Analysis and Management

of Production System
Lesson 12: Single Workstation Analysis

Prof. Giulia Bruno

Department of Management and

Production Engineering

giulia.bruno@polito.it
Variability of arrival time

We have considered only the variability of processing time. However,

there is a variability also in the arrival time of items to the workstation.

Low and high

Low variability – arrival variability
restricted
distribution of
process times

Time
High variability –
wide
distribution of
process times

Time
Variability of arrival time

Item arrival Process time

time (10 minutes)

Example
Units waiting for
Constant processing time: process

t0 = 10 min

Items waiting for process

Constant arrival rate:
A=1/30 pcs/min
B=1/20 pcs/min
C=1/10 pcs/min
D<1/10 pcs/min
Utilization
Variability of arrival time

Example Waiting time in queue = 3 min

Constant processing time = 10 min 6 units processing time = 65 min
Variable arrival rate Mean waiting time = 3/65 = 0.046 min
Inactivity time = 5 min = 7.7% of 65
Utilization = 92.3%

Units which enters

Item Arrival Start of End of Waiting Unused machine
Time process process time 1 min in a process with
Unused machine variable arrival
2 min
Waiting time in times and a
queue
1 min constant processing
Unused machine
2 min time (10 minutes)
Waiting time in queue
2 min

Time (minutes)
Variability of arrival time

Waiting items

Utilization

Mean waiting time = 3/65 = 0.046 min

Utilization = 92.3%
Mean waiting time

Utilization
Variability of arrival time

X = accepting long waiting

Decreasing times in order to obtain a high
Variability
utilization of the machine
Y = accepting a low machine
Utilization
utilization in order to obtain
short waiting time
Z = reducing the variability of
High utilization but long arrival times and processing
waiting times
times in order to obtain a
Reduction of variability
higher utilization and short
Short waiting times
but low utilization waiting times

Utilization
Queuing Systems

• Queues Theory: development of models to

study the waiting phenomena which appears
in presence of a service demand
• Applications
 Customers waiting at the bank or at the post office
 Patients waiting for the doctor
 People waiting for a taxi
 Parts waiting to be processed by a machine
• Defined in 1917 by Erlang and applied in
telephonic sector
Queuing Systems

A queuing systems includes:

1. An arrival process
 Single jobs or batches
 Deterministic or stochastic arrival times
2. A production/service process
 Single or parallel machines
 Deterministic or stochastic process times
3. A queue
 FCFS (First Come First Served), LCFS (Last Come First
Served), SIRIO (Service In Random Order),….
 Limited or illimited length
Parameters

• ra  arrival rate (job/time) to the station. In a serial line

with no losses or reworks, ra = TH for each station
• ta  mean time between arrivals (ta=1/ ra)
• ca  coefficient of variation of arrivals (ca=𝜎a / ta)
• m  number of parallel machine of the station
• b  maximum number of jobs allowed in the system
• te  mean effective process time (workstation capacity
re=m/ te)
• ce  coefficient of variation of the mean effective process
time (ce= 𝜎𝑒 / te)

ra, ta, ca

te, re, ce
Performance measures

• Workstation utilization percentage:

• Waiting time spent in the queue: CTq

• Waiting time spent in the station:

• Mean work in process at the station:

• Mean work in process in the queue:

Kendall Notation

Kendal Notation describes the station with four parameters:

A/B/m/b
• A arrival times distribution
 D = deterministic (constant) distribution
 M = exponential (Markovian) distribution
 G = general (normal, uniform,….) distribution
• B process times distribution
 D = deterministic (constant) distribution
 M = exponential (Markovian) distribution
 G = general (normal, uniform,….) distribution
• m number of machines in the station
• b maximum number of jobs allowable in the system (i.e.
buffer dimension + number of pieces under process)
General system to n-state

Hypothesis:
• N = number of jobs in the system
• pn = probability that the number of jobs in the system is n
pn = P {N=n}
• nMAX = maximum number of jobs the system can have (this mean
that in a single machine, the queue is limited to a maximum of
nMAX-1 jobs, and 1 job is in processing)

Assumptions:
• Job inter-arrival times are independent on the status of the
system
• The server will never be idle when there is a job in the system
that can be served (the server is idle only when there are no jobs
available to process)
• Processing time is assumed to be exponentially distributed
General system to n-state

• Since the system can have at most nMAX jobs, there are
nMAX +1 possible states {0,1,…. nMAX} (representing the
number of jobs in the system)
• In a system, state passages are:
1. From n to n+1 if there’s an arrival: system passage from
state n to state n+1 is related to the arrival rate ra
2. From n to n1 if there’s a departure: system passage from
state n to state n1 is related do the production rate re
• We define:
 as the rate at which the system goes from state
n-1 to state n
 as the rate at which the system goes from state
n to state n-1
General system to n-state

• Assuming the steady-state exists (in order to have the

system balanced), the flow into and out of each state must
balance and this happens when the following equation is
verified:

