Engine Model Calibration Using Bayesian Data Fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current model calibration methods, such as trial-and-error approaches, are time-consuming and computationally intensive due to the need for manual adjustment of tuning parameters to align model predictions with real-world data, especially when dealing with large temporal datasets containing noise and high-dimensionality, which complicates the calibration of complex models like engine performance models.
Innovation Solution
The use of data compression techniques combined with Bayesian data fusion and parallel Markov Chain Monte Carlo (MCMC) algorithms to reduce the dimensionality of data and efficiently identify potential values for tuning parameters, optimizing the calibration process by minimizing error between model and test data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual trial-and-error methods are used to adjust tuning parameters, then model predictions can be aligned with real-world data, but the calibration process becomes time-consuming and computationally intensive
Solution Approach 1:
The patent replaces manual trial-and-error adjustment with automated Bayesian inference algorithms. The system uses probabilistic models and computational methods to automatically identify optimal tuning parameter values, eliminating the need for human operators to manually iterate through parameter combinations. This substitution of automated computational mechanisms resolves the contradiction by maintaining prediction accuracy while dramatically reducing calibration time.
Solution Approach 2:
The patent transforms the calibration problem from adjusting individual parameters sequentially to simultaneously updating the probability distributions of multiple parameters using Bayesian inference. By changing the approach from deterministic parameter tuning to probabilistic parameter estimation, the system achieves both high accuracy and computational efficiency through parallel processing of parameter space exploration.
2Measurement precision
If manual trial-and-error methods are used to adjust tuning parameters, then model predictions can be aligned with real-world data, but computational resources are excessively consumed
Solution Approach 1:
The patent replaces computationally expensive manual trial-and-error simulations with efficient Bayesian inference algorithms. The system uses probabilistic forward models that compute likelihoods analytically or through simplified simulations, rather than requiring numerous full-scale model runs. This substitution reduces computational resource consumption while maintaining the ability to achieve accurate model predictions.
Solution Approach 2:
The patent applies partial action by using simplified probabilistic forward models that capture the essential system behavior without requiring complete physical fidelity. By using approximate likelihood computations and sampling methods rather than exhaustive simulations, the system achieves sufficient accuracy for calibration purposes while consuming far fewer computational resources than full trial-and-error approaches.
3Measurement precision
If large temporal datasets with noise and high-dimensionality are used for calibration, then model accuracy improves, but the complexity of identifying tuning parameters increases
Solution Approach 1:
The patent introduces probabilistic forward models as intermediaries between the complex noisy data and the tuning parameter estimation. These intermediate models serve as simplified representations that capture the essential input-output relationships while filtering out noise and irrelevant details. By using these intermediary probabilistic models to compute likelihoods, the system can effectively utilize large temporal datasets without being overwhelmed by their complexity and dimensionality.
4Measurement precision
If multiple output parameters are calibrated simultaneously, then comprehensive model alignment is achieved, but the calibration process becomes more complex and time-consuming
Solution Approach 1:
The patent merges the calibration of multiple output parameters into a unified Bayesian inference framework. Instead of calibrating parameters for each output parameter separately, the system simultaneously updates the joint probability distribution of all tuning parameters by combining likelihood information from multiple outputs. This merging approach achieves comprehensive model alignment across all outputs while reducing total calibration time through parallel processing and avoiding redundant computations.
Data Source
AI summary
Methods and systems for calibrating a model are provided herein. In some embodiments, the methods include receiving, via a control circuit, test data for an operational parameter of a real-world system, such as an engine, from operational tests. The control circuit also receives model data for the operational parameter from simulations performed via a model of the engine. The control circuit then compresses the test data and the model data to generate compressed test data and compressed model data and fusing the compressed test data with the compressed model data to generate fused data. The control circuit performs parallel Bayesian inference simulations using the fused data to identify at least one value for a tuning parameter of the model. The control circuit may identify and select tuning parameters to match the model data with test data, the model data may be one or more model outputs (i.e., output parameters).