Artificial Neural Network for Engine Air Charge Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional spark ignition engines rely on empirical calibration methods for air charge estimation, which are inaccurate and require substantial calibration efforts, especially under changing external conditions, leading to suboptimal combustion quality, fuel economy, and engine knock.
Innovation Solution
A combustion control system utilizing a trained feedforward artificial neural network (ANN) to estimate air charge mass based on engine speed, intake manifold absolute pressure, intake and exhaust camshaft positions, intake air temperature, and engine coolant temperature, allowing for improved combustion stability, torque response, and fuel economy, with optional downstream adjustments using a 2D empirical surface and VE correction factors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If conventional empirical calibration methods are used for air charge estimation, then the system is easier to implement, but the measurement precision and reliability deteriorate under changing external conditions
Solution Approach 1:
The patent replaces conventional empirical calibration methods with a neural network-based system. The neural network learns complex nonlinear relationships between engine parameters and air charge mass through training data, substituting the traditional mechanical/empirical calibration approach with an intelligent computational model that adapts to varying operating conditions without requiring extensive recalibration
Solution Approach 2:
The patent transforms the calibration approach by changing from fixed empirical lookup tables to a dynamic neural network model. The neural network takes multiple engine parameters (intake manifold pressure, temperature, camshaft positions, etc.) as inputs and dynamically computes air charge mass based on learned relationships, allowing the system to adapt to changing external conditions through parameter variations rather than fixed calibration values
2Measurement precision
If multiple calibration tables and surfaces are utilized for air charge estimation, then the measurement precision may improve, but the device complexity and calibration effort increase substantially
Solution Approach 1:
The patent merges multiple separate calibration tables and surfaces into a single unified neural network model. Instead of maintaining separate empirical calibrations for different operating conditions, the neural network integrates all relationships into one coherent computational structure that handles the entire operating range through a single model, reducing system complexity while maintaining accuracy
Solution Approach 2:
The neural network serves as a universal calibration system that handles multiple functions simultaneously - it estimates air charge mass across all operating conditions, adapts to varying external factors, and replaces multiple specialized calibration tables with a single multi-functional model that works throughout the engine's operating range
3Use of energy by stationary object
If conventional empirical methods are used, then the processor throughput requirement is reduced, but the reliability and adaptability to changing external factors deteriorate
Solution Approach 1:
The patent substitutes empirical lookup table methods with a neural network computational model. While the neural network requires more processing power than simple table lookups, it provides superior reliability by learning complex nonlinear relationships and adapting to changing conditions, achieving a balance between computational cost and control reliability
4Manufacturing precision
If extensive calibration data collection is performed during calibration periods, then the manufacturing precision improves, but the loss of time and productivity decrease
Solution Approach 1:
The patent performs preliminary neural network training using comprehensive calibration data collected during a calibration period before production deployment. This preliminary action allows the neural network to learn optimal relationships offline, so that during actual production and operation, the system can quickly adapt without requiring extensive real-time calibration data collection, reducing time loss while maintaining high precision
Data Source
AI summary
A combustion control method and system for an engine of a vehicle comprises a controller configured to access a trained feedforward artificial neural network configured to model a volumetric efficiency (VE) of the engine based on measured engine speed, engine intake manifold absolute pressure, intake and exhaust camshaft positions, intake air temperature, and engine coolant temperature, generate a base VE of the engine using the trained feedforward artificial neural network and the measured parameters, estimate an air charge mass flowing to each cylinder of the engine based on the base VE of the engine, and control combustion in the cylinders of the engine based on the estimated air charge mass to improve at least one of combustion stability, torque response, and fuel economy.