This paper presents the results of three uncertainty propagation and quantification methodologies, i.e., perturbation method, Taylor series expansion based approach, and Monte Carlo simulation, for deterministic multiphase fluid flow models. The sensitivity of pressure drop and liquid holdup predictions to slug length and liquid entrainment in gas core uncertainties is investigated for a multiphase fluid flow model, TUFFP Unified Model (1). The analysis shows that the Taylor series expansion consistently overestimates the output uncertainty for all cases when compared to perturbation method and the Monte Carlo simulation results. The results of perturbation method are similar to Monte Carlo simulation results.
Multiphase flow occurs in many systems encountered in chemical, petroleum, and nuclear industries. Because efficient design and operation of these systems depend on the ability to predict the flow characteristics, several multiphase flow models have been developed. These models are based on mechanistic approach, which utilizes the conservation equations of mass, momentum, and energy along with the empirical or semi-mechanistic closure relationships and adjustable parameters. The resulting equations are solved using various numerical methods and algorithms. The inputs to these models include fluid properties such as gas and liquid densities, viscosities, and surface tensions, pipeline geometry such as diameter and inclination angle, and operating conditions such as superficial gas and liquid velocities. The outputs are flow patterns, pressure gradient, liquid holdup, and flow characteristics. The multiphase flow models are used extensively to design and operate multiphase flow pipelines with considerable success. However, the predictions from these models can occasionally differ up to 100% compared to field data. We argue that part of this discrepancy is due to the inherent uncertainty in these models. However, these models are not equipped to represent, propagate, and quantify the resulting uncertainty in the model predictions, because they are deterministic.