Electro pump modeling using laboratory system data

Pumps are widely used in various industries and they are mostly driven by induction motors. For purposes such as system control, monitoring and model-based fault detection, a dynamic model of the motor-pump behavior is necessary. We used a nonlinear model to describe the behavior of an electro pump. The main problem to model the pump are the unknown parameters of hydraulic part of the pump. In this paper, two approaches are proposed to identify the hydraulic part of the pump. First, laboratory system data are used to train a neural network which identifies a black-box model for the hydraulic part of the pump. In the second approach, the parameters are estimated via least squares using laboratory system data. The main advantage of the proposed methods is that the impeller vane angel need not be measured to model the hydraulic part of the pump. Finally, the model is simulated with both proposed methods, and results are evaluated on a laboratory system. These results show the model describes the electro pump dynamic with good precision.