Development of Gas-Liquid Flow Regimes Identification Using a Noninvasive Ultrasonic Sensor, Belt-Shape Features, and...

Area of Science:

• Multiphase flow dynamics
• Signal processing
• Machine learning applications

Background:

• Classifying gas-liquid two-phase flow regimes is crucial for industrial process control.
• Ultrasonic signals offer a non-intrusive method for monitoring flow patterns.

Purpose of the Study:

• To develop an automated method for classifying gas-liquid two-phase flow regimes.
• To introduce novel feature extraction techniques and a deep learning classifier for this task.

Main Methods:

• Feature extraction using novel belt-shaped features (BSFs) from preprocessed ultrasonic signals.
• Classification using a multi-stage convolutional neural network (CNN/ConvNet) with convolution and pooling layers.
• Validation using experimental data from an industrial-scale multiphase flow facility.

Main Results:

• The proposed ConvNet classifier achieved high accuracy: 97.40% (training), 94.57% (testing), and 94.94% (validation).
• BSF features effectively capture relevant information for flow regime identification.
• The method demonstrates robust performance on diverse flow regimes (annular, churn, slug, bubbly).

Conclusions:

• The combination of BSF features and a ConvNet classifier is a viable and accurate method for gas-liquid two-phase flow regime classification.
• This approach shows significant potential for real-time industrial applications.
• The study validates the effectiveness of deep learning in analyzing complex multiphase flow data.