8/19/2019 MIT2_854F10_inv.pdf
1/78
Inventory
Lecturer: Stanley B. Gershwi
8/19/2019 MIT2_854F10_inv.pdf
2/78
Storage
• Storage is fundamental in nature, manageengineering.
⋆ In nature, energy is stored. Life can only exist if
acquisition of energy can occur at a different time
the expenditure of energy.
⋆ In management, products are stored.
⋆ In engineering, energy is stored (springs, batteri
capacitors, inductors, etc.); information is stored
hard disks); water is stored (dams and reservoir
8/19/2019 MIT2_854F10_inv.pdf
3/78
Storage
• The purpose of storage is to allow systemssurvive even when important events are
unsynchronized. For example,⋆ the separation in time from the acquisition or pro
something and its consumption; and
⋆ the occurrence of an event at one location (such machine failure or a power surge) which can prev
performance or do damage at another.
•
Storage improves system performance bydecoupling parts of the system from one a
8/19/2019 MIT2_854F10_inv.pdf
4/78
Storage
• Storage decouples creation/acquisition an
emission.
• It allows production systems (of energy or
be built with capacity less than the peak de
•
It reduces the propagation of disturbancesreduces instability and the fragility of comp
expensive systems.
8/19/2019 MIT2_854F10_inv.pdf
5/78
Manufacturing InveStorage
Motives/bene
• Reduces the propagation of disturbances
machine failures).
•
Allows economies of scale:⋆ volume purchasing
⋆ set-ups
• Helps manage seasonality and limited cap
• Helps manage uncertainty:
⋆ Short term: random arrivals of customers⋆ Long term: Total demand for a product n
8/19/2019 MIT2_854F10_inv.pdf
6/78
Manufacturing InveStorage
Costs
•
Financial: raw materials paid for, but no re
comes in until item is sold.
•
Demand risk: item may not be sold due to example)
⋆ time value (newspaper)
⋆ obsolescence
⋆ fashion
•
Holding cost (warehouse space)•Damage/theft/spoilage/loss
8/19/2019 MIT2_854F10_inv.pdf
7/78
NewsvendorProblem
... formerly called the “Newsboy Problem”.
Motive: demand risk.
Newsguy buys x newspapers at c dollars per paper (Demand for newspapers, at price r > c is w. Unsold
newspapers are redeemed at salvage price s < c.
Demand W is a continuous random variable with disfunction F (w) = prob(W ≤ w); f (w) = dF (w)/d
all w.
Note that w ≥ 0 so F (w) = 0 and f (w) = 0 for w
Problem: Choose x to maximize expected
8/19/2019 MIT2_854F10_inv.pdf
8/78
NewsvendorProblem
rx if x ≤ wRevenue = R =
rw + s(x − w)if
x > w
(r − c)x if x ≤ wProfit = P = rw + s(x − w) − cx
= (r − s)w + (s − c)x
8/19/2019 MIT2_854F10_inv.pdf
9/78
NewsvendorProblem
700 RevenueProfit
-100
0
100
200
300
400
500
600
-2000 200 400 600 800 1000 1200 1400
stock
8/19/2019 MIT2_854F10_inv.pdf
10/78
NewsvendorProblem
Revenue=Revenue=rw+s(x−w)
x < w x > w
Profit=(r−Profit=rw+s(x−w)−cx
f(w)
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
8/19/2019 MIT2_854F10_inv.pdf
11/78
xNewsvendorProblem
Expected Profit = EP (x) =
[(r − s)w + (s − c)x]f (w)dw−∞ ∞+ (r − c)xf (w)dw
x
8/19/2019 MIT2_854F10_inv.pdf
12/78
NewsvendorProblem
or EP (x)
x x
= (r − s) wf (w)dw + (s − c)x f−∞ −∞ ∞
+(r − c)x f (w)dwx
x
= (r − s) wf (w)dw−∞( ) F ( ) ( ) (1 F (
8/19/2019 MIT2_854F10_inv.pdf
13/78
NewsvendorProblem
EP is 0 when x = 0.
When x → ∞, the first term goes to a finite(r − s)Ew. The last term goes to 0.
The remaining term,
x(s − c)F (x) → −∞.
Therefore when x → ∞, EP → −∞.
