Development of Virtual Flow Sensor using Artificial Neural Networks

Abstract

In this paper, a virtual flow sensor using artificial neural networks (ANN) is proposed to improve the efficiency of an industrial flow control loops. In conventional flow-control loop, flow meters used for sensing flow rate in the feedback path cause pressure drop in the flow. This may increase the energy usage for propelling the fluid. The functional relation between the flow rate and the physical properties of the flow through the final control element such as control valve is known and the said properties namely pressure drop, temperature, and valve position are yielded from an experimental set-up. These properties are used as training data for ANN models to yield the fluid flow rate through the control valve. Here, the ANN acts as a virtual flow sensor. The feasibility of the proposed methodology is validated by using real measurement of flow and used them to model virtual flow sensor using the multi-layer perceptron artificial neural networks (MLP-ANN) with back propagation (BP) algorithm. Moreover, its practical proof of concept is demonstrated by implementing the trained MLP-ANN on a Spartan-3E-starter Field Programmable Gate Array (FPGA) unit through a hardware co-simulation.