MEchanics GAthering -MEGA- Seminar: Robust design of herringbone grooved journal bearings supported turbocompressors using artificial neural networks

Thumbnail

Event details

Date 21.04.2022
Hour 16:1517:30
Speaker Massoudi Soheyl (LAMD, EPFL)
Location Online
Category Conferences - Seminars
Event Language English
Abstract Although robustness is an important consideration to guarantee the performance of designs under deviations, systems are often engineered by evaluating their performance exclusively at nominal conditions. Robustness is sometimes evaluated a-posteriori through a sensitivity analysis, which does not guarantee optimality in terms of robustness. Herringbone grooved journal bearings (HGJB) supported turbocompressors need tight manufacturing tolerances on the bearing clearance and the groove depth to run in a stable mode. This presentation introduces the automated design of HGJB supported turbocompressors via multi-objective optimization including robustness as an additional competing objective. Robustness is computed as 1) the sampled hypervolume of the feasible region under constraints and as 2) the signal-to-noise ratio of the performance metric over the same sampled hypervolume. In order to address the high number of additional evaluations needed to compute robustness, ensembles of artificial neural networks are used to generate fast and accurate surrogates of high-fidelity models. The developed methodology was applied to the design of a small scale turbocompressor. Robustness was included as a hypervolume objective and a signal-to-noise ratio of the load capacity, alongside losses. An experimentally validated 1D HGJB model and an intersection mode identification model were used to generate the training data. The optimization results suggests a clear competition between the two definitions of robustness and the bearings and rotor losses, while the use of neural networks led to a speed up by 3 orders of magnitude compared to the 1D code.

Bio Soheyl Massoudi currently works as a Ph.D. candidate at the Laboratory of Applied Mechanical Design (LAMD). His research focuses on Design methodology via surrogate-model assisted optimization using artificial neural networks and evolutionary algorithms. The target application is gas-bearings supported turbocompressors for heat-pumps and fuel cells.

Practical information

  • General public
  • Free

Organizer

  • MEGA.Seminar Organizing Committee

Tags

Solids Structures Fluids

Share