8/19/2019 MIT2_854F10_inv.pdf
14/78
NewsvendorProblem
EP(x)
x
?
8/19/2019 MIT2_854F10_inv.pdf
15/78
NewsvendorProblem
EP(x)
x
?
8/19/2019 MIT2_854F10_inv.pdf
16/78
NewsvendorProblem
dEP= (r − s)xf (x)+ (s − c)F (x)+(s
dx +(r − c)(1 − F (x)) − (r − c)xf (x)
= xf (x)(r − s + s − c − r + c)
+r − c + (s − c − r + c)F (x)
= r − c + (s − r)F (x)
8/19/2019 MIT2_854F10_inv.pdf
17/78
NewsvendorProblem
Note that
• d2EP/dx2 = (s − r)f (x) ≤ 0. Therefor
concave and has a maximum.
• dEP/dx > 0 when x = 0. Therefore EP
maximum which is greater than 0 for some∗x = x > 0.dEP
• x∗ satisfies (x∗) = 0.dx r − c
Therefore F (x∗)
8/19/2019 MIT2_854F10_inv.pdf
18/78
NewsvendorProblem
EP(x)
x
x*
8/19/2019 MIT2_854F10_inv.pdf
19/78
NewsvendorProblem
x*
f(w)
r−c
r−s
F(x*)=
0
0.05
0.1
0.15
0.2
µ−4σ µ−2σ µ µ+2σ µ+4σ
8/19/2019 MIT2_854F10_inv.pdf
20/78
NewsvendorProblem
This can also be writtenr − c
F (x∗
) = (r − c) + (c − s)
r − c > 0 is the marginal profit when x <
c − s > 0 is the marginal loss when x > w
Choose x∗ so that the fraction of time you d
too much is marginal profit
8/19/2019 MIT2_854F10_inv.pdf
21/78
NewsvendorExample 1
r = 1, cProblem = .25, s = 0, µw = 10
x* x* vs. r
86
90
94
98
102
106
110
8/19/2019 MIT2_854F10_inv.pdf
22/78
NewsvendorExample 2
r =Problem 1, c = .75, s = 0, µw = 10
x* x* vs. r
105
101
97
93
89
85
81
8/19/2019 MIT2_854F10_inv.pdf
23/78
NewsvendorExample 1
r = 1, cProblem = .25, s = 0, µw = 10
x* x* vs. c
130
80
90
100
110
120
8/19/2019 MIT2_854F10_inv.pdf
24/78
NewsvendorExample 1
r = 1, cProblem = .25, s = 0, µw = 10
x* x* vs. s
130
110
114
118
122
126
8/19/2019 MIT2_854F10_inv.pdf
25/78
NewsvendorExample 2
r =Problem 1, c = .75, s = 0, µw = 10
x* x* vs. s
121
117
113
109
105
101
97
8/19/2019 MIT2_854F10_inv.pdf
26/78
NewsvendorExample 1
r = 1, cProblem = .25, s = 0, µw = 10
x* x* vs. Mean
240
40
80
120
160
200
8/19/2019 MIT2_854F10_inv.pdf
27/78
NewsvendorExample 2
r = 1, cProblem = .75, s = 0, µw = 10
x* x* vs. Mean
30
70
110
150
190
230
8/19/2019 MIT2_854F10_inv.pdf
28/78
NewsvendorExample 1
r = 1, cProblem = .25, s = 0, µw = 10
x* x* vs. Std
114
112
110
108
106
104
102
8/19/2019 MIT2_854F10_inv.pdf
29/78
NewsvendorExample 2
r = 1, cProblem = .75, s = 0, µw = 10
x* x* vs. Std
88
90
92
94
96
98
100
8/19/2019 MIT2_854F10_inv.pdf
30/78
�
NewsvendorWhy is x∗ linear in µw
Problem
r − cF (x∗) =
r − s
Assume demand w is N (µw, σw). Then
w − µwF (w) = Φ
σw
where Φ is the standard normal cumulative d
8/19/2019 MIT2_854F10_inv.pdf
31/78
�
NewsvendorWhy is x∗ linear in µw
Problem
Therefore
� x∗ − µw r − c
Φ =σw r − s
or
� x∗ − µw r − c= Φ−1σw r − s
or
r − cx∗ = µw + σwΦ
−1
8/19/2019 MIT2_854F10_inv.pdf
32/78
NewsvendorWhy is x∗ linear in µw
Problem
Note that
Φ−1(k) > 0 if k > .5
and
Φ−1(k) < 0 if k < .5
Therefore r − c• x∗ increases with σw if > .5, and
r − sr − c
•
x∗ decreases with σw if < .5.r − s
8/19/2019 MIT2_854F10_inv.pdf
33/78
EOQ
• Economic Order Quantity
• Motivation: economy of scale in ordering.
