ANN Engine Torque Prediction for Complex Gasoline Calibration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing thermodynamics-based models for predicting vehicle engine torque are complex and inaccurate, especially in modern gasoline engines, leading to increased calibration costs and difficulties in measuring inert gas effects, resulting in reduced model accuracy.
Innovation Solution
A method using an artificial neural network (ANN) to predict engine torque by learning from operating point data sets, including spark timing and torque information, which reduces the need for physical law expressions and simplifies calibration processes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a thermodynamics-based model is used to calculate engine torque, then the model is based on physical laws and formulas, but the model complexity increases and accuracy decreases with recent technology applications
Solution Approach 1:
The patent replaces the thermodynamics-based mechanical model with an artificial neural network data model. The ANN learns torque prediction directly from operational data without relying on physical law expressions, thereby simplifying the model structure while maintaining or improving accuracy in complex modern engine systems
Solution Approach 2:
The patent changes the fundamental parameters of the prediction model from physics-based equations to data-driven relationships. By training the ANN on extensive operational data including inert gas effects, the model adapts to real-world complexities that traditional thermodynamic models cannot capture accurately
2Reliability
If the thermodynamics-based model includes multiple physical effects (inert gas, air-fuel ratios, ignition efficiencies), then comprehensive torque prediction is attempted, but calibration costs increase significantly
Solution Approach 1:
The patent merges multiple separate calibration processes into a single unified ANN training process. By training the neural network on comprehensive datasets that include all physical effects simultaneously, the model learns integrated relationships without requiring separate calibration for each effect, significantly reducing overall calibration costs
Solution Approach 2:
The ANN model performs self-calibration through automated training on operational data. The system learns optimal torque predictions by processing historical data without requiring extensive manual calibration intervention, reducing both time and cost投入 for model development
3Ease of operation
If inert gas amount is predicted by modeling instead of measured in real time, then measurement complexity is reduced, but prediction accuracy becomes significantly difficult to achieve
Solution Approach 1:
The patent introduces the ANN model as an intermediary that processes multiple operational parameters (intake air amount, fuel injection amount, EGR rate) to predict inert gas effects indirectly. Rather than measuring inert gas directly or relying on simple modeling, the ANN synthesizes information from multiple sources to achieve accurate predictions
4Reliability
If non-linearity of the model is expressed on a table due to effect of each physical quantity, then physical effects are accounted for, but the value set in the table has significant effect on model accuracy and increases complexity
Solution Approach 1:
The patent replaces the table-based representation of non-linear relationships with an ANN-based continuous function. The neural network learns smooth non-linear mappings from operational parameters to torque predictions, eliminating the discontinuities and sensitivities inherent in table-based approaches while maintaining physical effect representation
Data Source
AI summary
A method of predicting vehicle engine torque using an artificial neural network is provided. A data-based artificial neural network model is applied to more accurately calculate torque and reduce development costs for calibration and logics.