Parameter Identification for Thermal Reduced-Order Models in Electric Engines
One part of the validation process of electric engines must check for thermal aging and damage of their components due to the high temperatures to which they are exposed. This way, the thermal requirements of the machine can be defined, and specific minimum service life can be guaranteed. For this p...
| Autor: | |
|---|---|
| Tipo de recurso: | tesis de maestría |
| Fecha de publicación: | 2021 |
| País: | España |
| Institución: | Universitat Politècnica de Catalunya (UPC) |
| Repositorio: | UPCommons. Portal del coneixement obert de la UPC |
| Idioma: | inglés |
| OAI Identifier: | oai:upcommons.upc.edu:2117/359319 |
| Acceso en línea: | https://hdl.handle.net/2117/359319 |
| Access Level: | acceso abierto |
| Palabra clave: | Electric motors Motors elèctrics Àrees temàtiques de la UPC::Enginyeria elèctrica |
| Sumario: | One part of the validation process of electric engines must check for thermal aging and damage of their components due to the high temperatures to which they are exposed. This way, the thermal requirements of the machine can be defined, and specific minimum service life can be guaranteed. For this purpose, drives must be validated against the most critical cases identified through simulations of representative driving scenarios. Since the thermal models require a long computation time to determine the temperatures of the components in each time increment, reduced-order models (ROM s) that can estimate them quickly are preferred instead. Also, there are positions in the engine where the temperature in the thermal model is determined only by sensor data when performing calculations online, like with the coolant temperature. Since it is also relevant to compute these values when performing simulations offline, ROM s can be applied for this purpose as well. This project focuses on creating and comparing different types of ROM s for tempera- ture estimation in electric engines. Several variants of discrete-time state-space models (SSM s) have been developed in the literature, showing promising results but requiring a high level of expert knowledge. This work introduces a set of SSM s that is entirely data- based and does not require knowledge of the physics and dynamics of the motor. They allow the user to adjust the parameters in different engines and create customized variants. Three models were developed for each engine temperature to be estimated. A preprocess- ing of the driving data divides it into three possible domains, and each model estimates the temperatures in their respective one. Model discretization based on different scenarios has shown an improvement in estimation accuracy. Finally, black-box approaches based on artificial neural networks (ANN s) were designed since they showed high potential in literature. Regression and Long Short-Term Memory (LSTM ) models were created, and their hyperparameter s were optimized, but the results were of low performance compared to the SSM s |
|---|