• Tradeoff:
⋆ Each time an order is placed, the compa
fixed cost in addition to the cost of the go⋆ It costs money to store inventory.
A i
8/19/2019 MIT2_854F10_inv.pdf
34/78
Assumptions
EOQ
•
No randomness.
• No delay between ordering and arrival of g
•
No backlogs.• Goods are required at an annual rate of λ
year. Inventory is therefore depleted at the
units per year.• If the company orders Q units, it must pay
the order. s is the ordering cost , c is the u
•
It costs h to store one unit for one year. h
P bl
8/19/2019 MIT2_854F10_inv.pdf
35/78
Problem
EOQ
•
Find a strategy for ordering materials that
minimize the total cost.
• There are two costs to consider: the order
and the holding cost.
S i
8/19/2019 MIT2_854F10_inv.pdf
36/78
ScenarioEOQ
• At time 0, inventory level is 0.
•
Q units are ordered and the inventory leve
instantaneously to Q.
• Material is depleted at rate λ.
•
Since the problem is totally deterministic, wwait until the inventory goes to zero before
next. There is no danger that the inventory
zero earlier than we expect it to.
S i
8/19/2019 MIT2_854F10_inv.pdf
37/78
ScenarioEOQ
•
Because of the very simple assumptions, w
assume that the optimal strategy does not
over time.
• Therefore the policy is to order Q each tim
inventory goes to zero. We must determineoptimal Q.
S i
8/19/2019 MIT2_854F10_inv.pdf
38/78
ScenarioEOQ
I n v e n t o r y
Q
jumpdecrease at rate λ
T 2T
T = Q/λ (years)
F l ti
8/19/2019 MIT2_854F10_inv.pdf
39/78
Formulation
EOQ
• The number of orders in a year is 1/T = λ
Therefore, the ordering cost in a year is sλ
• The average inventory is Q/2. Therefore t
average inventory holding cost is hQ/2.
•
Therefore, we must minimizehQ sλ
C = +2 Q
over Q.
F l ti
8/19/2019 MIT2_854F10_inv.pdf
40/78
Formulation
EOQ
ThendC h sλ
= − = 0dQ 2 Q2
or
2sλ
Q∗ =h
E amples
8/19/2019 MIT2_854F10_inv.pdf
41/78
Examples
EOQ
In the following graphs, the base case is
•
λ = 3000• s = .001
• h = 6
Note that
Q∗ = 1
Examples
8/19/2019 MIT2_854F10_inv.pdf
42/78
Examples
EOQ
0
10
20
30
40
50
0 2 4 6 8
o s t
s=.01s=.001
s=.0001
Examples
8/19/2019 MIT2_854F10_inv.pdf
43/78
Examples
EOQ
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
0 2000 4000 6000 8000 1
Q *
Examples
8/19/2019 MIT2_854F10_inv.pdf
44/78
Examples
EOQ
0
0.5
1
1.5
2
2.5
3
3.5
0 0 002 0 004 0 006 0 008
Q
*
Examples
8/19/2019 MIT2_854F10_inv.pdf
45/78
Examples
EOQ
0
1
2
3
4
5
6
7
8
0 2 4 6 8
Q *
8/19/2019 MIT2_854F10_inv.pdf
46/78
More Issues
• Random delivery times and amounts.
•
Order lead time.
• Vanishing inventory.
• Combinations of these issues and random
ordering/setup costs.
Make to Stock Qu
8/19/2019 MIT2_854F10_inv.pdf
47/78
Base Stock Make-to-Stock Qu
Policy
Usual assumptions:
• Demand is random.
•
Inventory is held (unlike newsvendor problem).
