IE

503 – Operations Analysis

Instructor - Narayan Rangaraj – IEOR, IIT Bombay
[email protected]
Teaching Assistants –
Santosh Palaskar – [email protected]
Kritika – [email protected]
and others
Super TA: Manoj Kumar – [email protected]
Lectures – Wed 930-1055, Fri 930-1055
Where – MS Teams – Team Code: q2gt2r3
 For whom is the course?
• Compulsory for IEOR M.Sc.-Ph.D. –
– Pl be in touch with me and the TAs

Aayushi Patil Isha Yadav

Rahul Vaishnav Rupesh Yadav
Abhishek Gupta Deepali Shrivastava
Aakash Roy Yannick Antonio Furtado
Vansh Chugh Shivam Mishra
Riya Sultana Shoaib Ahmad
Srija Mukherjee Anjali Bhagat

• Compulsory for IEOR M.Tech

• Minor course for some
• Additional learning for some
 Topics
• Location models [for plants, warehouses and other facilities]
• Layout models [for manufacturing facilities]
• Capacity planning
• Aggregate production planning [including MRP – Material
Requirements Planning]
• Principles of Just-in-Time systems
• Inventory models
– Deterministic [EOQ and Lot Sizing]
– Stochastic [Newsvendor, (R,Q) and(s,S) models]
• Scheduling
– Single machine sequencing
– Assembly line balancing
– Project planning
 After this course ...
After doing this course, you should have
• A basic idea of the major decision areas in
managing a modern manufacturing/service
facility
• Some of the main models in IE and OR that are
relevant to such decisions
• Some mathematical and computational principles
behind these models
• An idea of when and where to apply these
models
 Specifically, familiarity with
• Location and layout models
• Principles of materials requirement planning
and EOQ models
• Newsvendor model
• Basic scheduling theory applied to single
machine/resources and simple extensions
• An exposure to principles of modern
manufacturing like Just in Time, Quality etc.
Typical practical problem in Operations
Analysis
• A pharma plant has a production line with many
machines {1,2,3, …, m} – dryers, mixers, reactors etc.
– Why is the line defined this way?
• It makes many products {1,2, …, n} on the line
– Why many products on one line (why not one line per
product)? Why only these products?
• One ton of each product consumes some amount of
resource on each machine – this is given
• In a year, how much of each product should we make
on the line?
– When should we do it? Sequence and how often?
 Evaluation
• One end semester – 50 percent – Closed book
• Mid sem – 25 percent – Closed book
• Some number of quizzes – 7-8 percent each – Closed book

We plan to use SAFE exam for quizzes – details in next class

All complicated but relevant formulae/expressions will be

provided in question paper – no need for cheat sheet

Any cheating will result in grade penalty and reporting to the Dean AP’s office

No tutorials. Assigned reading, some assignments. TAs and I will be available at some
times for discussion on some topics, to be announced on Friday

TAs will do some grading of questions, but will not answer cribs about evaluation –
that is done only by me
 Background material and linkages with
other courses (IEOR students)
Modeling exercises in lab courses (IE 507)
– Especially formulation of decision models

Optimization models (IE 501)

– Linear Programming
– Network Flow models
– Integer Programming

Dynamic Programming
– Shortest Path Models
 B.Tech/BS/B.Des - DD students
You may have done works visits, internships,
projects etc. in some industry

You have done some courses involving modeling

and data analysis

No pre-requisites

Minor will be as per institute rules

 Reference material

• Material covered will be at the level of

Production and Operations Analysis –
S.A.Nahmias and co-authors, Wiley
• There are many books that are equivalent to this
and which cover most of the topics in this course,
e.g. – Chase Aquilano Jacobs – Buffa
• Very good reference books
– Factory Physics by Hopp and Spearman
– Matching Supply with Demand by Cachon and
Terwiesch
• Course notes
 Assignment 0 – problem 1

A new institute offers the following admissions every year

a) 100 students for a course with duration 2 years (e.g. M.Sc.)
b) 200 students for a course with duration 4 years (e.g. B.Tech)
c) 10 students for a course with (normal) duration 6 years (e.g. M.Sc.-Ph.D)
d) 30 students for a course with (normal) duration 5 years (e.g. Ph.D.)

