Dynamic modeling and identification of an axial flow ventricular assist device

An accurate characterization of the hemodynamic behavior of ventricular assist devices (VADs) is of paramount importance for proper modeling of the heart-pump interaction and the validation of control strategies. This paper describes an advanced test bench, which is able to generate complex hydraulic loads, and a procedure to characterize rotary blood pump performance in a pulsatile environment. Special focus was laid on model parameter identifiability in the frequency domain and the correlation between dynamic and steady-state models. Twelve combinations of different flow/head/speed signals, which covered the clinical VAD working conditions, were generated for the pump characterization. Root mean square error (RMSE) between predicted and measured flow was used to evaluate the VAD model. The found parameters were then validated with broadband random signals. In the experiments the optimization process always successfully converged. Even in the most demanding dynamic conditions the RMSE was 7.4 ml/sec and the absolute error never exceeded 24.9 ml/sec. Validity ranges for the identified VAD model were: flow 0-180 ml/sec; head 0-120 mmHg; speed 7.5-12.5 krpm. In conclusion, a universal test bench and a characterization procedure to describe the hydrodynamic properties of rotary blood pumps were established. For a particular pump, a reliable mathematical model was identified that correctly reproduced the relationship between instantaneous VAD flow, head and impeller speed.