Industrial Model Calibration Using Simulation Failure Screening
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing calibration methods for high-fidelity digital twin models of industrial systems often result in simulation failures due to complex parameter spaces and nonlinear interactions, leading to inefficient parameter estimation and significant resource wastage.
Innovation Solution
A failure-robust Bayesian optimization (FR-BO) algorithm that learns from simulation failures to identify regions in the parameter space where the system is likely to fail, informing a Bayesian optimization algorithm to avoid these regions and accelerate convergence, using a probabilistic parameter-to-cost regressor and acquisition function to select optimal parameters.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing calibration methods are used to estimate parameters in high-fidelity digital twin models, then parameter estimation can be performed, but simulation failures occur frequently due to complex parameter spaces and nonlinear interactions, leading to significant resource wastage
Solution Approach 1:
The patent applies preliminary action by performing a feasibility check before executing the full simulation. The system evaluates whether a candidate parameter set is likely to cause simulation failure using a trained classifier model. If the parameter set is predicted to fail, the simulation is skipped entirely, preventing resource wastage while maintaining the integrity of the calibration process
Solution Approach 2:
The patent introduces an intermediary component - a machine learning classifier model - that acts as a mediator between parameter selection and simulation execution. This classifier is trained on historical simulation data to predict failure-prone parameter combinations, thereby filtering out problematic parameters before they can cause simulation failures
2Measurement precision
If extensive simulations are performed to explore the parameter space, then near-optimal parameters can be found, but significant time and computational resources are wasted on simulations that are likely to fail
Solution Approach 1:
The system performs preliminary classification of candidate parameter sets before executing simulations. By evaluating the likelihood of simulation failure in advance using a trained model, the system avoids wasting time on parameter sets that would definitely fail, thereby reducing overall calibration time while maintaining estimation accuracy
Solution Approach 2:
The patent implements skipping by deliberately bypassing simulations for parameter sets predicted to fail. Instead of executing time-consuming simulations on all candidate parameters, the system skips those identified as failure-prone by the classifier, rushing through the calibration process by focusing only on promising parameter sets
3Adaptability or versatility
If the admissible parameter search domain is broadly defined to capture all possible parameter combinations, then comprehensive calibration can be performed, but the complexity of identifying failure regions increases significantly
Solution Approach 1:
The patent introduces a machine learning classifier as an intermediary that handles the complexity of failure region identification. Instead of requiring explicit mathematical characterization of complex failure regions in the parameter space, the classifier learns patterns from historical data and automatically identifies failure-prone regions, thereby managing the complexity while maintaining broad parameter search coverage
4Ease of operation
If candidate parameter sets are selected from user-defined search domains, then calibration can proceed with user-specified ranges, but parameters selected from these domains often result in simulation failure due to complex and non-intuitive shapes of the admissible parameter set
Solution Approach 1:
The system uses a trained classifier model as an intermediary layer between user-defined parameter ranges and simulation execution. The classifier analyzes candidate parameters within user-specified domains and predicts which ones are likely to fail, thereby protecting against simulation failures while maintaining the simplicity of user-defined search domains
Solution Approach 2:
Before executing simulations with user-selected parameters, the system performs a preliminary feasibility assessment using the trained classifier. This preliminary action filters out parameter sets that would likely fail, allowing users to define broad search domains while maintaining high simulation success rates through automated pre-screening
Data Source
AI summary
A calibration system and method for calibrating a model of dynamics of an industrial system is provided. The calibration method includes simulating the model multiple times with different combinations of parameters within an admissible range of values of the parameters to estimate success or failure of the simulation. Training the calibration system, iteratively, until a termination condition is met for defining a likelihood of failure of the simulation of the model and for a probabilistic parameter-to-cost mapping between various combinations of different values of the parameters of the model and their corresponding calibration errors. Further, calibrating, when the termination condition is met, the model with an optimal combination of the parameters having the largest likelihood of minimizing the calibration errors at the probabilistic parameter-to-cost mapping.