Multiple phase flow identification using computational simulation and convolutional neural network

Abstract

The Identification of gas-solid flow characterization in dense-phase pneumatic conveying particles is very important to a vast area of industrial fields such as chemical and pharmaceutical industries since a slight change in flow characteristics results in a completely different product. The motion of the gas-solid two-phase flow in densephase usually has a nonlinear and unsteady nature that needs to be examined and analysed to identify the particle flow behaviour in the pneumatic conveying pipelines. In this research a method to identify the type of flow pattern is proposed using a computational method where a gravity flow rig is modelled on Solidworks and multiple flow patterns are simulated with different mass flow rates ranging between 200 to 600 g/s. For changing the flow patterns inside the pipe, an Iris Mechanism is designed according to the specifications of the flow required to achieve the flow pattern control. A sectioning method is implemented to capture flow images at the plane of interest for different flow patterns. Afterwards images are fed to a Convolutional Neural Network which is trained and tested to identify the flowpatterns according to several flowfeatures which resulted in 100% accuracy. A GUI using PyQt is designed to better visualize the whole system and view the predicted flow pattern.