Predicting Air Superficial Velocity of Two-Phase Reactors Using ANFIS and CFD

Area of Science:

• Fluid Dynamics
• Chemical Engineering
• Artificial Intelligence

Background:

• Predicting turbulence in gas-liquid flows is computationally intensive.
• Computational Fluid Dynamics (CFD) methods are essential but time-consuming.
• Machine learning offers potential for accelerating these predictions.

Purpose of the Study:

• To integrate machine learning with CFD for predicting gas (bubble) flow turbulence.
• To reduce computational time in simulating multiphase flow systems.
• To identify effective input parameters and operating conditions impacting flow dynamics.

Main Methods:

• Combined machine learning and Eulerian methods for gas flow distribution estimation.
• Utilized Adaptive Neuro-Fuzzy Inference System (ANFIS) to train CFD data.
• Trained ANFIS with gas velocity and turbulent eddy dissipation rate from CFD nodes in a bubble column reactor (BCR).

Main Results:

• ANFIS successfully predicted artificial BCR performance, mimicking CFD results.
• Reduced the need for expensive and time-consuming numerical simulations.
• Demonstrated that the number of input parameters significantly affects prediction accuracy.

Conclusions:

• The ANFIS model accurately predicts turbulent eddy dissipation rate and gas flow patterns.
• Input data and membership function specifications critically influence model accuracy.
• This hybrid approach offers a computationally efficient alternative for multiphase flow analysis.