Iterative Learning Control of a Left Ventricular Assist Device

Due to the lack of donor organs, the importance of left ventricular assist devices (LVADs) increases. State of the art is to operate the pumps with a constant speed (CS) leading to effects such as underpumping, ventricular suction or the backflow of blood from the aorta in the ventricle. The end-diastolic volume (EDV) is influenced by venous return, as well as diastolic function and systolic pressure development. Thus it is a good medical indicator of ventricular load. In this paper a norm-optimal iterative learning control (NOILC) algorithm is designed to shape the EDV of a pathological ventricle. In addition, further constraints such as for example a uniform filling of the ventricle and the prevention of pumping during the systole are considered. A simplified model of the systemic circulation and the pump is used to study the system response to changes in pulmonary vein pressure and to benchmark controller performance. The results show that the algorithm obtains an excellent tracking performance and prevents the dilation of the ventricle. The approach offers the physician the opportunity to control multiple physiological variables even though the optimal control output trajectory is not known. Future work will focus on the improvement of the controller performance regarding the rejection of instantaneous disturbances and the incorporation of variable cycle duration.