Quantifying the Impact of Secondary Duties on Sailor Workload
Through Simulation
Victor Isaac
Centre for Operational Research and Analysis, Defence Research and Development Canada, Nepean, Canada
Keywords: Monte Carlo Simulation, Workload Modelling.
Abstract: Designing crewing concepts for ships requires complete information regarding the tasks that sailors must
perform, since an incomplete understanding could result in unreasonable crew workloads and fatigue. It is
therefore important to look at all sources of sailor workload. Primary duties, which belong to a billeted
position, have been studied extensively in the past and are well-represented in existing crew models.
Secondary duties, which are tasks assigned to individual sailors in addition to their primary duties, are not as
well understood. This creates a significant risk in crew design. In this study, a simulation model has been
developed to quantify the impact of secondary duties on sailor workload. The stochastic nature of the model
makes it suitable for use in Monte Carlo experiments and allows it to explore the impact of combining one or
more secondary duties with a sailor’s primary duties. This allows it to be used to create statistics suitable for
applications, such as predicting fatigue rates, where extreme values are important. This paper describes the
development, verification, and implementation of the model.
1 INTRODUCTION
1.1 Background
The state of the personnel aboard a vessel determines
how effectively and safely the vessel can be operated.
With this in mind, workload and fatigue are important
areas of study for maritime organizations. While lots
of effort goes into understanding the workload of
each billeted position on a ship, the work that sailors
do that does not belong to their assigned position can
be overlooked. This creates a risk in crew design in
that sailors may end up with far more workload and
far less rest than designers anticipated. While this can
be identified and corrected after the crew design is
implemented, improving the original designs with
more complete data would reduce the strain on first-
of-class crews and reduce the work necessary to
modify and validate crew designs after they are put
into practice.
This paper describes a simulation model that was
created to demonstrate the impact of secondary duties
on Royal Canadian Navy sailor workload.
1.2 Definition of Sailor Duties
For the purpose of this study, the following
definitions were developed:
Primary Duties: tasks listed in terms of
reference or a job description associated
with an assigned billeted position.
Secondary Duties: additional tasks
assigned to an individual by their chain of
command or nomination through voluntary
means that contribute to the good order and
functionality of a unit or crew.
Tertiary Duties: tasks assigned to
personnel on a watch and station bill that
may vary with each iteration of the watch
and station bill.
Note that because the definitions of primary,
secondary, and tertiary duties depend only on how
duties are assigned, the classifications may differ
from organization to organization. For example, a
task that is a primary duty in a navy that includes it in
the job description of a billeted position would be a
secondary duty in a navy where it is assigned to
individual sailors regardless of their position.
Isaac, V.
Quantifying the Impact of Secondary Duties on Sailor Workload Through Simulation.
DOI: 10.5220/0011923300003396
In Proceedings of the 12th International Conference on Operations Research and Enterprise Systems (ICORES 2023), pages 259-268
ISBN: 978-989-758-627-9; ISSN: 2184-4372
Copyright
c
2023 by His Majesty the King in Right of Canada as represented by the Minister of National Defence and SCITEPRESS Science and Technology Publications, Lda. Under
CC license (CC BY-NC-ND 4.0)
259
1.3 Literature Review
The United States Navy (USN) makes use of a Navy
Availability Factor (NAF) to describe the time each
sailor has available to complete tasks assigned to
them (Chief of Naval Operations, 2021). These hours
are broken down into productive and non-productive
time, with the non-productive time further divided
into training and service diversion. Service diversion
consists of actions required by personnel through
regulation or routine and includes things such as
inspections and participating in committees (Chief of
Naval Operations, 2021).
The USN’s Navy Total Force Manpower Policies
and Procedures provides breakdowns of the NAF of
personnel in various states, but these hours are not
broken down to a sufficient granularity to be able to
attribute hours to specific secondary duties or even
secondary duties in general (Chief of Naval
Operations, 2021). This issue is also seen in studies
that use the same definitions as the policy documents,
such as the study of sailor workload that was
performed by Garbacz (2019) and highlighted in
Cordle (2019). Many of what are considered
secondary duties in the modelling described in this
paper would be captured as own-unit support in
Garbacz (2019); however, values are only provided
for the productive workload of sailors, a category that
includes own-unit support along with primary duties
such as standing watch. The result is a lack of
available data describing the time required to
complete secondary duties.
