ANN Combustion Phasing Control for Off-Nominal Cam Timing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional combustion phasing control techniques for spark ignition (SI) engines rely on empirical approaches with substantial calibration efforts and are inaccurate at off-nominal camshaft positions, especially when variable valve control systems are involved, leading to decreased fuel economy and engine knock.
Innovation Solution
A calibration system using an artificial neural network (ANN) is implemented, which receives dynamometer data, weights it for high engine loads, generates training data, filters trained ANNs based on error metrics, and selects the best ANN for determining optimal spark timings, eliminating the need for empirical calibration surfaces.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional empirical calibration techniques with multiple calibration tables are used, then combustion phasing control can be implemented, but the calibration effort and processor throughput requirements increase substantially
Solution Approach 1:
The patent replaces the conventional empirical calibration approach with multiple calibration tables and surfaces with an artificial neural network (ANN) model. The ANN learns optimal spark timing relationships from training data during an offline calibration phase, then provides accurate spark timing predictions during engine operation without requiring complex real-time calculations or multiple calibration lookup tables. This substitution of the calibration methodology reduces both the calibration complexity and processor requirements while maintaining control accuracy.
2Reliability
If empirical calibration surfaces are used for variable valve control systems, then combustion phasing control is achieved, but accuracy decreases at off-nominal camshaft positions due to superposition methodology
Solution Approach 1:
The patent employs a dynamic ANN model that adapts to varying camshaft positions rather than using static empirical calibration surfaces. The ANN is trained on comprehensive data covering the full range of camshaft positions and operating conditions, enabling it to accurately predict spark timing at both nominal and off-nominal positions. This dynamic approach eliminates the superposition methodology limitations and maintains accuracy across the entire operating envelope.
3Reliability
If multiple calibration tables and surfaces are utilized, then combustion phasing control is implemented, but processor throughput requirements increase
Solution Approach 1:
The patent performs the complex calibration work in advance during an offline training phase, where the ANN model learns from extensive training data. The trained model is then deployed to the engine control unit, where it requires minimal processor throughput during real-time operation. This preliminary action separates the computationally intensive calibration process from the real-time control execution, significantly reducing in-vehicle processor requirements while maintaining control accuracy.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A calibration system and method for a spark ignition engine of a vehicle involve artificially weighting engine dynamometer data in high engine load regions and using it to generate training data for an artificial neutral network (ANN). A plurality of ANNs are trained using the training data and the plurality of ANNs are then filtered based on their maximum error to obtain a filtered set of trained ANNs. A statistical analysis is performed on each of the filtered set of trained ANNs including determining a set of statistical metrics for each of the filtered set of trained ANNs and then one of the filtered set of trained ANNs having a best combination of error at high engine loads and the set of statistical error metrics is then selected. Finally, an ANN calibration is generated using the selected one of the filtered set of trained ANNs