DTSMC and FNN control of left ventricular assist device to prevent suction and overperfusion

In this work, an intelligent physiological control strategy for left ventricular assist devices (LVADs) is proposed based on the Frank–Starling mechanism (FSM). The method introduces a mathematically derived target-flow formulation that defines the operating point of the pump using preload-dependent control lines constructed from the mean pump flow and flow pulsatility characteristics. This formulation enables adaptive regulation of pump support in response to changing cardiovascular demand. To robustly regulate the operating point and prevent adverse events such as suction and over-perfusion, a nonlinear control framework combining discrete-time terminal sliding mode control (DTSMC) and a fuzzy neural network (FNN) is developed. The DTSMC component ensures finite-time convergence and robustness against model uncertainties, while the FNN provides online approximation of unknown nonlinearities and disturbances, thereby reducing chattering and improving tracking performance. Lyapunov-based stability analysis is employed to rigorously establish closed-loop stability. The proposed controller is evaluated using a coupled cardiovascular system (CVS)–LVAD model to assess its physiological relevance and effectiveness. Two computational scenarios are conducted to emulate rest and physical exertion conditions in heart failure patients by systematically varying CVS parameters at 30 s intervals over a 3 s transition period. Simulation results demonstrate that the proposed control strategy successfully maintains the LVAD operating point within a safe and physiologically meaningful range, regulating pump flow between 3.6 L/min to prevent undersupply and 5.2 L/min to avoid over-perfusion. These results confirm the ability of the proposed approach to deliver adaptive, robust, and physiologically responsive LVAD support under varying cardiovascular conditions.