This lack of data means tools that model crew
usage rates are unable to explicitly include secondary
duties. Secondary duties cannot be included as discrete
tasks in task networks when using tools such as the
Improved Performance Research Integration Tool
(IMPRINT) as in Hollins and Leszczynski (2014). This
means that any workload related to the completion of
secondary duties would be unaccounted for in such
studies and the number of crew required to operate a
vessel underestimated. Similarly, the accuracy of a tool
developed by Defence Research and Development
Canada, the Simulation for Crew Optimization and
Risk Evaluation (SCORE), is reduced by the lack of
secondary duty data. The tool combines a sailor’s
regular duties with the roles they are assigned during
scenarios such as replenishment at sea to output a usage
rate for the sailor and a list of instances of conflict
where the sailor has concurrent taskings (Chow et al.,
2016). Since the model cannot include the impact of
secondary duties, the usage rates and number of
conflicts predicted by the model are underestimates for
any sailor assigned a secondary duty.
1.4 Goal of the Modelling
The goal of the simulation model described in this
paper is to generate daily work schedules that are
suitable for computing statistics describing the impact
that secondary duties have on sailor workloads. Using
Monte Carlo simulation, statistics can be computed
from daily schedules that allow for a more
comprehensive understanding of sailor workload than
broad measures such as the number of hours a sailor
must commit to various tasks per week, as is done in
the USN’s NAF (Chief of Naval Operations, 2021).
For example, suppose a secondary duty is
understood to require approximately 100 hours to
complete over the course of a year. A crew design
model could take this secondary duty into account by
reducing the availability of a sailor by two hours each
week, similarly to what is done in the NAF using the
non-productive time category (Chief of Naval
Operations, 2021). However, if the required hours are
not evenly distributed across the weeks of a year and
instead happen in a limited number of time periods
that are unpredictable in nature, a sailor assigned the
secondary duty may find themselves with extreme
short-term workloads. A simulated schedule
generated by the model contains this information, and
so it can be used to predict how often the sailor will
face extreme workloads. In this way, the model can
identify conditions that will lead to sailor fatigue in a
crew design that otherwise may go undetected until
crew validation takes place.
The simulation model also improves upon simple
estimates of secondary duty time requirements by
layering multiple stochastic processes. This will
allow for analysis that includes days where a sailor’s
primary duty workload is above average and they
must work on more than one secondary duty. This
sort of randomized schedule can then be combined
with other excursions from typical workload
including taking part in a special evolution such as a
resupply at sea or collective training to model the true
schedule of a sailor and identify unsustainable
combinations of primary and secondary duties.
2 MODEL ASSUMPTIONS
2.1 Workload
In most work settings, the expected workload of a
position is the sum of the time taken by each task
assigned to the position and any additional tasks taken
on by the individual worker filling the position. For
office workers, this is often as simple as 7.5 hours per
WAPAT 2023 - Special Session on Workforce Analytics - Practical Application and Theory
260
day of assigned work plus any additional time spent
taking part in committees, labour organizations, or
other voluntary activities. Sailor workload is similar,
in that sailors are responsible for their primary and
secondary duties, but there are several complicating
factors that present themselves when the workplace is
a vessel. This is especially true when the vessel is at
sea.
First, many sailors stand watches. Depending on
the nature of the watch, it may or may not be possible
for a sailor to work on other tasks while standing
watch. In the USN, hours spent standing watch are
included in the productive time within the NAF with
the acknowledgement that additional productive
work will have to take place outside of the hours spent
standing watch (Chief of Naval Operations, 2021).
This builds in the assumption that none of the non-
productive items (training or service diversion), and
by analogy many secondary duties, can be completed
on watch. For the simulation model, it was decided to
consider all watch hours as primary duty hours and
require that secondary duties be completed when
sailors are off watch. Since some sailors can complete
secondary duties during a watch, this assumption will
result in the model results being the upper bound of
sailor workload.
A further complication in analysing the workload
of sailors is that the length of the workday, which
tasks are to be completed, and how long many tasks
take to complete depend on what the ship or unit is
doing. In the broadest sense, sailor workload depends
heavily on whether they are currently ashore or at sea.
At a more detailed level, each individual primary and
secondary duty will likely change depending on
whether it is being performed on shore or at sea, with
further considerations often being necessary to
include the types of shore duty and sails.
