Developing semiautonomous system for robust performance of centrifugal pumping system

摘要

Centrifugal pumps are considered as one of the most common types of pumps in industry today. They are widely-used because their design is simple, they are available in wide range of capacity and head, they have high efficiency and smooth flow rate, and they are easy to operate and maintained. There are many factors that contribute to the efficiency and life time of centrifugal pumps. In this work, a control method has been implemented practically to keep the centrifugal pump working on Best Efficiency Point (BEP). This control governors two parameters: speed of the motor, valve angle of Control Valve (CV) that effect the response of the system. Two control systems have been devised in this work: the two-single control and dual control ones. The two-single control system has two systems that work in an alternative manner: the speed control and valve angle control systems. We test three kinds of controllers and compare among them to obtain the best performance, they are PID, predictive neural network and NARMA-L2 neural network controller. The speed and valve angle systems are designed based on the experimental data through finding the Transfer Functions (TF). Consequently, we use theses TFs to model the relationship between input and output of the two systems. The inputs to the proposed systems are the speed of the motor and valve angle of CV, respectively while the output is fluid flow. After completing the design of the systems, we use the flow rate as reference inputs. On the other hand, the dual control system controls simultaneously the speed of the motor and valve angle and use neuro-fuzzy controller. We train the neuro-fuzzy controller using the experimental data to achieve the BEP. From the obtained results, the NARMA-L2 NN has proved to be the best controller among the three suggested controllers for the single system. The NARMA-L2 NN has provided considerable reduction of settling time, overshoot and error steady state. For the dual controller, Neuro-Fuzzy shows good performance for centrifugal pump systems without pre-modeling requirements. Moreover, there is a good agreement between the reference input and the system output. This is true as the percentage of the error is 0,25% as a maximum value.