The International Journal of Artificial Organs / Vol. 23 / no. 10, 2000 / pp.

47-51

Artificial Heart and Cardiac Assist Devices

Modeling and control of a magnetic bearing system for

the magnetically suspended centrifugal blood pump
H. KIM 1 AND H.C. KIM 2
1
Interdisciplinary Program, Biomedical Engineering Major, Graduate School
2
Department of Biomedical Engineering, College of Medicine, and Institute of Medical and Biological Engineering, MCR,
Seoul National University, Seoul - Korea

ABSTRACT: We have developed a centrifugal blood pump with a magnetically suspended impeller to
reduce the hemolysis level for long-term use. The main advance we have made was simplifying the
traditional 5 axes controlled magnetic bearing system, since the total device should ultimately be small
enough to implant into a human body. A simplified magnetic bearing system with only 3 actively
controlled axes was designed to levitate the impeller. The state equations of the magnetic bearing
plant were established using Lagrange’s methods, and using the results, a linear quadratic optimal
controller was designed to regulate the position of the impeller. Flow rate and differential pressure of
5L/min and 300mmHg, respectively at a motor speed of 2750 rpm was achieved from a mock
circulation setup. These results show satisfactory performance as a cardiopulmonary bypass device.
The lowering of pumping efficiency as the clearance between the impeller and pump housing
increased was negligible. Thus for power efficiency, the smallest gap not leading to a hemolysis level
rise can be determined as a feasible operation clearance.

KEY WORDS: Centrifugal blood pump, Magnetic bearing system, Linear quadratic optimal controller

INTRODUCTION As a class of rotary pumps, the centrifugal blood pump

Mechanical circulatory support system are widely used
as cardiopulmonary bypass devices and implantable
ventricular assist devices. Recently, research on long-term
use of mechanical circulatory support systems, focusing
on implantable circulatory assist applications are in
progress by several groups. Rotary blood pumps are
attracting more and more interest by researchers as a next
generation implantable circulatory device. In the early
stages of development, arguments against their long-term
usage were wide spread due to the non-pulsatile output
flow and high risk of hemolysis. However, through
continuous research, rotar y pumps have proven to
overcome these drawbacks and have the advantages of
compact size, less hemolysis, less thrombosis, low power
consumption, easy control and low cost compared to the
pulsatile blood pump. A series of studies verifying the
suitability of nonpulsatile blood pumps have also been
performed by Jett et al (1).
has now become popular and widely used in clinical
application. This is because, in addition to the advantages
of the rotary pumps, they can operagte at an even lower
rotational speed than the axial pump. However, conventional
centrifugal blood pumps still have a comparatively high
hemolysis level not suitable for long-term use due to heat
and friction generated from the mechanical bearings or
pivots. This detriment to the blood cells limits centrifugal
blood pumps to short-term use only. Recent studies show
that active magnetic bearings can reduce the hemolysis
level in rotary blood pumps by eliminating this heat and
friction (2-5). This provides a significant improvement
towards long-term usage of centrifugal blood pumps.
We have developed a prototype magnetically suspended
centrifugal blood pump as a cardiopulmonary bypass
device or a ventricular assist device. Since the ultimate
goal is to make a small implantable device, we set our
sights on simplifying the active magnetic bearing system. In
traditional active magnetic bearing systems, shown in

©
by Wichtig Editore, 2000 0391-3988/000-09 $04.50/0
Centrifugal blood pump with a magnetically suspended impeller to reduce hemolysis level

a) b)

Fig. 1 - (a) Axis identification of the traditional 5 axes controlled magnetic bearing. The 5 axes of x1-x2, x3-x4, y1-y2, y3-y4, z1-z2 must have
their own dedicated sensors and actuators that make the total system bulky. (b) In our design we use the permanent magnet bearing to
regulate the axes x5-x6 and y5-y6. This strucutre has the advantage over (a) since only 3 axes (z, Θx, Θy) require the active magnetic bearing,
making the system small enough to implant into a human body.

