Assistant Professor Harvard Medical School Boston, Massachusetts, United States
Introduction: : Mechanical circulatory support (MCS) devices, such as Extracorporeal Membrane Oxygenation (ECMO) and Extracorporeal Cardiopulmonary Resuscitation (ECPR) systems, are vital for sustaining patients with severe cardiopulmonary failure. However, their effectiveness can be compromised by dynamic challenges, including variable hypovolemia (volume depletion), systemic hypertension (vascular overload), and device-related faults such as oxygenator degradation. Traditional control approaches often struggle to maintain stability in the face of such complex, evolving disturbances. In this work, we propose a dual-loop adaptive control framework designed to ensure robust hemodynamic stability across a wide range of clinical conditions. The architecture integrates a fast inner loop for direct regulation of blood flow and oxygenation with an adaptive outer loop that actively compensates for systemic disturbances in real time. Building upon prior modeling efforts, we introduce simulated fault injections and physiological variability to highlight safety, early fault detection and recovery, and patient-specific adaptation. Extensive simulations demonstrate that the proposed resilient control system significantly enhances the robustness of ECMO/ECPR support under scenarios of hypovolemia, systemic hypertension, and mechanical faults, ultimately advancing the development of safer, smarter, and more autonomous life support technologies.
Materials and
Methods: : We modeled the circulatory system using a simplified lumped-parameter cardiovascular framework. The control architecture employed a cascaded dual-loop feedback design, consisting of a fast inner Device-Level Flow and Oxygenation Controller and an adaptive outer Patient-Level Physiological Controller. The ECMO pump was integrated with two hierarchical control layers (Figure 1): (A) Inner Loop: A Proportional-Integral (PI) controller regulated the immediate pump flow rate to maintain target perfusion, compensating for sensor noise and high-frequency fluctuations. (B) Outer Loop: A Model Predictive Controller (MPC) dynamically adjusted the inner loop setpoints based on systemic vascular resistance, preload, and target perfusion pressure. We simulated various clinical disturbances including hypovolemia and systemic hypertension to evaluate system performance. Controller parameters were optimized to minimize deviation from target arterial pressure while maintaining physiological flow rates. The entire system was implemented and simulated in Python using SciPy and control libraries. Stability margins, overshoot, recovery times, and steady-state errors were systematically analyzed for validation.
Results, Conclusions, and Discussions:: Results Simulation results demonstrate that the proposed dual-loop system outperforms conventional single-loop control in resilience, recovery speed, and fault tolerance. Under normal conditions, the inner-loop PI controller achieved flow tracking within ±5% of target values. The outer-loop MPC layer enabled adaptive adjustment to dynamic vascular loads, maintaining arterial pressure within ±8 mmHg of the target during simulated hypotensive events. In disturbance recovery scenarios, flow recovery time after hypovolemic events was significantly reduced from 20.3 ± 3.1 seconds under single-loop control to 11.8 ± 1.9 seconds with dual-loop systems (p < 0.01) (Figure 2). Fault injection experiments (e.g., simulated pump partial occlusion) demonstrated that the resilient architecture restored stable hemodynamics within 15 seconds without inducing oscillations (Figure 3). Furthermore, the control system effectively prevented over-perfusion during low-resistance conditions and minimized ischemic injury risk during acute afterload elevation. Overall, the resilient control framework demonstrated superior robustness, reduced intervention time, and maintenance of more physiological hemodynamic profiles compared to conventional control methods. Conclusions and Discussions Our dual-loop resilient control architecture offers a promising advancement for next-generation circulatory assist devices. By combining rapid inner-loop stabilization with adaptive outer-loop regulation, the system consistently maintained perfusion targets despite patient-specific variability, device faults, and physiological disturbances. These features have the potential to significantly improve clinical outcomes by reducing adverse events associated with ECMO support, including hemodynamic instability and end-organ hypoperfusion. Future work will focus on expanding the model to capture more detailed patient-specific cardiovascular dynamics and validating the control system through in vitro bench testing and preclinical models. Additionally, integrating machine learning-based disturbance prediction modules could further enhance real-time adaptive performance. Ultimately, resilient intelligent control systems such as the one proposed here may substantially improve both short-term and long-term outcomes for critically ill patients requiring MCS.
