...Whole Fleet Management
- Case Study
The British Army is examining
a fundamental change in the way it manages its entire
fleet of Military vehicles (Fleet Management). Currently,
each Army Unit holds most of the vehicles it needs
to achieve all of its training objectives, resulting
in several key drawbacks...
- Low vehicle utilisation - vehicles are left idle
whilst other Units have deficits
- Intensive maintenance required - soldiers spend
much of their time maintaining / repairing
- Complicated and time consuming to assemble vehicles
quickly under 'Force Generation'
(FG)
Under 'Whole Fleet Management' (WFM)
several candidate schemes have been proposed. For example,
each Army Unit will only hold enough vehicles to carry
out low-level training operations. Additional vehicles
for more demanding training will be obtained from centrally
managed Training Pools (or potentially from other Units).
Surplus vehicles will be placed in Operational Pool(s)
- vehicles in this pool will be fully prepared and
stored in a 'Controlled Humidity Environment'
(CHE) in order to avoid deterioration (and maintenance)
over time.
WFM offers several advantages...
- Shared vehicle usage - high utilisation
- Low Unit holdings - less maintenance for soldiers
- Faster deployment under FG - can pull directly
from Operational Pool
- Cost savings - economies of scale
Use of Simulation
The initial task was to create a Model to represent
every vehicle in the British Army, every Unit, every
Training Area and every training exercise in the UK
and Germany (and later Canada) Theatres. In depth results,
captured by and extracted from model, would then help
to assist answering several key questions...
- How should the vehicles be distributed amongst
the Units, Training Pools and Operational Pool(s)
- What sort of training effectiveness can be achieved
with the above distribution - is it better than the
current system?
- How fast can the vehicles be deployed under 'Force
Generation' - is it faster than the current system?
- How much will the WFM solutions cost - is it more
cost effective than the current system?
The model was originally written (by iBright Ltd)
in eM-Plant, a leading object-oriented simulation tool
that was powerful enough to cope with the scale and
complexity of the WFM scenario. The model was later
converted (by iBright Ltd) to baseSim -
see 'Iteration Speed'.
The results from the initial study were used to feed
the live trials carried out in Germany in 2002. The
feedback from this study has been very positive and
has helped to enhance the status of the model.
|
|
 |
Figure 1 - Hierarchical
nature of the WFM model
|
Visualisation
The object-oriented (hierarchical) nature of the modelling
tool chosen enabled the model to be developed in an
intuitive way. The top-level of the model contained
the training Theatres, each Theatre contained Units,
Training Areas, Pools etc., each Training Pool, for
example, contained a vehicle park and the vehicle park
contained the relevant numbers and types of vehicles
- see Figure 1.
Using this technique of simulation modelling resulted
in several key advantages...
- Realism - the scenario was modelled at the lowest
level of detail i.e. every vehicle (each vehicle
is an object with its own properties), every training
event, every movement of every vehicle to the Training
Area etc. Therefore, the results are incredibly detailed
and all encompassing.
- Validation - it was possible to validate the accuracy
of the model, not only by running test cases and
analysing results but also by visually checking the
movements of vehicles, use of Training Areas and
also the logic of allocations i.e. was the correct
pool or Unit asked in the correct sequence?
- Marketing - the ability to show the entire training
operations of the British Army, for a full year,
including the movement of vehicles, transactions
and repair became a very powerful selling tool in
making the case for WFM.
Iteration Speed
The original model (eM-Plant) impressed many people
in terms of its capabilities, scale and speed. The
model took about 50 minutes to process
an entire years worth of training operations for the
British Army. This was just one iteration of the model.
Normally, in order to build confidence in a stochastic
simulation model, many iterations are carried out,
each with a change in random number seeds. This reduces
the chance of a 'fluke' run and allows sensitivity
analysis to be performed. This may take between 300-1000
iterations (could be less - problem specific).
Additionally, there was a desire to use optimisation
techniques (Genetic Algorithms) in
order to use the model to answer the key questions
e.g. what is the best (most cost effective) location
of the Training Pools (using the cost engine). This
would need even more iterations (maybe several thousand).
This would have taken the original model several
weeks to perform.
The decision was taken to convert the model into baseSim.
baseSim offered all the power and flexibility of eM-Plant
but, as it is a fully compiled
simulator, could perform the same job in a fraction
of the time.
The baseSim model was completed using the same input
driving data and produced exactly the same
output results as eM-Plant. The model was
also made to look and behave in a very similar manner.
The baseSim model did all this in 1 min 30
seconds (on the same development laptop as
the eM-Plant model) - including reading in data, dynamically
building the model, running the model and writing out
the results. The core simulation ran 75 times
faster than before (reading / writing data
did not change as it used the same ODBC link).
Further speed optimisations were made in order to
tailor the model for the experimentation and
optimisation tasks...
- Reading in the data once - not for every iteration
- Allowing the model to be reset without deleting
the model hierarchy
This brought down the core simulation time per iteration
(on the same development laptop) to around 30
seconds. On the simulation production desktop
computer, provided by Insys
Ltd, the time per iteration was around 17
seconds - see Figure 2.
This enabled experimentation (sensitivity analysis)
to be performed, 1000s of iterations can now be run
overnight (over 100 per hour) - see Figure
3.
This also enabled optimisation (using Genetic
Algorithms) to be performed, 1000s of iterations
can now be run overnight (over 100 per hour) - see Figure
4.
|
|
|
Figure 4 - Using Genetic
Algorithms to home in on an optimum solution
|
Benefits of baseSim
baseSim was used to create a large, complex and object-oriented
(hierarchical) simulation model to represent the future
transition of the British Army to Whole Fleet Management.
baseSim achieved the same level of accuracy and confidence
as an other leading simulation tool, whilst producing
exactly the same results in a fraction of the time.
The use of baseSim has enabled Insys Ltd and the WFM IPT to
experiment with the various options surrounding the
WFM case and also to perform optimisations, predicated
against cost.
Commercial
The Whole Fleet Management project has been carried
out under contract to 'Insys Ltd' (formerly Hunting
Engineering Ltd) on behalf of the 'Whole Fleet Management
Integrated Project Team' (Headquarters Land Command).
All screenshots are taken from the WFM simulation
model, written using the baseSim simulation components.
Video with Voice-Narration
Please feel free to have a look at the case study
video presentation, with voice-over, that you can stream
using Microsoft's Media Player (please ensure your
speaker volume is enabled). This shows the principal
characteristics of the WFM simulation model.
Glossary
CHE - see 'Controlled
Humidity Environment'
'Controlled Humidity Environment' -
protected from damp and dust - reduces need for preventative
maintenance
Experimentation -
the results taken from a single model run may not be
representative of the system being modelled - better
to do multiple runs and take average
FG - see 'Force Generation'
Force Generation -
assembling vehicles and soldiers in formations quickly
for deployment to military operations overseas
Fully Compiled Simulator -
compiled by a programming language into machine code.
Executes the model as quickly as possible. As opposed
to 'interpreted' simulation tools, where the model
logic is 'evaluated' at run-time
Genetic Algorithms -
numerical optimisation algorithms based on natural
selection. They can be applied to a wide variety of
problems and offer several key advantages over other,
more traditional, techniques (Genetic
Algorithm components)
IPT - Integrated
Project Team
WFM - Whole Fleet Management
|