• No ordering cost, batching, etc.
Policy• Try to keep inventory at a fixed level S .
Issue
• Stockout: a demand occurs when there is no stock
Make to Stock Qu
8/19/2019 MIT2_854F10_inv.pdf
48/78
Base Stock Make-to-Stock Qu
Policy
Issues
• What fraction of the time will there be stoc
how much demand occurs during stockout
• How much inventory will there be, on the a
•
How much backlog will there be, on the av
Make to Stock Qu
8/19/2019 MIT2_854F10_inv.pdf
49/78
Base Stock Make-to-Stock Qu
Policy
Production Dispatch Material
Raw SupplierFrom
Factory
Floor
FG
Bac
Material Flow Information Flow
• When an order arrives (generated by “Demand”),
⋆ an item is requested from finished goods invento⋆ an item is delivered to the customer from the finis
inventory (FG) or, if FG is empty, backlog is incre
and⋆ a signal is sent to “Dispatch” to move one item fr
Make to Stock Qu
8/19/2019 MIT2_854F10_inv.pdf
50/78
Base Stock Make-to-Stock Qu
Policy
Production Dispatch Material
Raw SupplierFrom
Factory
Floor
FG
Bac
Material Flow Information Flow
• There is some mechanism for ordering raw materia
suppliers.
• We do not consider it here.
• We assume that the raw material buffer is never em
Make to Stock Qu
8/19/2019 MIT2_854F10_inv.pdf
51/78
Base Stock Make-to-Stock Qu
Policy
Production Dispatch Material
Raw SupplierFrom
Factory
Floor
FG
Bac
Material Flow Information Flow
• Whenever the factory floor buffer is not empty and
production line is available, the production line take
from the factory floor buffer and starts to work on it
• When the production line completes work on a par
finished part in the finished goods buffer or, if there
it sends the part to the customer and backlog is red
S Make-to-Stock Qu
8/19/2019 MIT2_854F10_inv.pdf
52/78
Base Stock Make-to-Stock Qu
PolicyQ
Production Dispatch Material
Raw SupplierFrom
Factory
Floor
F
Ba
Material Flow Information Flow
• Q(t) = factory floor inventory at time t.
•
I (t) =
⋆ the number of items in FG, if this is nonnegative and thereor
⋆ – (the amount of backlog), if backlog is non-negative and F
• There are no lost sales. I (t) is not bounded from below. It cnegative value
B S k Make-to-Stock Qu
8/19/2019 MIT2_854F10_inv.pdf
53/78
Base Stock Make-to-Stock Qu
PolicyQ
Production Dispatch Material
Raw SupplierFrom
Factory
Floor B
Material Flow Information Flow
• Assume Q(0) = 0 and I (0) = S . Then Q(0) + I (0) = S .
• At every time t when a demand arrives,
⋆ I (t) decreases by 1 and Q(t) increases by 1 so Q(t) + Ichange.
• At every time t when a part is produced,
⋆ I (t) increases by 1 and Q(t) decreases by 1 so Q(t) + Ichange.
B S k Make-to-Stock Qu
8/19/2019 MIT2_854F10_inv.pdf
54/78
Base Stock Make-to-Stock Qu
PolicyQ
Production Dispatch Material
Raw SupplierFrom
Factory
Floor B
Material Flow Information Flow
• Also, Q(t) ≥ 0 and I (t) ≤ S .
• Assume the factory floor buffer is infinite.
• Assume demands arrive according to a Poisson pr
rate parameter λ.
• Assume the production process time is exponentia
distributed with rate parameter µ.
B S k Make-to-Stock Qu
8/19/2019 MIT2_854F10_inv.pdf
55/78
Base Stock Make-to-Stock Qu
Policy
• Assume λ < µ.
Therefore, in steady state,
prob(Q = q) = (1 − ρ)ρq, q ≥ 0
where ρ = λ/µ < 1.
• What fraction of the time will there be stoc
That is, what is prob(I ≤ 0)?
B St k Make-to-Stock Qu
8/19/2019 MIT2_854F10_inv.pdf
56/78
Base Stock Make-to-Stock Qu
Policy
∞
prob(I ≤ 0) = prob(Q ≥ S ) = (1 − ρ)ρq
q=S
∞ ∞ = (1 − ρ) ρq = (1 − ρ)ρS ρq = (1 − ρ)ρS
1q=S q=0
soprob(I ≤ 0) = ρS
• How much demand occurs during stockout periods?