1) In steady state, how many hostel rooms does the institute require,
assume all durations are deterministic?
2) Suppose c) students take 5-7 years and d) students take 4-6 years, how
will you plan capacity? You will have to assume something about the
uncertain duration.
3) For a)-d), past data is known regarding proportion of acceptance and
also male female split (which is different year to year). How do you plan
hostel accommodation for males and females separately, if desired?
 Assignment 0 – problem 2
On a (linear) stretch of beach, a vendor intends to set up a stall to sell food. Customers
are uniformly spread over the entire length L of the beach. Propose different criterion
on the basis of which the vendor could decide the location of the stall.
a) If the vendor wants to minimize total or average distance travelled by customers
where would she locate?
b) Customers accesses stall with probability p inversely proportional to distance to
stall. If we want to set up stall to maximize the expected number of customers that we
will get, where should we put it up?
c) If we can set up two stalls, where should we put them up?
d) If the two vendors are competing, how would they position themselves? Argue that
if one vendor is at location a in the interval [0,L], what would be the best location for
the other vendor, and thereby find the best location chosen by both vendors.
e) If the two vendors co-operate, then decide where they should set up.

Reference provided by Shardul Burde and Tanishq Awasthi

https://youtu.be/jILgxeNBK_8
But I suggest you first try and then watch the video.
 Basic location theory and
application to operations analysis
Narayan Rangaraj
IE 503 Lecture 2
Location decisions
• Where to locate factories?
• Where to locate warehouses?
• Where to locate a crane (when it is not in use?)
• Where to locate an elevator (when it is not in use?)
• Where to locate an ambulance depot?
• Where to locate a garbage dump?
• Where to locate the photocopying machine in the department office?
• Where to locate bus stops?
Factors influencing location
• Costs of moving people and material to some location
• Demand at some location, could be weighted demand
• Supply available at some points
• Cost per unit distance
• Route choice
• How is the material moved? What route do people select?
• Distance measure
• What are the possible locations (feasible locations)?
 Distance Measures
• Rectilinear distance(L1 norm) Pi = (ai, bi)

• d(X, Pi) = |x - ai| + |y - bi|

X = (x, y)

• Straight line or Euclidean distance (L2 norm) Pi = (ai, bi)

• d(X, Pi) = (x - a i ) 2 + (y - bi )2
X = (x, y)
• Squared Euclidean
• Chebyshev distance (L∞ norm) Pi = (ai, bi)

• d(X, Pi) = max{|x - ai|, |y - bi|}

X = (x, y)

• Minkowski distance
'
• d(X, Pi)= ∑ 𝑥 − 𝑎 % & (
 Distance measures - continued
• Rectilinear/Manhattan/Mannheim/Chandigarh
• Facility allocation in Grid like cities and warehouses (Look at maps of a few cities)
• Euclidian distances (as the crow flies)
• Squared euclidean also used
• Building of cell-phone towers
• Chebyshev distance
• Used to calculate the time taken for movement of overhead crane (crane can move on
the x and y axes at the same time but at the same speed along each axis)
• Network distance
Mannheim, in
Baden-Württemberg
Germany

From
Openstreetmap.org
Chandigarh

From
Openstreetmap.org
Example
• 5 demand locations on the plane, e.g. machines on a shop-floor
• P1 = (1,1), P2 = (6,2), P3 = (2,8), P4 = (3,6) and P5 = (8,4)
• New general purpose machine to be located on the shop-floor
• Number of demands (trips per day) from new machine to existing
machines
• 10, 20, 25, 20 and 25
• Cost of movement proportional to distance
• Where should new machine be located?
• In general, given locations (𝑎𝑖 , 𝑏𝑖 ) and weights 𝑤𝑖, where should we
locate the facility to minimize sum of weighted distance to all 𝑖’s?
 Solution concept for different distance measures
If distances are rectilinear distances, then solution by linear programming (LP)
or other algos
Basic idea: Min |𝑥| is equivalent to Min 𝑧, s.t. 𝑧 ≥ 𝑥, 𝑧 ≥ −𝑥
We can carefully replace all absolute value terms with such a construction
Simplified algorithms available which solve this LP