The USN addresses the disparity in sailor
workload based on unit state by generating multiple
NAFs: afloat, ashore (peacetime), and mobilization
(Chief of Naval Operations, 2021). Ashore is further
divided into whether the unit is stationed in the
continental US or internationally. A similar approach
was taken in this study after reviewing the secondary
duties to be considered.
In the simulation model, the state of a unit or ship
is divided into four categories: alongside home port,
alongside foreign port, at sea on routine sail, and at
sea on operation. These states were selected based on
the impact they are expected to have on secondary
duties. For example, a unit alongside a home port will
have extensive shore facilities to make use of,
reducing the need for many secondary duties that aim
to recreate these services at sea. A unit alongside a
foreign port will have some shore-based facilities to
make use of, but not as many as one alongside a
domestic port. Time at sea is divided into routine and
operational sails to take into account the difference in
crew composition and operational tempo in the two
states. The four states, and short labels used to refer
to them, are provided in Table 1.
Table 1: The four ship/unit states considered in the model.
Name Code
At sea on routine sail SR
At sea on operation SO
Alongside home port AH
Alongside foreign port AF
Combining all these considerations, a sailor’s
workload is described in the model as the time they
require to complete their primary and secondary
duties in each of the four unit states described above.
Watch standing, and therefore all associated tertiary
duties, are counted as primary duty workload in the
model.
2.2 Time Estimation
Due to a lack of historical logs or other means of
accurately tracking how sailors spend their time, the
workload modelling described here will rely on time
estimation done by sailors based on their experience.
Given the number of billeted positions, secondary
duties, and unit states to be considered for a typical
ship, many individual time estimates are required and
a quick means of expressing them is necessary. A
single-point estimate, such as the mean time spent on
a secondary duty each month, is simple but does not
describe the variance of the value or the uncertainty
in the estimate. The uncertainty in such estimates may
be large, especially if sailors are generating them
from memory. Two-point estimates, such as a
minimum and maximum, may describe the variance
and uncertainty of a time value but do not include an
estimate of the most-likely value. In this study, a
three-point estimation is used: all time estimates are
expressed as the minimum value, the most-likely
value, and the maximum value. These three values are
often used in studies that aim to predict how much
time a combination of tasks will take (Clark, 1962).
They are more intuitive than abstract values such as
the mean, variance, or standard deviation.
Multiple distributions can be built from three-
point estimation. Further knowledge of the
distribution of time values will be necessary to inform
a decision around the most suitable distribution for
this application. As of writing, the triangular
Quantifying the Impact of Secondary Duties on Sailor Workload Through Simulation
261
distribution is being used. Other three-variable
distributions, such as the Project Evaluation and
Review Techniques (PERT) distribution (Clark,
1962), or even two-variable distributions such as the
uniform distribution, will be considered if empirical
data supports it.
2.3 Treatment of Secondary Duties
With the understanding that secondary duties are
often made up of many distinct tasks, a simple way of
expressing them in the model is required, especially
since each secondary duty must be described four
times: once for each unit state. It was decided to
express instances of a secondary duty requiring
attention as a recurring event of variable length. The
rate of occurrence of instances of a secondary duty is
defined by a frequency, and the amount of time
required by each instance is described by a triangular
distribution defined by three-point time estimation. It
is the range of time requirements that is intended to
take the different tasks involved in a secondary duty
into account.
Some tasks associated with secondary duties
follow a set schedule or must be completed at a set
frequency. For example, a secondary duty may
include generating a monthly report. In its current
configuration, the model does not allow for rigid
scheduling of secondary duty events: the beginning of
secondary duty instances are computed stochastically.
This is to better capture tasks that must be completed
in response to unscheduled events or the fact that the
work required by a scheduled task may not itself be
scheduled. To return to the example of the generation
of a monthly report, a sailor may choose to complete
the report days before it is due if that is when they
have the required time available.
Each secondary duty instance is allowed to span a
number of days to allow for tasks that need not be
completed all at once. The number of days spanned
by a secondary duty instance is referred to as a
window, and a heuristic approach was devised to
allocate the required work hours within the window.
The heuristic approach for allocating hours will be
described in Section 3.4. Note that the length of the
window, like the other values that describe a
secondary duty, must be provided for all four unit
states.
Allowing separate occurrences of the same
secondary duty to overlap may or may not make sense
depending on the secondary duty being considered.
