Dominant Modes in a Gas Cyclone Flow Field Using Proper Orthogonal Decomposition

Mayank Kumar, Surya Pratap Vanka, Raja Banerjee, Narasimha Mangadoddy

Research output: Contribution to journalArticlepeer-review


The complex turbulent swirling flow inside cyclone separators was studied using high-fidelity large eddy simulations (LESs). Using the appropriate subgrid-scale LES model, flow features such as mean and turbulent statistics were studied for the variations in the inlet Reynolds number (Re) and geometric swirl number (S). Spectral analysis of the turbulent fluctuations shows multiple peaks pertaining to multiple coherent turbulent structures. In order to reveal the turbulent flow modes and their transient evolution, proper orthogonal decomposition (POD) of the time-dependent LES-computed values was performed. At lower S, the single-helix mode was found to be the predominant energy-containing turbulent mode. An increase in S results in a high-frequency double-helix mode. An increase in the energy content of the double-helix mode was observed with increasing S. The reconstruction using the reduced order POD model reveals that the coherent helical modes extract energy from the high-strain regions of the mean flow and contribute significantly to the increased fluctuation levels observed in the RMS velocity profiles near the axis of the cyclone. The reduced order model was able to reconstruct the RMS velocity profile capturing over 90% of the total turbulent kinetic energy using 200 of the most energetic POD modes and was multiple times faster than the transient LES simulation.

Original languageEnglish (US)
Pages (from-to)2562-2579
Number of pages18
JournalIndustrial and Engineering Chemistry Research
Issue number6
StatePublished - Feb 16 2022

ASJC Scopus subject areas

  • General Chemistry
  • General Chemical Engineering
  • Industrial and Manufacturing Engineering


Dive into the research topics of 'Dominant Modes in a Gas Cyclone Flow Field Using Proper Orthogonal Decomposition'. Together they form a unique fingerprint.

Cite this