Development and Research of the Neural Network Pump Efficiency Observer Based on the Programmable Logical Integral Scheme

Abstract

This paper deals with the actual task of design the technological parameter observer of the pumping unit. This approach can be applied into the systems where it is impossible to mount appropriate sensors without invasion into the hydraulic network.

For the technical implementation of the observer, the technology of neural networks is used, which, allows estimating the values of other coordinates based on the measured values, in this case, the efficiency of the pump. For the neural network training were used experimental captured arrays of pressure, efficiency and active power consumption of the pump unit.

Based on the choice of the neural network type and its settings, a mathematical description of the estimator for its implementation was used Altera DE1-SoC developer board, which was equipped with a field programmable logic device of the Cyclone V family and the Nios II processor.

Experimental investigation was carried out on a laboratory bench, consisting of a 0,33 kWt pump plant, a water supply system and Lenze 8200 Vector frequency converter operating in frequency control mode, to test the performance of the developed observer. Experiments were carried out for different operating points of the pump and at different load. In this case, the experimentally captured characteristics were compared with the calculated efficiency and the efficiency, which was estimated by the neural network.

The data analysis showed that the use of the neural network to evaluate the efficiency yields the maximum deviation of the estimated values in comparison with the cataloging characteristics is no more than 3 %, which is acceptable, given that a small training array of static characteristics had been formed. It is expedient to use more accurate sensors and train the neural network in dynamic modes to reduce the error.

This approach reduces the number of measuring values for the design control systems of technological coordinates and realizes energy-efficient turbo-mechanism control algorithms, where access to corresponding quantities measurement is complicated or impossible.