ρS λ
B St k Make-to-Stock Qu
8/19/2019 MIT2_854F10_inv.pdf
57/78
Base Stock Make-to-Stock Qu
Policy
•
How much inventory will there be, on the average?
on what you mean by “inventory”.)
ρ•
EQ =1 − ρ
• EI = S − EQ is not the expected finished goods
is the expected inventory/backlog.
• We want to know
I if I > 0E (I +) = E = E (I |I > 0)pro
h i
B St k Make-to-Stock Qu
8/19/2019 MIT2_854F10_inv.pdf
58/78
Base Stock Make-to-Stock Qu
PolicyS
E (I |I > 0) = i prob(I = i|I > 0)i=1
S
i prob(I = iS
i prob(I = i ∩ I > 0) i=1= = S prob(I > 0) i=1
prob(I = j j=1
S −1
(S − q) prob(Q = q)q=0
=S −1
prob(Q = s)
s=0
B St k Make-to-Stock Qu
8/19/2019 MIT2_854F10_inv.pdf
59/78
Base Stock Make to Stock Qu
Policy
S −1 S −1
S prob(Q = q) − q prob(Q =q=0 q=0
=S −1
prob(Q = q′)q ′=0
S −1 S −1
q prob(Q = q) (1 − ρ) qρq
q=0 q=0= S − = S − =
S −
1 S −
1prob(Q = q′) (1 − ρ) ρq
B St k Make-to-Stock Qu
8/19/2019 MIT2_854F10_inv.pdf
60/78
Base Stock Make to Stock Qu
Policy
ThereforeS −1
qρq
q=0
E (I |I > 0) = S − S −1
ρq
q=0
and S −1 S −1
E (I +) = S (1 − ρ) ρq − (1 − ρ) qq=0 q=0
S −1
= S prob(Q < S) − (1 − ρ) qρq
B St k Make-to-Stock Qu
8/19/2019 MIT2_854F10_inv.pdf
61/78
Base Stock Make to Stock Qu
Policy
• How much backlog will there be, on the av
•
This time, we are looking for −E (I −), whe
I if I ≤ 0
E (I −
) = E = E (I |I ≤ 0)0 otherwise−∞
E (I |I ≤ 0) = i prob(I = i|I ≤i 0
B St k Make-to-Stock Qu
8/19/2019 MIT2_854F10_inv.pdf
62/78
Base Stock Make to Stock Qu
Policy−∞
i prob(I−∞ prob(I = i ∩ I < 0) i=0= i =
−∞prob(I < 0) i=0prob(I
j=0∞
(S − q) prob(Q = q)
q=S =∞
prob(Q = s)s=S
where, because Q = S − I , we substitute q = S −
B St k Make-to-Stock Qu
8/19/2019 MIT2_854F10_inv.pdf
63/78
Base Stock Make to Stock Qu
Policy
∞ ∞
S prob(Q = q) − q prob(Q =q=S q=S
=∞
prob(Q = q′)′q =S
∞ S −1
q prob(Q = q) EQ − q prq=S q=0
= S − = S −∞ S −
1prob(Q = q′) 1 −
prob
Base Stock Make-to-Stock Qu
8/19/2019 MIT2_854F10_inv.pdf
64/78
Base Stock Make to Stock Qu
Policy
S −1
EQ − (1 − ρ) qρq
q=0= S −S −1
1 − (1 − ρ) ρq
q=0so
E (I −) = E (I |I ≤ 0)prob(I ≤ 0
( +) ( )
Base Stock Make-to-Stock Qu
8/19/2019 MIT2_854F10_inv.pdf
65/78
Base Stock Make to Stock Qu
Policyprob I pos
0.2
0.4
0.6
0.8
1
0 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
λ
Base Stock Make-to-Stock Qu
8/19/2019 MIT2_854F10_inv.pdf
66/78
10
Base Stock Make to Stock Qu
PolicyEIposEIneg
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
-10
-20
-30
-40
λ
Base Stock Make-to-Stock Qu
8/19/2019 MIT2_854F10_inv.pdf
67/78
Base Stock Make to Stock Qu
Policy
0.2
0.4
0.6
0.8
1
prob I pos, lambda=.6 prob I pos, lambda=.9
0
0 5 10 15 20 25 30 35
S
Base Stock Make-to-Stock Qu
8/19/2019 MIT2_854F10_inv.pdf
68/78
Base Stock Make to Stock Qu
Policy
-10
-5
0
5
10
15
20
25
30
35
40
0 5 10 15 20 25 30 35
EIpos, lambda=.6EIneg, lambda=.6EIpos, lambda=.9EIneg, lambda=.9
S
Base Stock Make-to-Stock Qu
8/19/2019 MIT2_854F10_inv.pdf
69/78
Base Stock a e to Stoc Qu
Policy Optimization of
We have analyzed a policy but we haven’t completely
How do we select S ?