Figure 1-(a), with five actively controlled axes the pump
structure is too bulky because each control axis must have
its own devoted sensors, electromagnets, amplifiers, and
feedback controllers. Seeking a compromise between
simple structure and precise positioning, we were able to
realize a simplified magnetic bearing system as illustrated
in Figure 1-(b), applicable to areas like implantable
centrifugal blood pumps, where the size of the system is
critical. With this simplified magnetic bearing system, which
consists of an active magnetic bearing at 3 axes of motion
and permanent magnetic bearing at 3 axes of motion, the
impeller could hover at a fixed position with a variation of
less than 10 µm. Since a low rotating speed is required to
reduce hemolysis, a pump showing high efficiency in
output flow and pressure head should be another design
target. We adopted the impeller structure from our former
studies to develop the prototype centrifugal pump (6,7). A
digital signal processor based controller was installed to
implement a linear quadratic control algorithm to suspend
the impeller. The digital controller enables sophisticated
modern control algorithms with flexibility in upgrading.
METHODS
Mechanical Structure
The schematic diagram of the pump head with the
magnetic bearing system is shown in Figure 2. It consists
of an active magnetic bearing system, a magnetically
suspended impeller and a magnetic coupling with a
brushless direct current motor. The implemented motor
(FBLM5120A-A, Oriental Motors, Tokyo, Japan) has a
maximum rotating speed and rated torque of 3000 rpm
and 0.4 N.m, respectively. On the magnetic bearing side,
the three proximity sensors and three in house fabricated
electromagnets with the geometry shown in Figure 3 face
the impeller.
A ferrite stainless steel plate on the impeller, face to
face with the magnetic bearing head, becomes the target
for the magnetic bear ing system. This magnetic
suspension cancels the attracting force of the magnetic
coupling and regulates the impeller position. Twenty four
neodymium-iron-boron (NdFeB) permanent magnets are
attached, one set on the motor shaft and the other set on
the impeller. The polarity of the driving permanent
magnets must be alternated in the manner fo N-S-N-S... to
transfer the rotating force from the motor to the impeller.
This magnetic coupling between the impeller and the
motor shaft also acts as a passive radial bearing,
regulating 3 degrees of freedom of the impeller.
The intended design point was a pressure head and
flow rate of 300mmHg and 5L/min, respectively at a
motor speed of 3000 rpm. With this target operating
point, calculation of the blade angles and volute specs
was done based on former design studies (6,7). The
weight of the pump head excluding the motor was 380 g
and the dimensions of the pump housing were 125 mm

2
H. Kim and H.C. Kim

Fig. 2 - Schematic diagram of

the prototype magnetically
suspended centrifugal blood
pump.

and 41.5 mm in diameter and height, respectively. The

inlet and outlet nipple dimensions were designed to fit
the standard 3/8” connection tubes used in cardiac
surger y. A data acquisition board (EZAD-512, Dail
Information, Seoul, Korea) in the pump controller was
connected to a laptop computer via an RS-232
communication link. A monitoring program coded with
Labview (National Instruments, Austin, TX) on the laptop
displays the displacement, power consumption and
additional data from the pump system. A mock circulation
setup in Figure 4 was established to evaluate the pump
performances.

Fig. 3 - Geometry of the 3 sensors and 3 electromagnets of the

Magnetic Bearing System Implementation magnetic bearing system.

The active magnetic bearing system consists of three

proximity sensors, a digital signal processor, power
amplifiers and three electromagnets. Eddy current type
proximity sensors (E2CA-X1R5A, Omron, Tokyo, Japan) and
signal amplifiers (E2CA-AL4C, Omron, Tokyo, Japan) were
used to detect the distance from the pump housing to the
impeller. An additional sensor signal conditioning circuit was
designed to reduce the high frequency noise and adjust the
gain and offset, providing the signal with a reliable resolution
and a desirable dynamic range for data acquisition. This
amplification circuit also acts as an anti-aliasing filter before
the analog-to-digital conversion. The overall resolution,
sensitivity and cut off frequency were 1.8µm, 20mV/µm and
3kHz, respectively. For high-speed sampling and
calculations, a digital signal processor (TMS320C31, Texas Fig. 4 - Mock circulation test setup of the developed pump system.

