1208 
MONITORING 
being controlled. A control system that is capa- 
ble of implementing such changes is said to be 
adaptive. 
THE ADAPTIVE CONTROL 
OF THERAPEUTIC PROCEDURES 
We chose to initially study the adaptive con- 
trol of intravenously administered vasopres- 
sors. The decision to start v^^ith the control of 
blood pressure v^as based more upon engineer- 
ing than upon medical expedience. Blood pres- 
sure can be measured accurately and reliably, 
and it can be altered by a number of rather spe- 
cific drugs. For these reasons, by choosing to 
control blood pressure, one could get on v^^ith 
the central control problem and not spend an in- 
ordinate time struggling with transduction and 
pharmacological questions. 
Before proceeding, let us further define adap- 
tive control. With simple control one attempts 
to maintain a given output by comparing some 
function of the output with a desired value any 
difference being the error. Figure 1. This error 
is used at the system's input to correct the out- 
put. It can be shown that such a system will be 
stable under a prescribed set of circumstances. 
It is potentially unstable if the plant delay or 
the open loop gain [(plant gain) X (feedback 
processing gain)] exceed prescribed limits. If 
these limits are known and fixed, they can be 
accounted for in the controller's design, but if 
not, some other allowances must be made. 
By making the feedback processor variable, 
Figure 2, and allowing it to observe the dynam- 
ics of the plant, one can hold the open loop char- 
acteristics within prescribed bounds. Such a 
system is said to be adaptive. Adaptive systems 
come in many forms, but they all have two 
things in common. One, somewhere within the 
controller there is a means of identifying and 
DESIRED^ 
OUTPUT >- 
SET 
ERROF^ 
PLANT 
(SYSTEM BEING 
CONTROLLED) 
FEED BACK 
PROCESSOR 
-OUTPUT 
DESIRED 
OUTPUT >- 
SET 
PLANT 
■7^ 
FEED-BACK 
PROCESSOR 
■ OUTPUT 
Figure 2. — The system in Figure 1 with provision for 
adapting the feed back loop. 
measuring the pertinent parameters in the 
plant, and two, there is a means for affecting 
the system's open loop response. 
At first a heuristic approach was used in an 
attempt to realize the system in Figure 2. This 
worked well for gain adaption, but due to natu- 
ral fiuctuations in blood pressure the end point 
of a response was hard to locate thus making 
delay adaption difficult. Correlation techniques 
could be used, but a better approach was at 
hand. It not only would allow for gain and delay 
adaption but also for modeling of the system's 
total dynamics. 
In this approach an "Adaptive Model" identi- 
fies the parameters of interest, and a "Fore- 
ward Time Calculation" affects the system's re- 
sponse. Figure 3. 
The Modeler observes the I.V. Control- 
ler/plant response and develops a mathe- 
matical model of these two components. The 
model weights are used by the "Foreward Time 
Calculator" to determine what Xj should be to 
obtain rj. 
At first glance it would appear that this is not 
a control system. No error is apparently deter 
mined. As we will see, internally, the model's 
output is compared with the plant's output. The 
difference between these, that is the error, is 
I 
I 
i 
DRUG 
FLOW- 
RATE/ 
BLOOD 
PRESSURE 
REFERENCE 
INPUT 
BLOOD 
PRESSURE 
COMMAND 
SIGNAL 
FORWARD 
TIME 
CALCULATION 
"j 
ZERO- 
ORDER 
HOLD 
PLANT 
r 
9 SAMPLER 
ADAPTIVE 
MODEL 
Figure 1. — A simpler linear control system. 
Figure 3. — A block diagram of the system now in use. 
