Digital Twin Surrogate Optimization With Constraint-Checked Parallel Search
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current optimization techniques for industrial processes using digital twins are computationally expensive and inefficient, often requiring multiple executions of expensive physics-based models, and surrogate models may not capture underlying physical processes or constraints, leading to suboptimal solutions.
Innovation Solution
A fast surrogate-based optimization method that uses a physics-based digital model in conjunction with a machine learning surrogate model to determine optimal input parameters, verifying constraints with the digital model to ensure feasibility, and employing parallel processing to enhance efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple executions of physics-based digital model are performed to converge to optimal value, then optimization accuracy is improved, but computational time and resources increase significantly
Solution Approach 1:
The patent creates a surrogate model that copies the input-output behavior of the expensive physics-based digital twin model. This surrogate model can be executed rapidly multiple times during optimization iterations, replacing the need to repeatedly run the computationally intensive original model while maintaining sufficient accuracy for finding optimal parameters.
Solution Approach 2:
The surrogate model acts as an intermediary between the optimization algorithm and the physics-based digital twin. The optimization process interacts with the fast surrogate model, and only occasionally validates results with the full digital twin, thereby mediating the computational burden while preserving optimization accuracy.
2Productivity
If surrogate model is used for optimization, then computational speed is improved, but fidelity to underlying physics and constraint satisfaction deteriorate
Solution Approach 1:
The patent implements a feedback mechanism where the surrogate model's predictions are periodically validated against the physics-based digital twin. When constraints are violated or accuracy thresholds are not met, the system adjusts the optimization approach or re-evaluates using the full model, ensuring that the fast surrogate model does not compromise solution reliability.
Solution Approach 2:
The optimization process uses the surrogate model for the majority of iterations (partial action) to achieve rapid convergence, then applies the full physics-based model for final validation and constraint verification (excessive action). This hybrid approach leverages the speed of the surrogate while ensuring constraint satisfaction through occasional full model evaluations.
3Measurement precision
If full exploration of parameter space is performed using physics-based model, then solution optimality is improved, but computational resources required become impractical
Solution Approach 1:
The surrogate model serves as a computationally efficient copy that enables comprehensive parameter space exploration. By replacing the expensive physics-based model with this faster surrogate, the system can evaluate numerous parameter combinations and perform thorough optimization searches without exhausting computational resources.
Solution Approach 2:
The patent performs preliminary optimization using the surrogate model to identify promising regions in the parameter space before conducting final validation with the physics-based model. This preliminary action narrows down the search space, reducing the number of full model evaluations needed while maintaining solution optimality.
Data Source
AI summary
A method of optimizing setting in an industrial process is described, along with associate media and systems, using a digital twin, computationally expensive physics-based model coupled with a faster machine learning (ML) model. Multiple initial guesses are subject to multiple branches of constrained objective function optimization using the ML model. In each branch, the optimized parameters are checked using the physics-based model, subject to constrain correction projection if they violate a constraint until they are in-bounds, and then logged. The best parameters are then selected from the logged parameters from all of the branches.