Existing locations at (𝑎3 , 𝑏3) , …, (𝑎4 , 𝑏4 ) – could be customers or service

locations or suppliers – Weightages 𝑤3, …, 𝑤4 for these locations

Objective for single facility location ∑% 𝑤% { 𝑥 − 𝑎% + 𝑦 − 𝑏% }

The solution is the median location, i.e. a co-ordinate where equal weight lies
on either side. Verify when this has multiple solutions.
Multiple distinct facilities
If multiple facilities have to be located at (𝑥3, 𝑦3 ), … , (𝑥:, 𝑦: ) – to be
determined – then let 𝑤%;be the weight representing movement
between existing 𝑖 and new 𝑗 and we use these weights in the above
expression to get
= 𝑤%; {|𝑥; − 𝑎% | + 𝑦; − 𝑏% }
%,;
Further if 𝑣;? be the weights representing movement between new
3
𝑗and new 𝑘, then we will add the term A ∑;,? 𝑣;? { 𝑥; − 𝑥? + 𝑦; − 𝑦? }

Note that these problems are separable in the 𝑥 and 𝑦 variables

Solution concept – cont’d
• If distances are squared euclidean, then calculus gives the centre of
gravity solution
𝑥 ∗ = Σ (𝑤𝑖𝑎𝑖) / Σ (𝑤𝑖) and 𝑦 ∗ = Σ (𝑤𝑖𝑏𝑖)/ Σ (𝑤𝑖)
Solution for optimal x and y co-ordinates can be computed
independently
• Not true for Euclidean distance – needs numerical iterative procedure
(try to formulate an algorithm for this)
For earlier example, solution with Rectilinear distance is (3,4), and
solution with squared Euclidean distance is (4.4, 4.7)
Try for other measures
Practical considerations
• All locations on the plane may not be feasible for location of facility
• Contour lines with identical costs
• Steiner locations allow for additional locations on the plane to be used to define a network
with minimal cost of construction (different from meeting weighted demand effectively)
• Do we look at average demand or max demand over time?
• For an emergency service, the max time away from a given demand location may be relevant
• Do we need to satisfy all demand?
• Do we need to locate one facility at a time? Why not all facilities together?
• Distinct facilities – e.g. several new machines, each with a specific purpose
• Easy to extend
• Identical facilities which are interchangeable
• e.g. many new service centres each of which can serve given demand equally well
• harder to compute (allocation also has to be decided)
 TOP STORIES / BUSINESS

BU SIN ESS

From dw.com Why are German chemical plants located near big cities?
(Deutsche Welle) à A fatal explosion at a chemical complex near Cologne has raised concerns about the park's proximity to one of Germany's
biggest cities. Chemical plants located near major cities are a common sight in the country.

Should big factories be near large

cities?

Often they are drivers of the cities

growth, but after some time, they
are inconvenient (and expensive) to
have in the midst of large
population centres.

Example: Textile mills in Mumbai

Even now, Mumbai has two

refineries, a fertilizer plant and a
The Chempark in Leverkusen is one of the largest chemical hubs in Europe
thermal power plant
An explosion at a chemical plant in Leverkusen, about a 30-minute drive from the western German city of Cologne, has killed at least two
persons and has left many more injured. Experts say the blast probably released toxic compounds into the surrounding area.
Big issue in managing safety, The tragedy has left many wondering what is a massive chemical complex doing so close to one of Germany's biggest cities, in the first place. In
pollution and logistics fact, other major German cities such as Mannheim and Leipzig also have large chemical complexes in their vicinity. 

