Gas-solid Flow Patterns Identification Based on Artificial Neural Network

Abstract

Flow patterns identification of the gas--solid flow in pneumatic transport pipelines is significant for the optimized design and operation of the pneumatic conveying system. The objective of this work is to training an Artificial Neural Network(ANN) to identify flow patterns (suspension flow, laminar flow, dense-dilute flow and dune flow) of the gas-solid flow in a horizontal pneumatic conveying pipeline. The performance of the ANN models was evaluated respectively using Hurst exponent of a ring-shaped electrode's output signal and Hurst exponent matrix of an electrostatic sensor array's output signals. Results show a higher recognition rate can be got by using the electrode sensor array, and the improvement is 5% for suspension flow, 9% for laminar flow and 13% for dense-dilute flow.