Reduced order modeling of the unsteady pressure on turbine rotor blades using deep learning
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2025
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866918128726835200 |
|---|---|
| author | Joachim, Dominique Lionel, Salesses Jean-François, Thomas Lieven, Baert Tariq, Benamara Franck, Mastrippolito Theo, Flament |
| author_facet | Joachim, Dominique Lionel, Salesses Jean-François, Thomas Lieven, Baert Tariq, Benamara Franck, Mastrippolito Theo, Flament |
| contents | In transonic turbine stages, complex interactions between trailing edge shocks from nozzle guide vanes and rotor blades generate unsteady wall pressure fields, impacting rotor aerodynamic performance and structural integrity. While shock-related phenomena are prominent, unsteady pressure fluctuations can also arise in subsonic regimes from wake interactions. Traditional methods like Unsteady Reynolds-Averaged Navier-Stokes (URANS) simulations are accurate but computationally expensive. To address this, a novel deep learning-based Reduced Order Model (ROM) is proposed, built on a database of URANS simulations, to predict unsteady pressure fields on turbine rotor blades at a fraction of the cost. The model consists of a Variational Auto-Encoder (VAE) integrated with a Gated Recurrent Unit (GRU) to capture time-series data, overcoming the limitations of traditional linear ROMs in capturing nonlinear phenomena, such as moving shocks. The goal is to develop a ROM that accurately reproduces unsteady pressure fields from URANS simulations while reducing computational costs. The ROM is applied to the Turbine Aero-Thermal External Flows (TATEF2) project configuration, a representative test case in turbomachinery research. Model performance is evaluated using machine learning quality metrics and design-oriented criteria, including the accuracy of the first harmonic in the Fourier transform of the unsteady pressure field. The impact of the simulation database size on model accuracy is also analyzed, considering the number of training simulations required for task-specific accuracy as a key factor in industrial applicability. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2503_00013 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Reduced order modeling of the unsteady pressure on turbine rotor blades using deep learning Joachim, Dominique Lionel, Salesses Jean-François, Thomas Lieven, Baert Tariq, Benamara Franck, Mastrippolito Theo, Flament Fluid Dynamics In transonic turbine stages, complex interactions between trailing edge shocks from nozzle guide vanes and rotor blades generate unsteady wall pressure fields, impacting rotor aerodynamic performance and structural integrity. While shock-related phenomena are prominent, unsteady pressure fluctuations can also arise in subsonic regimes from wake interactions. Traditional methods like Unsteady Reynolds-Averaged Navier-Stokes (URANS) simulations are accurate but computationally expensive. To address this, a novel deep learning-based Reduced Order Model (ROM) is proposed, built on a database of URANS simulations, to predict unsteady pressure fields on turbine rotor blades at a fraction of the cost. The model consists of a Variational Auto-Encoder (VAE) integrated with a Gated Recurrent Unit (GRU) to capture time-series data, overcoming the limitations of traditional linear ROMs in capturing nonlinear phenomena, such as moving shocks. The goal is to develop a ROM that accurately reproduces unsteady pressure fields from URANS simulations while reducing computational costs. The ROM is applied to the Turbine Aero-Thermal External Flows (TATEF2) project configuration, a representative test case in turbomachinery research. Model performance is evaluated using machine learning quality metrics and design-oriented criteria, including the accuracy of the first harmonic in the Fourier transform of the unsteady pressure field. The impact of the simulation database size on model accuracy is also analyzed, considering the number of training simulations required for task-specific accuracy as a key factor in industrial applicability. |
| title | Reduced order modeling of the unsteady pressure on turbine rotor blades using deep learning |
| topic | Fluid Dynamics |
| url | https://arxiv.org/abs/2503.00013 |