The chemical industry in the Rhineland — a region of the German state of North Rhine-Westphalia — which includes the chemical park in
Leverkusen, is one of the largest hubs for the industry in Europe. The region has over 260 chemical companies employing more than 70,000
people. The firms are often clustered together in so-called chemical parks.
https://www.dw.com/en/why-are-german-chemical-plants-located-near-big-cities/a-58685101
From Wesseling in the south to Dormagen in the north: Global companies such as Bayer, ExxonMobil Chemical, Ineos, Covestro,
Lyondellbasell and Lanxess are located within a few kilometers of Cologne, a city with a population of over 1 million people.

To understand how a major city and a chemical park can exist in such close proximity, one has to look back at the long history of the Rhineland
as an industrial powerhouse. 

Carl Leverkus and the paint factory

Network problems
• Movements happen on a network
• We can locate at nodes on the network (sometimes along the arcs of the
network also)
• Demands at nodes (sometimes along arcs – e.g. garbage collection)
• Costs are proportional to shortest path movements on network
• For multiple facilities, combinatorial problem
• Median problems (minimized weighted sum) – regular facilities
• Center problems (minimize max distance from facilities) – emergency
facilities
• Obnoxious facility location problem – usually maximize min distance from
facilities
 Numbers on nodes are demands
Exercise Numbers on arcs are distances

One facility to be located anywhere

Try your hand on the following network on network i.e. on any arc
C

3
(4) Objective, to minimize sum of
weighted distance to all nodes
A 5
B
(5) (2) 4
Example: If on arc A-B, x units from A,
6 Weighted distance =
D (5)*(x) + (2)*(5-x) +
(3) (4)*(8-x) + (3)*(11-x)

Be careful in writing this expression

Try other objectives also, if you can for the other arcs, though
 IE503:‌‌Operations‌‌Analysis‌‌
   ‌
Location‌‌Problem‌  ‌
 ‌

The‌‌google‌‌map‌‌link‌‌(h
‌ ttps://goo.gl/maps/UZ6cSeb1F6mpnMoe8‌)‌‌shows‌‌the‌‌map‌‌of‌‌the‌‌  
centre‌‌of‌‌Mannheim‌‌town‌‌in‌‌Germany.‌‌It‌‌has‌‌a‌‌grid-like‌‌structure‌‌(this‌‌area‌‌is‌‌called‌‌the‌‌
 
Quadrate).‌‌    ‌

There‌‌are‌‌many‌‌restaurants‌‌of‌‌various‌‌cuisines‌‌in‌‌the‌‌town‌‌and‌‌offer‌‌some‌‌particular‌‌kinds‌‌of‌‌
 
service‌‌(dine-in,‌‌takeaway‌‌and‌‌home‌‌delivery)‌‌to‌‌the‌‌customers.‌  ‌

A‌‌few‌‌of‌‌these‌‌restaurants‌‌are‌‌considered‌‌for‌‌further‌‌analysis‌‌and‌‌the‌‌table‌‌below‌‌describes‌‌
 
their‌‌name‌‌and‌‌expected‌‌no.‌‌of‌‌weekly‌‌orders‌‌belonging‌‌to‌‌the‌‌category‌‌of‌‌home‌‌delivery‌‌in‌‌
 
this‌‌area.‌  ‌

Restaurant‌‌name‌  ‌ Weekly‌‌orders‌‌at‌‌lunch‌‌time‌  ‌

Lenok’s‌‌Burger‌  ‌ 22‌  ‌

Royal‌‌Donuts‌  ‌ 13‌  ‌

Little‌‌India‌  ‌ 9‌  ‌

Subway‌  ‌ 25‌  ‌

Surin‌‌Kuche‌  ‌ 7‌  ‌

Ponte‌‌Pizzeria‌  ‌ 18‌  ‌

Suppose‌‌that‌‌the‌‌customers‌‌are‌‌uniformly‌‌distributed‌‌over‌‌time‌‌and‌‌a‌‌single‌‌delivery‌‌  
operator‌‌in‌‌the‌‌town‌‌proposes‌‌to‌‌offer‌‌a‌‌common‌‌service‌‌to‌‌fulfil‌‌the‌‌demand.‌‌Construct‌‌an‌‌  
appropriate‌‌grid‌‌structure‌‌and‌‌locate‌‌the‌‌above‌‌restaurants‌‌to‌‌carry‌‌out‌‌the‌‌first‌‌cut‌‌analysis.‌‌
   ‌

