter variations (ΔP);

selecting, by a parameter-feature selector ( 823 ), parameter features comprising mathematical expressions derived from the multi-variate functions used to describe the boundaries of the model validation domain (MVD);

determining, by a validator ( 825 ) of the filter module, whether the first scaled-down experimental responses (Φ E 1 , Φ E 2 , . . . ) and the second scale-down experimental responses, (Φ E′ 1 *, Φ E′ 2 *, . . . ), corresponding to the selected parameter features are within the boundaries of the model validation domain (MVD); and

in response to the parameter features falling outside the boundaries of the MVD, removing, by a remover ( 827 ), the parameter features for which the first scaled-down experimental responses (Φ E 1 , Φ E 2 , . . . ) and the second scaled-down experimental responses (Φ E′ 1 *, Φ E′ 2 *, . . . ) are outside the boundaries of the model validation domain (MVD).

3. The method of claim 2 , wherein

the first implementation ( 810 ) of the physics system simulator is a high-fidelity implementation of the physics system simulator, and

the second implementation ( 821 ) of the physics system simulator is a low-fidelity implementation of the physics system simulator.

4. The method of claim 2 , wherein selecting the parameter features is performed using one or more of singular value decomposition, project pursuit techniques, or neural networks.

5. The method of claim 1 , wherein the filtering is based upon an increase in mutual information beyond a threshold determined by comparison of scaled-down experimental responses (Φ E1 , Φ E2 , . . . ), (Φ E 1 , Φ E 2 , . . . ) from two separate physics system simulator instances.

6. A system comprising: