Model Ensemble Weighting for Faster Engine Control Calibration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for calibrating combustion engine control devices are time-consuming and expensive due to the need for extensive test bench testing, especially when dealing with a small number of data points and unknown model structures, as they rely on the Akaike Information Criterion which is inadequate for such conditions.
Innovation Solution
A method using empirical complexity measurements to determine weighting factors for a model ensemble, which evaluates the deviation of model outputs from actual physical process outputs over a complete input variable range, allowing for the formation of a surface information criterion that can derive weighting factors without knowledge of model structures, thereby reducing the number of necessary data points and test bench time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the Akaike Information Criterion is used to determine weighting factors for model ensembles, then model plausibility can be evaluated, but it requires a large number of data points and knowledge of model structures, which increases test bench time and cost
Solution Approach 1:
The patent changes the parameters used for model evaluation from traditional Akaike Information Criterion (requiring large N and known model structures) to an empirical complexity measurement based on output variable deviation over input variable ranges. This parameter change enables accurate model plausibility evaluation with fewer data points
Solution Approach 2:
The patent replaces the mechanical requirement of extensive test bench data collection with a computational approach using empirical complexity measurements. By substituting physical test bench iterations with computational evaluation of model outputs against process data, the method reduces test bench time while maintaining evaluation accuracy
2Measurement precision
If extensive test bench testing is conducted to collect sufficient data points for traditional model identification, then model accuracy can be improved, but the calibration process becomes time-consuming and expensive
Solution Approach 1:
The patent applies partial action by using only the necessary minimum of test bench data points rather than extensive testing. The empirical complexity measurement method achieves accurate model evaluation with fewer data points, performing only the essential measurements needed rather than excessive test bench time
Solution Approach 2:
The patent performs preliminary action by collecting process data during normal operation before model identification is needed. This preliminary data collection eliminates the need for extensive subsequent test bench testing, as the process data is already available from regular engine operation
3Productivity
If traditional model identification methods are used with small numbers of data points, then test bench time is reduced, but the resulting models lack sufficient accuracy for reliable calibration
Solution Approach 1:
The patent changes the evaluation parameters from traditional metrics that require large N to an empirical complexity measurement that works accurately with small N. The method evaluates model plausibility based on output variable deviation over the complete input variable range rather than relying on large sample sizes
Solution Approach 2:
The patent transitions from evaluating models in the data point dimension (requiring large N) to evaluating them in the input variable range dimension. By assessing model outputs across the complete input variable range using process data, the method achieves accurate evaluation with fewer data points through a dimensional shift in the evaluation approach
Data Source
AI summary
A method for generating a model ensemble that estimates at least one output variable of a physical process as a function of at least one input variable, the model ensemble being formed from a sum of model outputs from a plurality of models that have been weighted with a weighting factor.