That is,
General system to n-state

• If we write all the equations starting from state 1:

• Since pn values are probabilities, their sum must

be equal to 1:

That is,
General system to n-state

• From previous passages we have

and so


• The geometrical series converges to

if 0 < u < 1
• So, we have that and
M / M / 1 / ∞ model

Notation: M / M / 1 / ∞
Markovian Markovian Single Queue of
distribution distribution machine in unlimited
of arrival of process the station dimension (it
times times can be omitted)

Data:
• Mean time between arrivals: ta (arrival rate ra = 1/ ta)
• Mean effective process time: te (production rate re = 1/ te)
• Number of jobs present in the system: n
• The maximum number of jobs allowed in the system is
unlimited (i.e., an infinite number of states exists)
M / M / 1 / ∞ model

Hypothesis:
• pn = probability that the system is at the state n, namely that the
number of jobs in the system is n
• Jobs arrives one at a time, are produced one at a time and they
cause the change of state from n-1 to n or from n to n-1
Description:
• M/M/1 queue is the simplest one. It consists of a single machine
with exponentially distributed arrival times and processing times
 Exponential distribution is an advantage because of its memoryless
property: this implies it is possible to describe the queue state just by
using the number of jobs in the system and not the time the jobs
already spent processing.
 The state of the system, indeed, can be defined through one single
number n, that is the number of jobs
M / M / 1 / ∞ model – Performance Measures

• Work in process on the station (expected number of pieces in

the system) is the product between the number n of pieces in
each possible state and the probability pn for the system to be in
that state
M / M / 1 / ∞ model – Performance Measures

• Work In Process:

• Cycle Time:

• Queue Cycle Time:

• Queue Work In Process:

• Performance measures: Moreover:

 Increase as u and te increase
 Tend to infinity as u  1
Example

Consider a WS with the following parameters:

• ra = 2,875 jobs/hr, with ca=1 (exponential
arrival times distribution), thus ta = 20,87 min
• re= 3 jobs/hr, with ce=1 (exponential process
times distribution), thus te = 20 min

Compute the WS performance measures

Example – Solution

𝑟𝑎
• 𝑢= = 2,875/3 = 0,9583
𝑟𝑒
𝑢
• 𝑊𝐼𝑃 = = 23 jobs
1−𝑢
𝑊𝐼𝑃
• 𝐶𝑇 = = 23/2,875 = 8 hrs
𝑇𝐻

• 𝐶𝑇𝑞 = 𝐶𝑇 − 𝑡𝑒 =7,6667 hrs

• 𝑊𝐼𝑃𝑞 = 𝑇𝐻 ∙ 𝐶𝑇𝑞 = 22 jobs

G / G / 1 / ∞ model

Notation: G / G / 1 / ∞
General General Single Queue of
distribution distribution machine in unlimited
of arrival of process the station dimension
times times

Description:
• Arrival and process time has general and not exponential
distribution
• Without exponential memoryless property, we cannot use
the previous formulas
G / G / 1 / ∞ model

KINGMAN RELATION (VUT)

• By empirical considerations, it is possible to obtain an

approximated relation which relates M/M/1 and G/G/1
queues except for a variability coefficient
When ca = ce = 1, the formula is equal to the M/M/1 case
CTq (M/M/1)
G / G / 1 / ∞ model

• The other measures are obtained by the

usual formulas:
Example

Consider a WS with the following parameters:

• ra = 2,875 jobs/hr = 0,0479 jobs/min
• ca= 1 (exponential arrival times distribution)
• te = 20 min = 0,33 h
• ce=2,5 (not exponential process times
distribution)
• u = 0,9583

Compute the WS performance measures

Example – Solution

1+6,25 0,9583
• 𝐶𝑇𝑞 (𝐺/𝐺/1) = 𝑉 ∙ 𝑈 ∙ 𝑡 = 2 1−0,9583
∙ 20 =
27,7685 hrs

• 𝑊𝐼𝑃𝑞 = 𝑟𝑎 ∙ 𝐶𝑇𝑞 = 80 jobs

• 𝐶𝑇 = 𝐶𝑇𝑞 + 𝑡𝑒 = 28,1 hrs

• 𝑊𝐼𝑃 = 𝐶𝑇 ∙ 𝑇𝐻 = 81 jobs
M / M / m / ∞ model

Notation: M / M / m / ∞
Exponential Exponential m parallel Queue of
distribution distribution machines in unlimited
of arrival of process the station dimension
times times

Sakasegawa Relation:
When m=1, the formula is equal to the M/M/1 case
G / G / m / ∞ model

Notation: G / G / m / ∞
General General m parallel Queue of
distribution distribution machines unlimited
of arrival of process in the dimension
times times station

WHITT 1983:
CTq (M/M/1)

CTq (M/M/m)

VUT for parallel

machines
M / M / m / ∞ and G / G / m / ∞

• The other measures are obtained by the usual

formulas:
Exercise

A production line consists of a workstation composed of 3 identical

machines which work in parallel.
Jobs arrive to the line at a rate of 207 jobs/day. One day is
composed of 24 working hours.
Arrival times are exponentially distributed. The effective production
rate is equal to 3 jobs/hour. Also process times follow an
exponential distribution.

