Real-time automatic flow regime classification and mapping for vertical pipes using dynamic pressure signals

Accurate flow regime identification is essential for modeling two-phase flow systems, but the literature on real-time applications in vertical pipes is scarce. This work aims to develop a real-time, automated, data-driven flow regime classifier for vertical pipes using dynamic pressure signals. These signals were collected using a numerical model to represent three distinct flow regimes—slug, churn, and annular—in a 3-inch vertical pipe. Features were then extracted from these signals using Discrete Wavelet Transform (DWT). To optimize classification performance, twelve dimensionality reduction techniques were evaluated, followed by the application of an AutoML framework to identify the most effective machine learning classifier among K-Nearest Neighbors (KNN), Artificial Neural Networks, Support Vector Machines (SVM), Gradient Boosting, Random Forest, and Logistic Regression, with hyperparameter tuning incorporated. Kernel Fisher Discriminant Analysis (KFDA) demonstrated the best clustering performance, while KNN emerged as the top classifier with 90.2% accuracy and excellent repeatability. Leveraging DWT, KFDA, and KNN, a virtual flow regime map was constructed, enabling real-time flow regime identification with a moving window of pressure signals. Verification of the model using a 50.8 mm (2-inch) diameter pipe at different locations confirmed its robustness and scalability. In the final stage, a unified flow regime map was developed for both horizontal and vertical pipes, achieving 100% training and 92.5% testing accuracy using DWT, KFDA, and ANN. The proposed workflow represents a significant step forward in automating flow regime identification, enabling its application to opaque pipes fitted with pressure sensors for flow assurance and monitoring in process industries.