Identification of two-phase flow patterns based on capacitance data of electrical capacitance tomography with...

Area of Science:

• Multiphase flow analysis
• Tomographic imaging techniques
• Machine learning applications

Background:

• Electrical Capacitance Tomography (ECT) is crucial for flow pattern identification.
• Existing ECT algorithms struggle with accuracy for complex flows and require extensive labeled data.
• Generative Adversarial Networks (GANs) offer potential for improved pattern recognition.

Purpose of the Study:

• To propose a novel flow pattern identification method using ECT capacitance data.
• To leverage a semi-supervised generative adversarial network (SGAN) for enhanced accuracy and reduced data dependency.
• To compare the performance of SGAN against traditional machine learning models.

Main Methods:

• Developed a SGAN model tailored for ECT capacitance data.
• Constructed a comprehensive dataset of 11,400 random flow patterns via COMSOL and MATLAB simulations.
• Trained and validated SGAN, Back Propagation (BP), and Support Vector Machine (SVM) models.
• Conducted static experiments on a self-developed ECT system for comparative analysis.

Main Results:

• SGAN demonstrated higher average identification accuracy compared to BP and SVM.
• SGAN achieved superior performance even with ten times fewer labeled samples than other algorithms.
• The study validated the effectiveness of SGAN in complex flow pattern identification.

Conclusions:

• The proposed SGAN method significantly enhances flow pattern identification accuracy using ECT data.
• SGAN offers a more efficient approach by requiring substantially less labeled training data.
• This research presents a promising advancement for industrial multiphase flow monitoring.