3
Centrifugal blood pump with a magnetically suspended impeller to reduce hemolysis level
Instruments, Dallas, TX) based control unit was installed (8).
The digital signal processor reads the displacement from the
proximity sensors through a 12 bit resolution analog-to-
digital converter (MAX122, Maxim, Sunnyvale, CA) at a
sampling rate of 10 kHz. The digitized displacement data
are lowpass filtered by an 80-tap finite impulse response
(FIR) digital filter. For maximum calculation speed, all
filtering and linear quadratic control algorithms were coded
with assembly language using application specific
instructions such as MPYF,ADDF or STF,STF. Using these
parallel instructions we could i) multiply and add or ii) store 2
floating point operands in a single cycle, taking full
advantage of the digital signal processor’s micro-parallelism
(9). The electromagnet coil current calculated by the digital
control algorithm is sent to the current amplifiers via a 12-bit
resolution digital-to-analog converter (AD664, Analog
Device, Norwood, MA). A current amplifier was designed
using a power operational amplifier (OPA548, Burr-Brown,
Tucson, AZ) to convert the voltage from the digital-to-analog
conver ter to a corresponding driving current. Each
electromagnet was organized with 287 turns of enameled
copper wire around a ferrite El lamination core.
System Modeling and Controller Design
To levitate the impeller with the pump structure as
shown in Figure 5, a centralized controller would be
desirable for sufficient stability and robustness (10).
Description of the pump system verifying the system
parameters must be made before designing this optimal
4
Fig. 5 - Block diagram of the
magnetic bearing plant and
controller.
digital controller.
Assuming the impeller as a rigid body, the total
magnetic bearing system requires that all six degrees of
freedom (x, y, z, Θx, Θy, Θz) in Figure 6 to be controlled.
Since the motor, which is magnetically coupled, controls
the rotation Θz and this permanent magnetic bearing also
regulates the radial coordinates x, y, the active magnetic
bearing system’s duty is to control the remaining three
degrees of freedom z, Θx, Θy. Forces acting on the impeller
besides these three axes (i.e. shear stress) have not been
included for simplicity.
As a first step in deriving the equation of motion, we
define the vector notations of the generalized coordiantes
in Figure 6 as:
[1]
A preliminary equation of
[2]
with
m: impeller mass 6.26x10–2 kg
I: impeller inertia 1.6172x10–5 kg·m2
Z1, Z2, Z3: generalized force [N]
H. Kim and H.C. Kim
can be constructed applying the Lagrange equation,
ignoring the gyroscopic effects. The generalized forces,
which are the vector sum of the acting forces (including
the electromagnetic forces and load forces), can be
expressed as:
[3]
with
I: distance from the center to the electromagnets in Figure 6
1.9x10–2 m
fa, fb, fc: force applied to the impeller at the 3 electromagnet
position in Figure 6 [N].
Based on the literature on handling electromagnetics or
magnetic bearings, force activating within a magnetic
circuit can be expressed as a linear function of the
clearances za, zb, zc between the impeller and the housing,
Fig. 6 - Cross-sectional view of
the magnetic bearing facing
the impeller. Coordinates are
defined for system modeling.
and coil currents ia, ib, ic (10).
[4]
Here, ks, ki are the force-displacement factor and force-
current factor, respectively. Force, displacement and
current are all regarded as incremental quantities. The
dimension of k s and k i were 1600 N/m and 21.6 N/A
respectively, and were measured by a separate setup. In
real control applications, it is convenient to express the
equations using the sensor coordinates. After transforming
[4] into the sensor coordinates from the geometrical
constraints, we derive the following ultimate equation of
motion from which we can design an optimal controller.
[5]
where vectors, matrices and system parameters are
defined as:
5
Centrifugal blood pump with a magnetically suspended impeller to reduce hemolysis level

Transformation of this formula into the digital domain

with a sampling period of 100 µs was made since we are
designing a digital controller. Finally, using the difference
of the displacements instead of the actual derivative, a
simple centralized linear quadratic optimal controller, as
in Figure 7, was designed (12). The optimal state
feedback gain K minimizing the cost function J = *`0 (u’
Ru + x B + x B Q xB ) dt was found using Matlab
— — —
(Mathwor ks, Natick, MA) from the der ived state
equations in [5]. u is the system input vector and xB is
—
the state vector. The respective weights R and Q were
fixed as unit matrices.

[6]
RESULTS

Controller Performance

The performance of the optima linear quadratic (LQ)

p: distance from the center to the sensor in Figure 6
2.3x10–2 m
ZBa, ZBd, ZBc: clearance between the impeller and housing
at the 3 sensor positions in Figure 6. [m]
controller was ver ified and compared with the
propor tional der ivative (PD) controller. The step
response without motor rotoation and the steady state
response at a motor speed of 3000 rpm are shown in
Figure 8. The rising time (<5 ms) and overshoot (<10
µm) during the step response were similar in both the LQ
and PD cases. But at high motor speed, the LQ controller
showed better perfor mance in less peak-to-peak
fluctuation (45 µm) than the PD controller (60 µm). Since
the centralized LQ controller also considers the coupling
gains between the channels, it would be more robust
under varying conditions.
Fig. 7 - Multiple input multiple
output linear quadratic
controller for the magnetic
bearing system. r, x B , u are
the reference input —vector,
feedback input vector and state
vector of the system, respecti-
vely.

6
H. Kim and H.C. Kim

Fig. 8 - The step response and

steady state response of the
linear quadratic (LQ) and
proportional derivative (PD)
controller are compared. The
LQ controller shows less
fluctuation in steady state
response at a motor speed of
3000 rpm.

Output Characteristics and Power Consumption

T h e f l ow ra t e a n d d i f fe r e n t i a l p r e s s u r e o f t h e
prototype pump with the magnetic suspension showed
higher hydraulic parameters than our mechanical
bearing pump developed in 1998 (7). The principal
reason for this improvement is that the clearance
between the impeller and the pump housing in the
magnetic bearing version is less than in the mechanical
b e a r i n g ve r s i o n . T h u s, t h e p r e c i s e p o s i t i o n i n g
characteristics of the magnetic bearing system can also
p r ov i d e b e t t e r p u m p i n g e f f i c i e n c y. Fo r d i f fe r e n t Fig. 9 - Pressure and flow curves for different rotating speeds.

