Activities

Home > Activity > Newsletter > 2023.10

Newsletter  2023.11  Index

Theme : "Mechanical Engineering Congress, 2023 Japan (MECJ-23)"

  1. Preface
    Hideo MORI, Tetsuya KANAGAWA
  2. Why do mistakes communicate and spread? (Diffusion and prevention of misperceptions regarding fluid mechanics)
    Ryozo ISHIWATA (Kanagawa Institute of Technology)
  3. Toward Digital Twin Numerical Turbine
    Satoru YAMAMOTO (Tohoku University)
  4. Measurements of turbulent wall pressure fluctuation field in a turbulent boundary layer and the wing-flat plate juncture flow using the many-channel microphone array
    Yoshitsugu NAKA (Meiji University)
  5. PSP and TSP for measuring pressure and temperature fields on wall surfaces and their applications
    Yasuhiro EGAMI (Aichi Institute of Technology)
  6. Flow measurement using MEMS differential pressure sensor
    Hidetoshi TAKAHASHI (Keio University), Takuto Kishimoto (Keio University), Kei Ohara (Keio University), Kyota Shimada (Keio University)
  7. Flexible Sheet Sensor for Advanced Flow Monitoring
    Masahiro MOTOSUKE (Tokyo University of Science)

 

Toward Digital Twin Numerical Turbine

Abstract

Satoru YAMAMOTO,
Tohoku University

 

As one of my works within past 40 years’ research, Digital Twin Numerical Turbine (DTNT) was successfully developed. Numerical Turbine (NT) can simulate full-annulus flows through a gas turbine compressor and a steam turbine considering moist air and wet steam.  Recently, NT was applied to the simulation of full-annulus flows of a real gas turbine compressor under partial-load operation at a power plant. Fig. 1 represents the simulated instantaneous pressures on whole blade surfaces assuming full-load and partial-load operations. A stable flow was obtained in the case of full-load operation, whereas an unstable flow with stall cells was captured in the case of partial-load operation. The results indicate that the current partial-load operation to suppress the total electric power exceeded by solar-power generation may reduce the lifetime of gas turbines. DTNT finally forms a Self-Organizing Map (SOM) created by machine learning from hundreds of time-dependent data obtained from dozens of NT simulations setting different flow conditions. Fig. 2 shows the schematic of the process forming SOM from the NT simulations. The cases in the dotted line were supposed to be abnormal cases, clustering successfully. The most effective contribution to achieve the process was the significant reduction of computational time. Currently only 1.3 days are spent for one simulation of a full-annulus flow in 1.5-stage compressor using the supercomputer AOBA (SX-Arora Tsubasa) at Tohoku university. The activities of this work were reported by news medium in Oct. 2021 and June 2023. It is expected that normal and abnormal flow conditions of compressor operation can be predicted by the SOM at the design and in the operation.

Key words

CFD, Gas Turbine, Compressor, Numerical Turbine, Digital Twin

Figures

Fig. 1 Instantaneous pressure distributions on whole blade surfaces 
(Left:Full load,Right:Partial load)


Fig. 2 Data Clustering by Self-Organizing Map (SOM)

Last Update:11.29.2023