Identify the kind of queue corresponding to the workstation, and

evaluate its performance.
Compare the obtained performance with the case of three
separates queues, one for each machine. Which is the most
efficient solution?
Solution M/M/3
𝑟𝑎
ൗ𝑟𝑒
• u= 3
= 0,958
𝑢 2(𝑚+1)−1
• 𝐶𝑇𝑞 (𝑀/𝑀/𝑚) = 𝑚(1−𝑢)
𝑡𝑒 = 2,467 hrs

• 𝑊𝐼𝑃𝑞 = 𝑟𝑎 ∙ 𝐶𝑇𝑞 = 21,27 jobs

• 𝐶𝑇 = 𝐶𝑇𝑞 + 𝑡𝑒 = 2,8 hrs

• 𝑊𝐼𝑃 = 𝐶𝑇 ∙ 𝑇𝐻 = 24,14 jobs

Comparison 3 M/M/1 with M/M/3

3 queues system 1 queue system

M/M/1 M/M/3
Limited queues

• So far, unlimited queues had been analysed. Now we

focus, instead, on limited ones

• The system can contain b=B+1 jobs: B in the buffer and

one on the station
 Given that the number of jobs allowed in the system is limited,
the system allows only b+1 states:
Limited queues

• ra represents the potential arrival rate of a not stuck

system
• u = rate represents utilization the system would have
with no stops. It can be u>1.
• The system works with a TH equal to ra until it arrives
to state b where we have:
M / M / 1 / B model

Notation: M / M / 1 / B
Markovian Markovian Single Queue of
distribution distribution machine in limited
of arrival of process the station dimension
times times
Description:
• Consider the case in which process and arrival times are exponential, as
in a M/M/1 queue, but that in the system only b items are allowed
(instead of ∞)
• When the buffer reaches the maximum dimension (B), arrival process
stops and jobs are sent back
• ra = “potential” arrival rate, assuming the system is not at its maximum
capacity
• u = it is no longer the long term probability that the station is busy, but it
is the utilization the system would have if no jobs was sent back (u could
be >=1)
M / M / 1 / B model

• This happens, for instance, in a line composed by two

stations with a limited buffer between them. The first WS
processes raw materials and then stores them in a limited
inventory. The second one, instead, can be represented with
a M/M/1/B model.
• As previously said, the probability that the system is in the
state n is:

• The sum of probabilities is equal to 1 and must be limited to

b components, thus:
M / M / 1 / B model

The order b sum of the geometric series is:


Valid when u≠1
If u=1, then

(1+b) times

Namely, all states

have the same
probability.
M / M / 1 / B model

• If u≠1, WIP and TH of the system are:

• If u = 1, WIP and TH are:

M / M / 1 / B model

• Other measures are computed with the Little’s

Law:
Exercise 1

• Let’s consider a line composed of a single

station with interarrival times equal to 20
minutes (exponentially distributed) and
effective processing times equal to 21 minutes
(exponentially distributed).
• Compute performances with limited buffer B=3
Exercise 1 – Solution

• b = B+1 = 4 pieces
• ra = 1/20 = 0,05 pcs/min = 3 pcs/hr
• u = 21/20 = 1,05
• WIP = u/(1-u)-(5u5)/(1-u5) = 2,1
• TH = (1-u4)/(1-u5)ra = 0,039 pcs/min = 2,34
pcs/hr
• CT = WIP/TH = 53,8 min
• CTq = CT-te = 32,8 min
• WIPq = TH*CTq = 1,28 pieces
Exercise 2

• Consider a line composed of a single station:

 Interarrival time of 21 minutes, exponentially
distributed
 Effective process times of 20 minutes,
exponentially distributed
• Compare analytically the performances
a) With unlimited buffer
b) With limited buffer B=3
Exercise 2 – Solution

ta = 21 min, ca = 1, ra = 1/21 = 0,0467 pcs/min

te = 20 min, ce = 1, re = 1/20 = 0,05 pcs/min

u = 20/21 = 0,95238

M/M/1 M/M/1/4

• WIP = u/(1-u) = 20 • b = B+1 = 4 pieces

• TH = ra • WIP = u/(1-u)-(5u5)/(1-u5) = 1,9
• CT = WIP/TH = 420 min • TH = (1-u4)/(1-u5)ra = 0,039 pcs/min
• CTq = CT-te = 400 min • CT = WIP/TH = 48,8 min
• WIPq = TH*CTq = 19 parts • CTq = CT-te = 28,8 min
• WIPq = TH*CTq = 1,12 parts