Related Experiment Video
Updated: Aug 16, 2025

Microfluidic Devices for Characterizing Pore-scale Event Processes in Porous Media for Oil Recovery Applications
Published on: January 16, 2018
Soft measurement of oil-water two-phase flow using a multi-task sequence-based CapsNet
Lei OuYang1, Ningde Jin1, Landi Bai1
1School of Electrical and Information Engineering, Tianjin University, Tianjin, 300072, People's Republic of China.
This study developed an improved CapsNet for predicting oil-water two-phase flow. The enhanced model accurately predicts flow patterns and superficial velocity in large pipes.
Area of Science:
- Multiphase flow dynamics
- Fluid mechanics
- Chemical engineering
Background:
- Accurate measurement of flow parameters is crucial for understanding two-phase flow.
- Predicting superficial velocity in oil-water two-phase flow within large diameter pipes remains a significant challenge due to complex flow structures.
Purpose of the Study:
- To develop an advanced deep learning model for accurate prediction of flow patterns and superficial velocity in vertical upward oil-water two-phase flow.
- To address the challenges posed by high-dimensional, time-varying, and nonlinear characteristics of multiphase flow.
Main Methods:
- Conducted vertical upward oil-water two-phase flow experiments in a 125 mm ID pipe using a vertical multi-electrode array (VMEA) conductance sensor.
- Employed novel data pre-processing (1D to 2D) and information fusion techniques (network channels).
- Optimized a multi-task sequence-based Capsule Network (CapsNet) with attention blocks, residual structures, inception blocks, and an improved dynamic routing algorithm for enhanced performance.
Main Results:
- The proposed multi-task sequence-based CapsNet demonstrated superior performance in both flow pattern classification and superficial velocity prediction compared to its variants and other networks.
- The optimized network architecture and improved dynamic routing algorithm effectively handled the complexities of the two-phase flow data.
- Extensive experiments validated the effectiveness of the improved modules and the overall network.
Conclusions:
- The developed multi-task sequence-based CapsNet shows significant potential for real-world applications in multiphase flow analysis.
- The model offers a robust solution for high-dimensional, time-varying, and nonlinear problems encountered in industrial fluid dynamics.
- This research advances the capabilities of sensor data analysis for precise flow parameter estimation.
Related Concept Videos
Uniform Depth Channel Flow: Problem Solving
Pipe Flowrate Measurement: Problem Solving
Pipe Flowrate Measurement
The orifice meter is a simple,...
Measurement of Fluid Pressure
A basic form of manometer is the piezometer, a vertical tube open at the top and filled with the same...
Uniform Depth Channel Flow
Rapidly Varying Flow

