Transcript Investigation in LVAD & CVS interaction & A Non

20
10
Investigation in LVAD&CVS interaction and
A Non-Invasive control approach
Author – Huaiyu Scott Lin
Supervisor : Andrey Savkin / Co-Supervisor: Nigel Lovell / Assessor: Victor Solo
Introduction
Existing designs of Left Ventricular Assist Device have their
limitations in responding to recipients’ changing physiological
status. The aim of this thesis is to investigate the interaction
between the Cardiovascular system and its assisting device,
then take a non-invasive approach to control the device thus
avoid the extreme scenarios in terms of assistance failure
which further damages the native heart. While the invasive
methods are commonly used in current researches, it is
essential to seek for a non-invasive approach that minimise
the risks of thrombus formation caused by implementing
physiological sensors and transducers. In additional, a control
strategy which is able to track for a changing optimal
operating point is the key in designing future fully adaptive
control algorithms.
Part2: Non-invasive control approach
1. Optimal operating point searching using
Extremum Seeking Algorithm for suction
avoidance
Part1: Biological Signal Analysis
1. Measurement of
system Preload
Pulsatile
waveforms
present throughout the
blood circulation of body
due to the contractile
force of cyclic heart
beats. Such pulsatile
signal indicates the status
of left ventricle function
as
it
responds
to
physiological
blood
demand. It can be
obtained as:
PI = LPF(abs(HPF(X)))
PI(w) Vs Pump Speed
mean Pump speed (rpm)
2800
2600
2400
2200
2000
Low Afterload
Medium Afterload
Medium High Afterload
High Afterload
1800
1600
0
20
40
60
80
PI(w)
100
120
140
160
PI(w) Vs Mean Pump Flow
7
6.5
Mean Pump Flow(L/min)
6
5.5
5
4.5
4
3.5
Low Afterload
Medium Afterload
Medium High Afterload
High Afterload
3
2.5
2
0
20
40
60
80
PI(w)
100
120
140
160
Maintaining a desire pump flow by
adjusting a right amount of pump
speed is critical in order to achieve
a target assistance. Experimental
results showed that an increase in
speed is required for maintaining a
target flow for each increment of
systemic afterload as shown on
the left. This is true for high flow
values and it provides us with an
approximated linear relationship
between PI(speed) and flow within
the drawn box. Clinically, this is
the expected working range for an
appropriately controlled LVAD
since PI(w) can not be less than a
certain level which induces
suction, and non-linearity beyond
the right boundary of the box.
3. Optimal operating point
UNSW
Result on the left shows a
convergence of output
value as input is stable.
Output
2. Feedback approach to compensate
system afterload
2. Non-linearity between pump speed
and flow for varying afterload
3000
Input
ESC tracks the optimal
point of PI by watching its
cost function, which is
defined as gradient of PI
w.r.t speed
GPI = dPI/dw
A
small
sinusoidal
perturbation is added in
order to track the slope of
changing convex function
of GPI.
A simulation work has
been carried out by
assigning a second order
polynomial as plant.
An optimal operating point
for full assist can be defined
as the minimum gradient of
PI w.r.t speed thus gives the
maximum
possible
assistance with leaving
enough safe margin before
suction.
Due to the major system linearity between a mean
pump flow and the mean rotational speed is caused
by changing systemic peripheral resistance
(afterload) once a high pump flow is required. We
required a speed compensation in order to keep up
with the desired flow hence follow the starling law.
Result
is able to track the
reference PI value with inversely
change of the input rotational
speed. The right top figure
shows the tracking process for
reaching the desired optimal
value obtained from ESC. The
bottom
figure
shows
the
changing speed with respect to
that.
PI
speed
ENGINEERING @ UNSW