One possible way: choose S to minimize a function texpected costs due to finished goods inventory (E (I
backlog (E (I −)). That is, minimize
EC (I ) = E (hI + − bI −) = hE (I +) − bE (
where h is the cost of holding one unit of inventory founit and b is the cost of having one unit of backlog fo
Base Stock Make-to-Stock Qu
8/19/2019 MIT2_854F10_inv.pdf
70/78
Base Stock Q
Policy Optimization of
cost
−b
h
Base Stock Make-to-Stock Qu
8/19/2019 MIT2_854F10_inv.pdf
71/78
Base Stock Q
Policy Optimization of
5
10
15
20
25
30
35
40
Cost, lambda=.6Cost, lambda=.9
0
0 5 10 15 20 25 30 35
S
Base Stock Make-to-Stock Qu
8/19/2019 MIT2_854F10_inv.pdf
72/78
Base StockPolicy Optimization of
Problem: Find S ∗ to minimize hE (I +) − bE
Solution
1
: S
∗
= ⌊S ˜⌋ or ⌊S
˜⌋ + 1, where
S ̃+1 hρ =
h + b
Note: ρS ̃+1 = prob(Q ≥ S ̃+ 1) = prob(I
Base Stock Make-to-Stock Qu
8/19/2019 MIT2_854F10_inv.pdf
73/78
Base StockPolicy Optimization of
That is, we choose S so thath
prob(I < 0) =h + b
Look familiar? This is the solution to the newsvendor problem ifI = x − w,h = c − s, and b = r − c because
r −F (x∗) = prob(w ≤ x∗) = prob(I ≥ 0) =
(r − c) +so
r − c c − sprob(I < 0) = 1 − =
(r − c) + (c − s) (r − c) + (
c − s is the cost of one more unit of inventory when x > w, ie I
Base Stock Make-to-Stock Qu
8/19/2019 MIT2_854F10_inv.pdf
74/78
Base StockPolicy Optimization of
profit
r−c
s−c
w
Newsvendor profit function
Base Stock Make-to-Stock Qu
8/19/2019 MIT2_854F10_inv.pdf
75/78
Base StockPolicy Optimization of
Questions:
•
How does the problem change if we considfloor inventory? How does the solution cha
• How does the problem change if we consid
finished goods inventory space ? How doesolution change?
•
How else could we have selected S ?
8/19/2019 MIT2_854F10_inv.pdf
76/78
Q, R Policy
• Fixed ordering cost, like in EOQ problem
• Random demand, like in newsvendor or ba
problems.
• Policy: when the inventory level goes down
a quantity Q.• Hard to get optimal R and Q; use EOQ an
stock ideas.
Inventory
8/19/2019 MIT2_854F10_inv.pdf
77/78
InventoryPosition
Inventory position
Q
Lead time
Inventory
Q/ λ
Material ordered Material arrives
The current order must satisfy demand until the next
MIT OpenCourseWare
8/19/2019 MIT2_854F10_inv.pdf
78/78
MIT OpenCourseWarehttp://ocw.mit.edu
2.854 / 2.853 Introduction to Manufacturing Systems
Fall 201
0
For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.
http://ocw.mit.edu/http://ocw.mit.edu/termshttp://ocw.mit.edu/termshttp://ocw.mit.edu/termshttp://ocw.mit.edu/