For tasks such as inspecting equipment, overlaps are
senseless; however, when a task is in response to a
need that can arise at random, such as in response to
a workplace accident, overlaps may occur. Since most
secondary duties contain multiple tasks, instances of
the secondary duty overlapping can also be treated as
different tasks overlapping and not a single task
overlapping. For this reason, the model allows for
overlaps in secondary duty instances. Upon review of
collected data, the code can be modified to disallow
overlaps for specific secondary duties if needed.
The model is flexible in the number of sailors and
secondary duties it can consider in a single run: the
scope of the simulation can range from a single sailor
to an entire unit. To allow this flexibility, the model
simulates each sailor’s schedule independently with
no knowledge of the schedule of other sailors.
For simulations with more than one unit position,
secondary duties can be assigned to multiple sailors
simultaneously. To account for the reduction in
workload required of a single sailor when a secondary
duty is shared, a number in the interval [0,1] is used
to represent the fraction of a secondary duty that each
sailor is responsible for. These fractional assignments
scale the three-point time estimates of the associated
duty for each sailor, but they do not affect the
frequency of occurrences. Note that the schedule of
each sailor is simulated independently, so while
sailors may share the total workload of a secondary
duty, instances of the secondary duty requiring
attention will not correspond in their schedules.
3 OVERVIEW OF THE
SIMULATION MODEL
3.1 Brief Description of the Model
The simulation model is a stochastic, discrete model
that utilizes a time step of one day as it simulates
sailor workload. It makes use of an idealized calendar
consisting of months of 30 days each, with each day
being assigned a day of the week. This allows
flexibility in including things like weekly routines or
leave into the model, if desired. The product of the
model is a schedule for each input unit position that
contains the amount of time a sailor occupying that
position would spend on their primary and secondary
duties each day. In its current configuration, the
model generates three schedules for each position: the
time spent on primary duties, the time spent on
secondary duties, and total workload.
Each simulation consists of a user-specified
number of replications of a single year. Since events
do not carry over from one simulated year to the next,
it is incorrect to interpret the replications as
WAPAT 2023 - Special Session on Workforce Analytics - Practical Application and Theory
262
consecutive years of a multi-year period. Instead,
each replication should be considered as an
independent simulation of the same year. As such,
they are suitable for use in Monte Carlo experiments.
3.2 Model Inputs
A sailor’s workload depends on the state of their unit
or vessel, so the first input the model makes use of is
a table containing the state of the unit being studied
in each month of the simulation. The four states used
in the model are described in Section 2.1 and
summarized in Table 1.
The system being studied by the model ranges
from a single sailor to an entire unit or vessel’s crew.
Information regarding each position is loaded into the
model as a table that contains the number, name, and
daily primary duty time requirement estimate for each
position. Given that each daily time requirement
estimate consists of a minimum, most-likely, and
maximum value, and that each value must be given
for each unit state, a total of 12 time values are
provided for the primary duty workload of each
position.
Next, a list of secondary duties is required by the
model. To fully describe a secondary duty, the model
requires the minimum, most-likely, and maximum
time required by the secondary duty during each
instance it occurs, the length of the window (in days)
that these hours must be completed within, and the
frequency at which instances occur. Since this
information is required for each unit state, a total of
20 numerical values are necessary to fully describe a
secondary duty in the model.
The final piece of information required by the
model is the assignment of secondary duties to the
sailors occupying the unit positions. This takes the
form of a matrix with each unit position represented
as a row and each secondary duty as a column. The
value of each matrix element determines the portion
of the corresponding secondary duty that is assigned
to the sailor occupying the position in question.
Fractional values allow duties to be split between
multiple sailors, although their individual schedules
are still modelled independently as is discussed in
Section 2.3.
3.3 Generation of Simulated Schedules
The model begins by generating a 14-month, 420-day
calendar and then attaching the day of the week and
unit state to each day using the unit state input file.
The length of the calendar corresponds to an idealized
12-month year with an additional month added to the
beginning and end. These additional months are
included so that the 12-month schedule they
encompass does not include any edge effects such as
a lack of secondary duty events beginning before the
first day of the simulation or secondary duty events
carrying over past the last day of the simulation. The
additional months are not included in the simulation
outputs and do not contribute to analysis of simulated
schedules. 30-day months are used in place of true
month lengths for ease in computing monthly
statistics.
The model then simulates the primary duty
schedule of each sailor by determining the number of
hours worked by each sailor each day using a
triangular distribution and the inputs that correspond
to the unit state of that calendar day.
