<p>This study proposes a novel machine learning-based approach for fast and accurate classification of the wavy flow sub-regimes in gas-liquid pipelines using supervised Logistic Regression Models (LRM). While most previous works focused on main flow regimes, the classification of wavy sub-regimes remained largely unexplored. Early detection is crucial, as wavy flow, especially roll waves, often precedes slug formation, which can cause sudden pressure surges and operational hazards. The model aims to enable timely interventions to ensure pipeline safety. Using Photron high-speed camera, a comprehensive dataset of 100,000 images was captured at 4000 fps from horizontal pipe flows with diameters of 12.5, 25, and 50 mm. The dataset includes five sub-regimes: stratified, 2D Small Amplitude waves (2DSA), 3D Small Amplitude waves (3DSA), 2D Large Amplitude waves (2DLA), and 3D Large Amplitude waves (3DLA). Four machine learning based LRMs- Multimodal (MLRM), Lasso (LLRM), Ridge(RLRM), and Elastic Net (ENLRM) Logistic Regression Models were trained and optimized using random search for hyperparameter tuning. The models demonstrated excellent performance across all flow regimes, achieving near-perfect classification accuracy (<InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="11760_2025_4410_Article_IEq1.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="49" /> </InlineMediaObject> <EquationSource Format="TEX">$\approx 98\%$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mo>≈</mo> <mn>98</mn> <mo>%</mo> </mrow> </math></EquationSource> </InlineEquation>) for stratified flow. Among the sub-regimes, LLRM consistently outperformed other models, showing superior accuracy in distinguishing complex patterns such as 2DSA, 3DSA, and 2DLA. Validation accuracies ranged from 91.06% to 96.91%, with MLRM and LLRM consistently exceeding 94% across evaluation metrics. Therefore, this study provides an effective ANN-based solution for wavy flow sub-regime identification, supporting reliable flow monitoring in industrial pipelines. Future work includes real-time implementation on embedded platforms, hybridization with CNNs to enhance accuracy, and incorporation of pressure data to improve identification.</p>

Classification of Wavy Flow Sub-Regimes Using Supervised Logistic Regression Models

期刊论文
发表日期：

2025年6月26日

查看全文

Tarannum Sallauddin Mujawar,
Jyotirmay Banerjee

摘要

This study proposes a novel machine learning-based approach for fast and accurate classification of the wavy flow sub-regimes in gas-liquid pipelines using supervised Logistic Regression Models (LRM). While most previous works focused on main flow regimes, the classification of wavy sub-regimes remained largely unexplored. Early detection is crucial, as wavy flow, especially roll waves, often precedes slug formation, which can cause sudden pressure surges and operational hazards. The model aims to enable timely interventions to ensure pipeline safety. Using Photron high-speed camera, a comprehensive dataset of 100,000 images was captured at 4000 fps from horizontal pipe flows with diameters of 12.5, 25, and 50 mm. The dataset includes five sub-regimes: stratified, 2D Small Amplitude waves (2DSA), 3D Small Amplitude waves (3DSA), 2D Large Amplitude waves (2DLA), and 3D Large Amplitude waves (3DLA). Four machine learning based LRMs- Multimodal (MLRM), Lasso (LLRM), Ridge(RLRM), and Elastic Net (ENLRM) Logistic Regression Models were trained and optimized using random search for hyperparameter tuning. The models demonstrated excellent performance across all flow regimes, achieving near-perfect classification accuracy ( $\approx 98\%$ $\approx 98 %$ ) for stratified flow. Among the sub-regimes, LLRM consistently outperformed other models, showing superior accuracy in distinguishing complex patterns such as 2DSA, 3DSA, and 2DLA. Validation accuracies ranged from 91.06% to 96.91%, with MLRM and LLRM consistently exceeding 94% across evaluation metrics. Therefore, this study provides an effective ANN-based solution for wavy flow sub-regime identification, supporting reliable flow monitoring in industrial pipelines. Future work includes real-time implementation on embedded platforms, hybridization with CNNs to enhance accuracy, and incorporation of pressure data to improve identification.