IEEE Circuits and Systems Magazine - Q3 2021 - 75

Dm VS dt
t0
[]
3
t1
==
## (1)
AirPipePipe
t0
))T
t
t1
2 P Sdt
Actuators are Sanyo Denki 103H5208-5240 3D printer
stepper motors driven by a Allegro A4988 motor
driver. A serial link has been implemented in order
to monitor all parameters on a standard PC, without
adding a difficult to find or expensive LCD directly to
the ventilator.
V. User Interface
User graphic interfaces for supervision and set-up
are developed in C#, thus it can be run on a standard
PC connected to the ventilator through a RS232
link. Sensors data is monitored in real time (intralung
pressure and tidal volume). This interface allows
the user to configure the following ventilator
medical parameters:
■ Inspired volume: can be set-up between 0.2 and
15 . L.
■ Maximum intra-lung pressure (in millibars).
■ Breath cycles per minute.
Interface also allows to set ventilator active mode among
the following two ones:
■ Reanimation: breath is initiated by the ventilator
(usually used when patient is unconscious)
■ Assistance: breath is initiated by the patient (patient
is conscious and needs only an assistance)
It is important to note that computer interface is only
used for monitoring and set-up. Artificial ventilator is
fully controlled by its microcontroller, allowing an autonomous
and reliable behaviour, even if computer link
is lost for any reason. The full interface program is open
source and published in [9].
VI. Hardware Failure Automatic Detection
As low-cost recycled materials quality is unknown, a
mechanical breakdown may occur. Therefore, in order
to improve robustness, a solution based on cooperation
between embedded artificial intelligence (AI) and medical
staff is proposed.
Pitot Tube Sensor
Holes
Air Flow
Embedded AI algorithm is based on characterizing
interactions between the patient and the ventilator.
This is done by extracting features from sensors
signals, as described in Section VI-A, during breathing
cycles, in order to later detect hardware failures
or anomalies. It uses in situ unsupervised learning:
every patient is different, so that it is not possible to
learn system behaviour once and for all. Consequently,
training process will be done every time the ventilator
is used, under supervision of medical staff for
a short time (about one minute), to check its proper
functioning during that period.
Life of the patient cannot depend on the serial link
between embedded microcontroller and computer interface.
Thus, failure detection is fully implemented
on the embedded microcontroller, so that a problem
with RS232 communication will not affect ventilator
function, nor anomalies detection. Computer user interface
software is used only for supervision and control.
Hardware failure automatic detection system is
presented in Fig. 5. Each block is described in the following
subsections:
A. Features Extraction
First step of the proposed algorithm is to extract relevant
features from sensors signals on each breath cycle as
Figure 4. Volume sensor implemented using home-made Pitot
tube assembled on the Omron D6F-PH0505AD3 sensor.
Embedded Signal Processing
Volume
Features
Extraction
Pressure
K-Means
Classification
Failures
Detection
Figure 5. Hardware failure automatic detection system.
THIRD QUARTER 2021
IEEE CIRCUITS AND SYSTEMS MAGAZINE
75
IEEE Circuits and Systems Magazine - Q3 2021

Table of Contents for the Digital Edition of IEEE Circuits and Systems Magazine - Q3 2021