To simulate the secondary duty schedule of a
sailor, the model iterates over each secondary duty
assigned to the sailor. In larger crews, most sailors
will have a single or no secondary duty assigned to
them; however, smaller crews will see more instances
of sailors being responsible for multiple secondary
duties.
To begin, the probability of an instance of the
secondary duty requiring attention beginning is
computed for each day. This is done by dividing the
annual frequency for the unit state of the day by 360.
Random number generation using a uniform
distribution is then performed to determine when
instances occur. For example, if the probability of an
event starting on a given day is 0.1, a random number
is generated between zero and one and an event is
created beginning on that day only if the number is
less than or equal to 0.1. An illustrative example of
this process is shown in Table 2.
Table 2: Example of the stochastic treatment of secondary
duty occurrences. Only a single unit state is considered.
Frequency of secondary
duty
20 events per year
Probably of instance
beginning each day
20 ÷ 360 = 0.056
Distribution of random
number generation
Uniform distribution
Interval of random
number generation
[0,1]
Requirement for instance
to begin on day
Random number 0.056
Each instance of a secondary duty requiring
attention is spread across a number of days equal to
the window length of that secondary duty during the
Quantifying the Impact of Secondary Duties on Sailor Workload Through Simulation
263
unit state at the beginning of the instance. From this,
an end date is computed for each date on which an
instance begins. In cases where an end date would
extend beyond the end of the calendar, the beginning
date is moved up. Recalling that the final month is
discarded to avoid edge effects, the shifting of the
start and end dates will not impact the simulated
schedules output by the code unless a very long
window is being considered.
Secondary duty instances that begin in one unit
state are allowed to extend to the following one, with
the secondary duty values for the entire instance
being those of the first day of the instance. For
example, if a secondary duty instance begins on the
last day a ship is at sea on operation before returning
to its home port, the three-point time estimate and
window length corresponding to being at sea on
operation are used for the entire instance even though
some days occur while the ship is no longer at sea.
The model then simulates the completion of
secondary duties by assigning hours to days within
the corresponding windows. The number of hours to
be allocated is generated using random number
generation from the triangular distribution defined by
the three-point estimate available for that secondary
duty and unit state. Rather than distribute the hours
among the days within the secondary duty window
evenly, a heuristic approach is used to balance the
total workload of the days. The heuristic approach is
described in Section 3.4.
In cases where a sailor is assigned more than one
secondary duty, the model iterates over them, adding
the hours they require to both the secondary duty and
total workload schedule. Once all secondary duties
for a unit position are completed, the model moves on
to simulating the schedule of the next unit position
until all schedules have been generated. At this point,
the primary duty, secondary duty, and total workload
schedules for all unit positions are output separately
as comma-separated-value files for analysis or import
into other programs.
3.4 Heuristic Approach for Assigning
Hours
In the heuristic approach, what is known of the total
workload of the sailor is used to decide how the time
required by a secondary duty instance is distributed
between the days within that instance’s window.
Before any secondary duties are considered, the total
workload of the sailor is simply their primary duty
workload. As the schedule of each of their secondary
duties are simulated, the hours required by those
duties are added to the total workload.
The goal of the approach is to follow how sailors
manage their time by attempting to distribute the
hours required by a secondary duty in a way that
avoids creating days with extreme workload. In
practice, this means assigning more hours to days
with light workloads than days with heavy workloads.
This is done by dividing the amount of work required
by the secondary duty instance by the number of days
in the window of that secondary duty instance to create
several units of time that must be allocated. In an
iterative approach, each unit of time is added to the day
in the window with the lowest number of hours in the
total workload schedule, with the total workload
schedule being updated to account for each added unit.
As an example, consider a scenario where a sailor
is serving aboard a ship at sea and is working roughly
12 hours per day. The total workload of the sailor
before any secondary duty hours are scheduled is
shown in Figure 1 where Day 1 corresponds to Monday.
Figure 1: Plot of one week of total workload before
secondary duty hours are added.
The sailor is assigned a secondary duty. On
Thursday morning, the sailor becomes aware that
they have two days to complete four hours of work
for their secondary duty. This means that four hours
of secondary duty work have to be assigned to
Thursday and Friday, which correspond to the two-
day window of this instance. If the time is divided
equally between the two days, they end up working
16.3 hours on the Friday and only 12.7 hours on the
Thursday, as shown in Figure 2.
