Flow pattern identification of liquid-liquid (oil and water) in vertical pipelines using machine learning techniques

Given the importance of process control in the petrochemical industry, there is a need to determine the behavior of the fluids inside the pipes. In this work a methodology is developed for the identification of flow patterns in vertical pipes with diameters between 0.01 m and 0.10 m, from the implementation of artificial intelligence techniques, for a liquid combination of two phases composed of oil with viscosity in the range of 792 Kg/m3 to 1823 Kg/m3 and water at room temperature. The predictive models generated in the structuring of the methodology were trained with 70% of data based on viscosity parameters, pipe diameter, volume fraction and surface velocities of the working fluids stored in a database. The remaining information, equivalent to 30% of the total, was used to develop the automatic model validation. The flow patterns identified by the intelligent system for oil and water flow, without considering the predominant substance, are churning, dispersed, very fine dispersion, transition flow, intermittent, and annular

Boiling Flow Pattern Identification Using a Self-Organizing Map

In the paper, a self-organizing map combined with the recurrence quantification analysis was used to identify flow boiling patterns in a circular horizontal minichannel with an inner diameter of 1 mm. The dynamics of the pressure drop during density-wave oscillations in a single pressure drop oscillations cycle were considered. It has been shown that the proposed algorithm allows us to distinguish five types of non-stationary two-phase flow patterns, such as bubble flow, confined bubble flow, wavy annular flow, liquid flow, and slug flow. The flow pattern identification was confirmed by images obtained using a high-speed camera. Taking into consideration the oscillations between identified two-phase flow patterns, the four boiling regimes during a single cycle of the long-period pressure drop oscillations are classified. The obtained results show that the proposed combination of recurrence quantification analysis (RQA) and a self-organizing map (SOM) in the paper can be used to analyze changes in flow patterns in non-stationary boiling. It seems that the use of more complex algorithms of neural networks and their learning process can lead to the automation of the process of identifying boiling regimes in minichannel heat exchangers.