A new hybrid physics–ML model for predicting over-reading in differential pressure measurements under wet gas flow

Accurate measurement of wet gas flows remains a longstanding challenge across various industries. Differential pressure (DP) meters, particularly Venturi tubes, are commonly used for metering wet gas flows, but the presence of liquid induces an over-reading of the actual gas flow rate. This study proposes a new over-reading correction method using a hybrid modeling strategy that combines the physics-based homogeneous flow model with a machine learning model trained to predict its residuals. The hybrid model was developed and evaluated using an extensive database compiled from seven experimental studies on horizontally mounted Venturi meters. Compared to the base model, the hybrid approach demonstrated a significantly enhanced performance, reducing both RMSRE and maximum relative error by more than 80% and effectively addressing its systematic errors. Moreover, it provided more accurate predictions relative to a purely data-driven model developed in this work. It also outperformed existing correlations, achieving a 45% reduction in RMSRE compared to the most accurate among them, the ISO/TR 11583:2012 correlation. The proposed method represents a promising solution to the challenges associated with over-reading in wet gas flow metering, and can also be adapted to other configurations and types of DP meters.