Figure 2: Plot of one week of total workload when the
secondary duty hours are evenly distributed. Two hours are
added to both Thursday and Friday in red.
WAPAT 2023 - Special Session on Workforce Analytics - Practical Application and Theory
264
If the sailor knew that they were going to be
working more hours on Friday than Thursday, they
may choose to complete all four hours of the
secondary duty on Thursday. The heuristic approach
has the same effect, since the four required hours are
divided into two units of two hours each, both of
which are assigned to Thursday. This leads to a total
workload of 14.7 hours on Thursday and 14.3 hours
on Friday, as is seen in Figure 3.
Figure 3: Plot of one week of total workload when the
heuristic approach is used to allocate the secondary duty
hours. All four hours are added to Thursday in red.
Through use of the heuristic approach to allocate
hours, the maximum workload encountered on a
single day is reduced from 16.3 to 14.7. More
importantly, the heuristic approach follows how
individuals manage time, making the model more
representative of reality.
4 NOTIONAL EXAMPLE OF
RESULTS
An example of the application of the simulation
model is presented in this section. Notional data is
used.
Consider a sailor serving aboard a ship that will
spend the first half of a year at sea in an operational
footing before returning to home port for the
remainder of the year. The unit state inputs for this
case are shown in Table 3, where an additional month
has been added to the beginning and end to avoid edge
effects, as discussed in Section 3.3. It is the year
spanning from the 2nd to 13th months that is output
for analysis.
Table 3: Unit state for each month in the notional example.
Months Unit state
1-7 At sea on operation (SO)
8-14 Alongside home port (AH)
The sailor’s primary duties require between 10
and 15 hours per day when at sea on operation, with
a most-likely value of 12.5. When in home port, their
primary duty time requirements are shorter and less
varied and range from seven to nine hours per day
with a value of eight being the most likely. This is
summarized in Table 4, which shows only the
relevant data from the primary duty input file.
Table 4: The primary duty input values for the sailor being
described in the notional example.
SO: Minimum required hours per day 10
SO: Most-likely required hours per day 12.5
SO: Maximum required hours per day 15
AH: Minimum required hours per day 7
AH: Most-likely required hours per day 8
AH: Maximum required hours per day 9
On top of their primary duties, the sailor is also
responsible for two secondary duties. They do not
share the duties with any other sailors, so they are
responsible for the full time requirements of both. The
first duty requires just as much work when the ship is
at sea on operation as when it is alongside its home
port, but the other is much more demanding when the
ship is at sea. The relevant data from the secondary
duty input file are shown in Table 5.
Table 5: The input values for the two secondary duties
assigned to the sailor in the notional example.
Duty 1 2
SO: Instances per year 26 8
SO: Minimum required hours of instance 2 10
SO: Most-likely required hours of instance 4 15
SO: Maximum required hours of instance 6 20
SO: Days in window 2 5
AH: Instances per year 26 2
AH: Minimum required hours of instance 2 3
AH: Most-likely required hours of instance 4 6
AH: Maximum required hours of instance 6 9
AH: Days in window 2 2
Based on the input values, and without running
the simulation, the sailor is expected to spend 12.5
hours on their primary duties each day when at sea on
operation. Over the course of the six months spent on
operation, they would expect to spend 52 hours on
their first secondary duty and 60 hours on their second.
With no further knowledge of the secondary duties,
one could assume that they can be divided up over all
180 days of the time period and result in a daily
workload of approximately 37 minutes per day. This
would bring the length of the overall sailor’s workday
from 12.5 hours to roughly 13.1 hours, leaving 10.9
Quantifying the Impact of Secondary Duties on Sailor Workload Through Simulation
265
hours for rest and recovery. It is unlikely that such a
case would be flagged as unreasonable or as putting
the sailor at high risk of becoming fatigued.
When the simulation model is run using the inputs,
the impact of the secondary duties on individual days
becomes clear. An example of a single simulated year
(a single replication) is shown in Figure 4, where it is
seen that the sailor works many more hours during
some days than others regardless of the state of the
ship.
Figure 4: Primary duty, secondary duty, and total workload
of a single simulated year for the sailor in the notional
example.
Looking specifically at the first six months of this
replication, corresponding to when the sailor is at sea
on operation, the sailor works between 10.39 and
17.52 hours each day with a mean of 12.89 hours. The
sailor is assigned secondary duty work hours on 26 of
the 180 days.
