![](bg2.png)
1.2 Outline
This paper contains four more sections and one ap-
pendix. Section 2 provides an outline of the stock-
piling problem, related work, and the implementation
of FEAST, which touches upon the DES modelling
paradigm, the main assumptions and modeling fea-
tures, and the inputs and outputs. Section 3 intro-
duces a notional case study pertinent to a segment of
the CAF which has been simplified, altered and re-
stricted to a single ammunition nature for purposes of
presentation herein. For example, demand has been
modified from a subset of FG and FE requirements.
Section 4 includes a discussion on practical consid-
erations for usage of the tool, limitations and some
aspects under development. Concluding comments
are made in Section 5. Finally two verification cases
are presented in the Appendix for which the simu-
lated output can be compared to the expected annual
steady-state demand for ammunition.
2 THE FUTURE EVENT
AMMUNITION STOCKPILE
TOOL
FEAST is driven in a “black box” manner by specify-
ing inputs. Here we sketch some approaches to stock-
piling and introduce the key features and elements of
FEAST.
2.1 Approaches to Stockpiling
Approximate Dynamic Programming (ADP) is a
widely used approach for operational research ques-
tions involving sequential decisions under uncer-
tainty. ADP has been applied to help optimize in-
ventory management policies (Powell, 2009) and to
address a variety of other challenges facing the mil-
itary (Rempel and Cai, 2021). At its core, ADP is
a discrete-time approach where the state of the sys-
tem transitions under a decision based on observed
information and there is a cost/reward associated with
how decisions play out. With these elements ADP
attempts to find functions, or policies, that optimally
map the k-th state to the k-th decision. In terms of
stockpiling, the state encodes the stockpile, the infor-
mation is the (generally stochastic) demand over the
time period, the decision is the amount of ammuni-
tion to purchase in that time step, and the costs are
ordering, storage, and stockout penalty (real or vir-
tual) costs. While ADP is a powerful and general
framework, demand is required as an input, a suit-
able time step size selected, and constraints on good
policies to be searched for imposed. Here we will
be focusing on more primary concerns, considering
a rolling-horizon approach (Powell, 2009) to gener-
ate demand and ask “what if” questions, leaving opti-
mization concerns and approaches that further exploit
these elements for future work.
For ammunition stockpiling, more pedestrian ap-
proaches are often taken due to the numerous con-
straints, complicating factors, and unknowns in-
volved. For example, one complicating factor is
that a national stockpile is not a single reserve,
but is partitioned into several sub-stockpiles, includ-
ing those reserved for ongoing operations, train-
ing, war contingency, experiments and disposal (see
page 77 of (Brown, 2008)). Additionally, stock-
piles are physically partitioned and located in differ-
ent places geographically, e.g., at continental hold-
ings, off-continent support hubs, and deployed bases
(Bacot, 2009). Another complicating factor is man-
ufacturer stockpiles awaiting sale and partner stock-
piles are “potential” stockpiles as they are extant but
there may be limited visibility into, and uncertainty
ability to draw from, these sources and it is difficult
to model such situations. Ref. (Guy, 2010) discusses
a number of issues for strategic stockpiling in a North
Atlantic Treaty Organization (NATO) context that can
pose severe modelling challenges. One issue, for
example, is that “engagements may not follow pre-
conceived doctrine”. This issue can result from the
evolution of asymmetric warfare capabilities and em-
ployment approaches, to include “overkill” weapons,
where precision or battle decisive munitions to engage
unmounted opponents may undercut existing battle
assumptions. A recent example of this is the use of
drones in Ukraine as loitering weapons, for recon-
naissance, and to improve artillery precision (see, for
example, Ref. (Vershinin, 2022)).
Since stockpiles sit between ammunition supply
and ammunition demand, these two fundamental as-
pects are often treated separately in the construction
of models. However, with FEAST we enable the
modeling of both aspects together.
To determine the demand there are two ap-
proaches used by NATO members, the Target Ori-
ented Methodology (TOM) and the Level of Effort
(LoE) methodology (Andrews and Hurley, 2004; Guy,
2010). TOM requires a precise list of targets, a list
of the targeting platforms, as well as probability of
single shot or multiple shot kills. As such TOM can
be onerous to work with and validity depends an the
adequacy of the lists and success models. On the
other hand LoE takes an intensity level, which sets
a usage rate, and duration to determine demand—
where the number of days at given intensity, multi-
A Simulation Tool for Exploring Ammunition Stockpile Dynamics
39