Development of a Fast-Running Injector Model with Artificial Neural Network (ANN) for the Prediction of Injection Rate with Multiple Injections
- Delivery
- Provide download link
- Format
- Price
- Non-members (tax incl.)：￥6,600 Members (tax incl.)：￥5,280
- Paper/Info type
- SAE Paper
No.2021-24-0027
- Pages
- 1-12(Total 12 p)
- Date of publication
- Sep 2021
- Publisher
- SAE International
- Language
- English
- Event
- International Conference on Engines and Vehicles 2021
Detailed Information
| Author(E) | 1) Dominik Golc, 2) Stefania Esposito, 3) Heinz Pitsch, 4) Joachim Beeckmann |
|---|---|
| Affiliation(E) | 1) RWTH Aachen University, 2) RWTH Aachen University, 3) RWTH Aachen University, 4) RWTH Aachen University |
| Abstract(E) | The most challenging part of the engine combustion development is the reduction of pollutants (e.g. CO, THC, NOx, soot, etc.) and CO2 emissions. In order to achieve this goal, new combustion techniques are required, which enable a clean and efficient combustion. For compression ignition engines, combustion rate shaping, which manipulates the injected fuel mass to control the in-cylinder pressure trace and the combustion rate itself, turned out to be a promising opportunity. One possibility to enable this technology is the usage of specially developed rate shaping injectors, which can control the injection rate continuously. A feasible solution with series injectors is the usage of multiple injections to control the injection rate and, therefore, the combustion rate. For the control of the combustion profile, a detailed injector model is required for predicting the amount of injected fuel. Simplified 0D models can easily predict single injection rates with low deviation. However, the prediction of injection rates with multiple injections is more challenging, because of the impact of past injections on future ones. In this work, an advanced 0D injector model is presented, which takes into account the effect of injection history. In order to develop and calibrate the model, an injection rate testbench has been used to generate an extensive and suitable database. This database is used to train an artificial neural network (ANN), which is integrated in the model. The developed multi-injection model predicts with high accuracy (R2>0.85) the experimental injection rate up to four injections. Additionally, the model is real-time capable and therefore usable for controller application. |