The stochastic nature of the model makes it
necessary to analyze multiple replications to ensure
accurate statistics are generated, since consecutive
replications will differ. For example, the next four
replications of the notional experiment are plotted in
Figure 5. The minimum number of hours worked in
each replication are 10.23, 10.07, 10.25, and 10.31,
respectively. The corresponding maximum values are
Figure 5: Four additional replications of a simulated year
for the sailor in the notional example.
20.17, 18.65, 16.64, and 21.12. Far more variation is
seen in the maximum total workload values than the
minimum values due to the difference in sample size
of days with low and high workload. This is due to
small total workload values occurring when no
secondary duties are performed, which is relatively
common, and high total workload values occurring
when multiple instances of secondary duties being
performed overlap, which is less common.
In total, 100 replications of the notional example
simulation were completed and aggregate statistics
were generated from those. Focusing on the first six
months of the simulated year, when the ship is at sea,
a total of 18,000 simulated days are contained in the
100 replications. The mean of all the simulated days
is 13.10 hours, which agrees with the calculation
performed without using the model. The shortest
workday seen is 10.04 hours and the longest is 23.71
hours. The maximum value represents a day where
many secondary duty instances are occurring
concurrently. While an outlier, the existence of such
a day demonstrates the risk associated with assigning
multiple secondary duties to a single sailor.
The total workloads of all 18,000 days are
represented in the distribution shown in Figure 6. It is
seen that the triangular distribution of the primary
duty workload, which ranges from a minimum of 10
to a maximum of 15, is the dominant feature of the
distribution, although the addition of secondary duty
hours adds significant positive skew.
Figure 6: The distribution of total workload in the 18,000
simulated days falling within the first six months of the
notional example.
In Table 6, the probability of workloads
exceeding certain thresholds are given. This
probabilistic treatment provides a much better sense
of the workload the sailor would experience during
the time at sea than the mean value of 13.1 hours
computed before completing the simulation. For
example, more than five percent of days require the
sailor to work at least 16 hours, leaving less than eight
WAPAT 2023 - Special Session on Workforce Analytics - Practical Application and Theory
266
hours for rest and recovery. Such a statistic is more
suitable for predicting fatigue than the daily mean.
Table 6: Probability of daily workloads more than various
numbers of hours computed for the first six months of the
notional example.
Number of hours
worked
Probability of day requiring
more work
10 1
11 0.9415
12 0.7530
13 0.4679
14 0.2426
15 0.1082
16 0.0524
17 0.0197
18 0.0078
19 0.0034
20 0.0014
21 0.0008
22 0.0003
23 0.0002
24 0
5 VERIFICATION
The model was first implemented in the R
programming language version 4.1.1 (R Core Team,
2021). Verification of the R version was first
completed by comparing statistics generated using
simulated schedules to those computed analytically.
Next, several experiments were designed to isolate
and challenge code features so that errors in the
implementation could be identified and corrected.
The simulation model was then implemented a
second time in Python version 3.8.8 (Python Software
Foundation, 2021). The Python implementation made
use of the Pandas package (Pandas Development
Team, 2021) to allow it to closely follow the R
implementation. Identical experiments could then be
run in both implementations and the results could be
compared to check for issues with either
implementation or the functionality of the packages
involved.
As an example, the experiment discussed in the
notional example was run in both Python and R.
Table 7 summarizes results from both implemen-
tations. Values from the first six months are shown to
avoid conflating time at sea with time in port. The
results of single replications differ in R and Python,
but this is expected due to the stochastic nature of the
model. Similar differences were seen when
comparing individual replications performed in the
same programming language. The results converge
when many replications are considered, as is done in
the table. It is also noted that values that include
secondary duties tend to display more variation in the
mean between replications, which manifests as a
larger standard error. This is due to the limited
number of secondary duty instances in a year,
meaning the sample size of days with secondary duty
hours is smaller than that for days with primary duty
hours.
The distributions of daily total workload hours
simulated for the sailor from the notional example in
one replication and 100 replications are compared in
Figure 7. Again, only the first six months are included.
When considering all 100 replications, only the
outliers are discernibly different when comparing the
Python and R outputs.
Figure 7: Distributions of total workload hours computed in
both Python and R.
Table 7: Summary of results for an experiment performed in both R and Python. 100 replications were completed.
