Development of a virtual pressure transducer for engine control
For environmentally sound, efficient and robust large engines, it is essential to have a detailed knowledge of the combustion process. In-cylinder pressure transducers are often applied in advanced combustion control systems. This allows each cylinder to be operated efficiently while avoiding anomalies such as knocking or misfiring. Since high-quality cylinder pressure transducers come with a limited lifetime and at a relatively high cost, alternative sensing methods which are more durable and less expensive are being sought. While such sensing methods may provide significantly different information than pressure transducers, it is possible to use machine learning to reproduce the corresponding pressure traces.
| Target
The aim of the present thesis is to develop a model that determines the in-cylinder pressure trace based on a corresponding signal from an acceleration sensor. To this end, a measurement database has been created from engine tests under different operating conditions, where both an in-cylinder pressure transducer and a simple acceleration sensor were installed. These data will be modeled with data-driven methods from the field of machine learning to develop a virtual pressure transducer based on the acceleration sensor data. A special focus of the thesis is the implementation close to the engine control unit capabilities (e.g., constraints in data transmission or availability have to be considered during model development).
Tasks
- Familiarization with the experimental test setup (engine testbed, engine control unit, measurement technology and measurement parameters)
- Preprocessing of the measurement data from the experimental investigations
- Basic exploratory data analysis on signal characteristics and correlations with engine operating parameters
- Definition of a model training strategy considering different engine control unit constraints
- Development of a virtual pressure transducer for engine control
Your profile
Programming skills in Python or R; experience in data analysis
Starting date: As soon as possible
Duration: Approximately 6 months
Contact:
Univ.-Prof. Dr.-Ing. Nicole Wermuth, +43 (316) 873-30087, nicole.wermuth@lec.tugraz.at
Dipl.-Ing. Christian Laubichler, +43 (316) 873-30089, christian.laubichler@lec.tugraz.at |
 |
