MEchanics GAthering –MEGA- Seminar: Enabling Robust and Integrated Design of Gas-Bearing Supported Turbocompressors Through Neural Network Ensemble Surrogate Models
Event details
| Date | 18.04.2024 |
| Hour | 16:15 › 17:15 |
| Speaker | Soheyl Massoudi (LAMD, EPFL) |
| Location | Online |
| Category | Conferences - Seminars |
| Event Language | English |
Abstract: Modern engineering challenges in the design of small-scale turbocompressors call for a multi-disciplinary approach, necessitating the consideration of robustness against manufacturing imperfections alongside traditional performance metrics. This presentation introduces a framework utilizing ensembles of neural networks to create surrogate models, significantly enhancing the robustness and integration in the design process of gas-bearing supported turbomachinery.
We examine the limitations of conventional design optimization, which focuses on component-level metrics rather than holistic system robustness. Our work demonstrates that maximizing nominal performance metrics often fails to account for the manufacturing realities, potentially compromising system reliability and safety margins.
In response, we propose an automated, multi-objective optimization framework that incorporates robustness as a core design objective, evaluated against geometrical and operational deviations. This approach is empowered by the speed and accuracy of neural network surrogates, which replace time-intensive high-fidelity models, expediting the design process while maintaining precision.
Case studies on micro-turbocompressor applications reveal clear trade-offs between efficiency, operational range, and robustness. The application of our methodology to herringbone grooved journal bearings underlines the importance of integrated design, offering practical design guidelines and an interactive design and simulation tool, DARTS-NETGAB, which harnesses real-time analysis and 3D construction capabilities for engineers and researchers.
This research embodies a substantial stride forward in the domain of turbocompressor design, advocating for a paradigm shift towards integrated optimization and robustness. The seminar will discuss the implications and applications of the proposed design framework in enhancing the resilience and performance of turbocompressors.
Bio:Soheyl Massoudi is a Ph.D. candidate at the Laboratory for Applied Mechanical Design (LAMD), EPFL. He joined the lab after doing his B.Sc. and M.Sc. specializing in fluid dynamics with a minor in energy at the EPFL. His research focuses on design methodology via surrogate-model assisted optimization using artificial neural networks and evolutionary algorithms. The target application is gas-bearing supported turbocompressors for heat pumps and fuel cells.
We examine the limitations of conventional design optimization, which focuses on component-level metrics rather than holistic system robustness. Our work demonstrates that maximizing nominal performance metrics often fails to account for the manufacturing realities, potentially compromising system reliability and safety margins.
In response, we propose an automated, multi-objective optimization framework that incorporates robustness as a core design objective, evaluated against geometrical and operational deviations. This approach is empowered by the speed and accuracy of neural network surrogates, which replace time-intensive high-fidelity models, expediting the design process while maintaining precision.
Case studies on micro-turbocompressor applications reveal clear trade-offs between efficiency, operational range, and robustness. The application of our methodology to herringbone grooved journal bearings underlines the importance of integrated design, offering practical design guidelines and an interactive design and simulation tool, DARTS-NETGAB, which harnesses real-time analysis and 3D construction capabilities for engineers and researchers.
This research embodies a substantial stride forward in the domain of turbocompressor design, advocating for a paradigm shift towards integrated optimization and robustness. The seminar will discuss the implications and applications of the proposed design framework in enhancing the resilience and performance of turbocompressors.
Bio:Soheyl Massoudi is a Ph.D. candidate at the Laboratory for Applied Mechanical Design (LAMD), EPFL. He joined the lab after doing his B.Sc. and M.Sc. specializing in fluid dynamics with a minor in energy at the EPFL. His research focuses on design methodology via surrogate-model assisted optimization using artificial neural networks and evolutionary algorithms. The target application is gas-bearing supported turbocompressors for heat pumps and fuel cells.
Practical information
- General public
- Free
Organizer
- MEGA.Seminar Organizing Committee