|
CARDIOVASCULAR
|
Volume 8, 2003, No 1-2 |
A Neural Network-Based Controller for an
Artificial Heart
M. J. Conlon, D. L. Russell, T. Mussivand
Background: A control system for a ventricular assist device
(VAD) must be reliable and flexible, able to react to changes in the human
circulatory system. Compared to traditional control strategies, neural network
controllers adapt more easily to systemic changes. A neural network-based
controller was designed to regulate the operation of a simple VAD and its
performance was evaluated using a numerical simulation.
Methods: The human circulatory system was modelled using thirteen first
order differential equations, and this numerical model was used to tune and
evaluate a neural network controller. The controller has four inputs (current
pump output, current systolic fraction, previous diastolic force, and previous
systolic force) and two outputs (suggested changes to the systolic and diastolic
forces for the next beat) and is trained to behave as a fixed-rate controller
under normal physical conditions, in an attempt to maintain a constant
peripheral blood flow rate. The change in peripheral flow resistance is
estimated by the controller, and used to adjust the cardiac output. In this
study, the controller was trained to react to a drop in peripheral resistance by
increasing the cardiac output.
Results: The controller is able to achieve its design goals, maintaining
a peripheral blood flow rate of 4.8 l/min under nominal conditions. If the
controller detects a drop in peripheral resistance, which might indicate (for
example) a heightened activity level, it reacts by increasing the cardiac
output. Simulated peripheral flow rates using the neural network controller are
superior to those achieved using either a fixed-rate or full-fill, full-eject
control strategy, as the neural network controller responds more effectively to
changes in the circulatory system parameters.
(CVE. 2003; 8 (1/2): 32-39)
Key words: ventricular assist device, neural network,
control, computer simulation, cardiovascular system, dynamics
Prof. Donald L. Russell
Department of Mechanical and Aerospace Engineering
Carleton University
1125 Colonel By Drive
Ottawa K1S 5B6, Ontario
Canada
E-mail: drussell@mae.carleton.ca
PABST SCIENCE PUBLISHERS
Lengerich, Berlin, Riga, Rom,
Wien, Zagreb