Data-Driven Modeling: An AI Toolchain for the Powertrain Development Process
- Delivery
- Provide download link
- Format
- Price
- Non-members (tax incl.)：￥6,600 Members (tax incl.)：￥5,280
- Paper/Info type
- SAE Paper
No.2022-01-0158
- Pages
- 1-25(Total 25 p)
- Date of publication
- Mar 2022
- Publisher
- SAE International
- Language
- English
- Event
- WCX SAE World Congress Experience 2022
Detailed Information
| Author(E) | 1) Sasa Milojevi, 2) Sebasian Bodza, 3) Valerian Cimniak, 4) Michael Angerbauer, 5) Dominik Rether, 6) Michael Grill, 7) Michael Bargende |
|---|---|
| Affiliation(E) | 1) University of Stuttgart, 2) University of Stuttgart, 3) University of Stuttgart, 4) University of Stuttgart, 5) FKFS, 6) FKFS, 7) University of Stuttgart |
| Abstract(E) | Predictive physical modeling is an established method used in the development process for automotive components and systems. While accurate predictions can be issued after tuning model parameters, long computation times are expected depending on the complexity of the model. As requirements for components and systems continuously increase, new optimization approaches are constantly being applied to solve multidimensional objectives and resulting conflicts optimally. Some of those approaches are deemed not feasible, as the computational times for required single predictions using conventional simulation models are too high. To address this issue it is proposed to use data-driven model such as neural networks. Previous efforts have failed due to sparse data sets and resulting poor predictive ability. This paper introduces an AI Toolchain used for data-driven modeling of combustion engine components. Two methods for generating scalable and fully variable datasets will be shown. Prior to the actual training of neural networks, resulting data is evaluated regarding the distribution of variation parameters in a multidimensional space and the distribution density of prediction variables. Second, datasets will be post-processed according to evaluation findings. Eventually, layout and training of neural networks predicting combustion characteristics, e.g. MFB (Mass Fraction Burned) and pressure quantities, are described. A final summary will conclude the statistical accuracy of corresponding neural networks. Finally, the potential use of neural networks in the development process of internal combustion engines is discussed. The ultimate goal of the research project is the holistic realization of a data-driven engine model. This work represents the current status concerning this effort. |