Output variable Mean Standard Error
Python R || Python R
Mean daily primary duty workload (hours) 12.49865 12.50428 0.00563 0.00676 0.00748
Mean daily secondary duty workload (hours) 0.61707 0.59184 0.02523 0.01640 0.02001
Mean daily total workload (hours) 13.11571 13.09612 0.01959 0.01752 0.02125
Minimum daily total workload (hours) 10.25510 10.26931 0.01421 0.01256 0.01384
Maximum daily total workload (hours) 18.93740 18.58606 0.35134 0.17311 0.16486
Days with > 16 hours of total workload (days) 9.68 9.43 0.25 0.44 0.53
Quantifying the Impact of Secondary Duties on Sailor Workload Through Simulation
267
6 PERFORMANCE
Performance benchmarking was completed using an
experiment designed to imitate simulating a large
crew: 100 replications of schedule simulation for a
crew consisting of more than 200 positions that is
assigned 58 secondary duties. Primary duty workload
inputs were identical for most unit positions, with a
few excursions inserted manually. Similarly, the
inputs of only a few secondary duties were unique.
The assignment of secondary duties to unit positions
was mostly ordinal: the Nth secondary duty was
assigned to the unit position occupying the Nth
position in the list. A few exceptions were inserted
manually so shared secondary duties and cases of
sailors being assigned multiple secondary duties
would exist in the test data.
Testing was done on a laptop with a four-core 1.9
GHz processor. The results are presented in Table 8,
where it is seen that the Python version executes
roughly 12 times faster than the R version despite
being structured similarly through the use of the
Pandas package and data frames.
Table 8: Benchmarking of the R and Python versions.
Run time
(minutes:seconds)
R
6:01
Python
0:34
7 CONCLUSION
A model that simulates sailor workload has been
presented. By combining primary and secondary
duties stochastically, it is seen that schedules
produced by the model provide a fuller description of
sailor working life than deterministic estimates of
expected workday lengths. The simulated schedules
are ideal for use in applications where extreme values
in daily workload must be considered including crew
design and research into crew fatigue.
Monte Carlo experimentation using the model can
be used to identify individual secondary duties or
combinations of secondary duties that have a large
impact on sailor workload. In this way, the model can
be used to guide the assignment of secondary duties
or identify secondary duties that may need to be
adjusted or shared between multiple sailors.
ACKNOWLEDGEMENTS
The author would like to thank Cdr Karine
L’Archevêque, LCdr Aislinn Joiner, CPO2 Rayon
Murdock, CPO1 Nathalie Scalabrini, and CPO1
Phillip Gormley for acting as patient and attentive
subject matter experts.
REFERENCES
Chief of Naval Operations (2021). Naval total force
manpower policies and procedures. OPNAVINST
100.16L CH-3. Department of the Navy, Washington,
DC.
Chow, R., Lamb, M., Charest, G., Labbé, D. (2016).
Evaluation of current and future crew sizes and
compositions: Two RCN case studies. Naval Engineers
Journal, 128(4), 59-64.
Clark, C., E. (1962). Letter to the editor—The PERT model
for the distribution of an activity time. Operations
Research, 10(3), 405-406. https://doi.org/10.1287/opre.
10.3.405.
Cordle, J., P. (2019). Manning matters. U.S. Naval Institute
Blog (December 20). https://blog.usni.org/posts/
2019/12/20/manning-matters.
Garbacz, B., D. (2019). How increased manning affects
crewmember’s workload inport and underway: Results
of a study onboard two U.S. Navy destroyers in basic
phase. Naval Postgraduate School, Monterey, CA.
Hollins, R., Leszczynski, K., M. (2014). USN manpower
determination decision making: A case study using
IMPRINT Pro to validate the LCS core crew manning
solution. Naval Postgraduate School, Monterey, CA.
Pandas Development Team (2021). pandas-dev/pandas:
Pandas 1.2.4 (v1.2.4). Zenodo. https://doi.org/10.5281/
zenodo.4681666
Python Software Foundation (2021). The Python language
reference (3.8.13). https://docs.python.org/3.8/referen
ce/index.html#reference-index.
R Core Team (2021). R: A language and environment for
statistical computing. R Foundation for Statistical
Computing, Vienna, Austria.
WAPAT 2023 - Special Session on Workforce Analytics - Practical Application and Theory
268