Fig. 10 - Pressure and flow curves for different clearances between the Fig. 11 - Current consumption of the magnetic bearing system for dif-
impeller and the pump housing at the active magnetic bearing side. ferent clearances. The supply voltage for the current amplifier was 12 V.

7
Centrifugal blood pump with a magnetically suspended impeller to reduce hemolysis level
rotational speeds, the flow-pressure curves are shown
i n F i g u r e 9 . T h e d e s i ra bl e o p e ra t i o n p o i n t fo r a
cardiopulmonary bypass device (5L/min, 300mmHg)
was achieved at a motor speed of 2750 rpm. Flow and
pressure relations in Figure 10 show little deviation as
the nominal clearance was changed from 200 µm to
100 µm. Since this decline is negligible, there is no
need to consider the lowering of pumping efficiency due
to the change of nominal clearance in determining the
optimal clearance.
The current consumption of the magnetic bearing
system almost halved as the nominal clearance changed
from 200 µm to 100 µm (Fig. 11). As the rotational speed
increased, the input current slightly decreased due to
the buoyancy. It is advantageous to operate the pump at
a smaller gap to consume less power. But if this
clearance is reduced to a value comparable to the size
of the blood cells, the hemolysis level will increase (7).
Conclusively, the optimal gap must be the smallest
possible one that wont’t result in higher hemolysis. The
developed pump showed a power consumption of 5W at
a clearance of 100 µm and rotational speed of 2700
rpm.
DISCUSSION
We h ave d e s i g ned a prot ot ype magne ti ca l l y
suspended centr ifugal blood pump showing high
differential pressure and flow rate at a comparatively low
rotation speed. The magnetic bearing system with only 3
actively controlled axes could provide sufficient stability
and robustness during in vitro tests. A digital LQ control
algorithm was implemented in a digital signal processor
to regulate the position of the magnetically suspended
impeller. Less fluctuation was observed than with the PD
controller, when the impeller was rotating at a high
speed. Laboratory tests to determine the optimal gap
between the impeller and the pump housing were
performed considering power consumption, pressure
and output flow. Since the decline of pressure and flow
rate was negligible as the clearance between the
impeller and pump housing varied, the least clearance
not causing a rise in hemolysis level should become the
operating clearance for maximum power efficiency.
Further in vitro and in vivo hemolysis test results should
be considered.
ACKNOWLEDGEMENTS
The authors wish to acknowledge the financial support given
by the Korea Research Foundation in the program year of 1998.
Reprint requests to:
Hee Chan Kim, Ph.D.
Department of Biomedical Engineering
College of Medicine, Seoul National University
28 Yongun-Dong, Jongno-Gu
Seoul, 110-744, Korea
e-mail: hckim@snu.ac.kr
REFERENCES
01. Jett GK. Physiology of nonpulsatile circulation: acute versus
chronic support. ASAIO J 1999; 45: 119-22.
02. Nojiri C, Kijima TR, Maekawa J, et al. More than 1 year
continuous operation of a centr ifugal pump with a
magnetically suspended impeller. ASAIO J 1997; 43: M548-
52.
03. Hilton EF, Allaire PE, Baloh MJ, et al. Magnetic suspension
controls for a new continuous flow ventricular assist device.
ASAIO J 1997; 43: M598-603.
04. Akamatsu T, Nakazeki T, Itoh H. Centrifugal blood pump with
a magnetically suspended impeller. Artif Organs 1992; 16:
305-8.
05. Malson EH, Bearnson GB, Allaire PE, et al. Feedback control
applicaitons in artificial hearts. IEEE Control Systems
December 1998; 26-34.
06. Kim WG, Chung CH, Yang WS, et al. Development of a cen-
trifugal pump with thck blodes. Artif Organs 2000: 24: 160-4..
07. Yang WS. Design methods. In: Design and performance test
of centrifugal blood pump. MS Dissertation: Seoul National
University 1998; 4-13.
08. Bleuler H, Guler C, Herzog R, et al. Application of digital
signal processors for industrial magnetic bearings. IEEE
Trans Control Systems Technology 1994; 2: 280-9.
09. Parallel-Operations Instructions. In: TMS320C3X User’s
Guide. Dallas: Texas Instruments, 1997; 13.6-13.8.
10. Schweitzer R, Bleuler H, Traxler A. Magnetic force. In: Active
magnetic bearingo. Zürich: vdf Hochschulverlag AG, 1994; 63-7.
11. Phillips CL, Nagle HT. Discrete state equations. In: Digital
control system analysis and design. New Jersey: Prentice-
Hall, 1995; 156-9.
12. Anderson BDO, Moore JB. The standard regulator problem-1.
In: Optimal Control. New Jersey: Prentice-Hall, 1989; 7-34.