Instrumentation & Measurement Magazine 24-4 - 85

Table 1 - MF levels boundaries
MF1
MF2
MF3
MF4
MF5
< PSD t ≤ 0.8
ratio() 
0.6 < PSD t
0.8 < PSD t
1.0 < PSD t
1.2 < PSD t
PSD t 
ratio
()
1
t
ratio ≤ 0.8  PSD t
PSD tratio i
ratio ≤ 1.0  PSD t
PSD tratio i
ratio ≤ 1.2  PSD t
PSD tratio i
ratio()  N  PSD t
ti  0 ratio() ()
1
1
1
1
1
t
N  PSD t
PSD t
t
ratio
ti  0
()
t
PSD tratio i
t
ti  0
ratio i
ti  0 ratio() ()
t
ti  0 ratio() ()
t
()
ratio i
()
(2)
MF t MF t MF t MF t exp

3 45()
()
()
The next step is to define how variations on the normalized
PSD ratio can be related to changes in the MF state. We start by
considering a baseline state (MF3
MF2) and two higher (MF4
) and also two lower (MF1
and MF5
and
) MF states. These five states
define boundaries for the evaluation of the normalized PSD ratio.
The value for these boundaries is shown in Table 1.
The main criteria we consider for evaluating the degradation
in the MF state is how long the operator spends in each
one of the different MF levels. Looking at isolated time stamps
can give the false impression that a high MF state was reached
while that could only be a natural EEG oscillation or a signal
spike due to some sources of noise. That is one of the reasons
for MF researchers to look into the tendency of variation in the
EEG data instead of focusing on absolute values [14]. As the
operator fatigues there is an increase in the percentage of time
in the MF4
and MF5 states. Based on the proportion of time in
each state, we can derive an MF score function, which can be
used to track the tendency of progression of MF:
5
MF k MF

score   MF
i 1
i
The MF multiplier kMF
i
(3)
i can be selected to make the MF detection
algorithm more or less compliant. Using higher values
for the higher MF states multipliers leads to higher MF score
values faster and the system tends to assess higher MF degradation.
We always consider kMF
values for kMF and kMF
1
2 and for kMF
3 = 1, since it is the baseline. The
4 and kMF
5 should be lower
and bigger than 1, respectively. The selected values are shown
in Table 2. More importance is given to the higher MF, since we
want the assessment system to detect with a good safety margin
any possible dangerous operational situations.
Before MF multipliers, a damping function is needed to
minimize the effect of sporadic spikes in the beginning of
the assessment period. The occurrence of these oscillations
when we have a short data series can lead to situations where
MF4
and MF5
states are overestimated, leading to the wrong
kMF1
0.80
June 2021
Table 2 - Selected MF multipliers
kMF2
0.90
kMF3
1.00
kMF4
1.15
Experimental Procedure
In order to assess the performance of the proposed algorithm,
we performed experiments using a vessel simulator, Fig. 3.
The simulator facility is a realistic full mission bridge, which
provides pilots in training with an accurate reproduction of
a real vessel bridge and experienced pilots with familiarity to
real vessels.
Participants were monitored using the wireless EEG headset
described earlier. We performed experiments in which we
varied the performed task and the participant's role in it as
well as the participants. During the experiments we made use
of the Karoslinska Sleepiness Scale (KSS) [15] as an easy way
for the participants to report their tiredness state. The questionnaire
is composed of nine affirmations about how sleepy
the participant is feeling. We applied the KSS questionnaire
immediately before the participant started and immediately
after he finished the experiment. The KSS is already validated
against EEG data, and the information obtained with the questionnaire
was be later cross-referenced with the results of our
MF assessment algorithm.
We tried to ensure a good baseline condition between participants
by performing all experiments during the morning
and asking participants to avoid the consumption of stimulants.
However, these measures by themselves are not enough
to ensure all participants start the experiment in the same MF
N  PSD t
()  Nt
()  Nt
()  Nt
PSD t
t
ratio
()
ti  0 ratio() ()
t
assessment of a dangerous MF situation. We chose to use an
exponential damping function that converts a time-dependent
portion of early MF4
and MF5
assessment to MF3
MF t MF t
() 1


D()   exp

5t
T


D()   exp
MF t MF t( ) 1 55 
5t
T
3 ()  T
D
is 0.67% and can be considered neglectable.

 

5t
state, which
can be described by the following set of equations:
44
(4)
(5)
(6)
where T is the desired time stamp where the damping of MF4
and MF5
kMF5
1.30
Fig. 3. Simulator facilities used in the study.
IEEE Instrumentation & Measurement Magazine
85
Instrumentation & Measurement Magazine 24-4

Table of Contents for the Digital Edition of Instrumentation & Measurement Magazine 24-4