Using‌‌the‌‌data‌‌and‌‌information‌‌given‌‌above,‌‌answer‌‌the‌‌following:‌  ‌

a)‌‌If‌‌the‌‌operator‌‌wants‌‌to‌‌minimize‌‌total‌‌or‌‌average‌‌distance‌‌travelled,‌‌where‌‌would‌‌he‌‌
 
locate‌‌himself?‌  ‌

Try‌‌the‌‌grid‌‌distance‌‌along‌‌the‌‌roads‌‌and‌‌the‌‌resulting‌‌solution‌‌and‌‌also‌‌the‌‌centre‌‌of‌‌gravity‌‌
 
solution‌‌and‌‌check‌‌how‌‌different‌‌they‌‌are.‌  ‌

b)‌‌If‌‌there‌‌are‌‌two‌‌operators‌‌located‌‌such‌‌that‌‌the‌‌demand‌‌of‌‌a‌‌restaurant‌‌is‌‌fulfilled‌‌by‌‌the‌‌
 
operator‌‌nearest‌‌to‌‌it,‌‌then‌‌comment‌‌where‌‌they‌‌should‌‌locate‌‌themselves.‌‌Note‌‌that‌‌we‌‌are‌‌  
not‌‌talking‌‌about‌‌dynamically‌‌allocating‌‌orders‌‌to‌‌the‌‌operators,‌‌but‌‌we‌‌do‌‌not‌‌decide‌‌a‌‌priori‌‌  
that‌‌so‌‌and‌‌so‌‌restaurant‌‌will‌‌be‌‌served‌‌by‌‌so‌‌and‌‌so‌‌operator.‌‌After‌‌the‌‌one-time‌‌  
calculation,‌‌the‌‌allocation‌‌is‌‌fixed.‌  ‌

 ‌

 ‌

 ‌
 More on location and
location-allocation problems
Location on networks
• p-median problem, to locate p facilities at nodes in a network, with the
objective of minimizing weighted distance to n given demand or service
locations (note that demand is at discrete locations on a network)
• For a large network can be tough, nCp possible combinations of locations
• Each combination will result in an allocation of demand nodes and then the
objective needs to be computed
• Generally speaking, no efficient way available to do a much simpler
computation (NP hard)
• Many good heuristics available
Location along arcs
• What if we can locate facilities along arcs (if we have the choice)?
• Check for simple networks (one arc, then two, then a triangle) that if
a location inside an arc is optimal, then one of the (neighbouring?)
nodes is also optimal
• Check on all (weighted) demands met through one end point of arc
and through the other end point
• Summary: It is enough to look at solutions at nodes
• Solutions on arcs may still be optimal
• For practical reasons, location at some nodes may not be possible
 Try the following

Try demands at A and B and some value of d (cost of arc A-B).

d

B 1 median problem here is relevant. Verify Hakimi’s theorem.

A

Try 1 and 2 median problems on this network with

different values of demands at A, B and C. Verify
Hakimi’s theorem.
d

B As per discussion in class, assumptions about how

A
demand is satisfied (e.g. how do customers react
when two or more facilities are accessible,
including tie breaking rules) are all relevant and
C different assumptions could give different
solutions.

For this network, also try a 1 median solution which is as far away as possible from any of the nodes, i.e. if x is the
chosen location, then min(d(x,A), d(x,B), d(x,C)) is as large as possible. This is the obnoxious facility location problem. Is
this also guaranteed to have a solution at a node? What about min(max d(x,A), d(x,B), d(x,C)) - the center problem?
 Median problems on networks – Hakimi’s
theorem
• A network is given and there are demands at nodes to be satisfied
• Demand at node j is hj
• If facilities are set up at points Xk = x1, x2, …, xk anywhere on the network, then every
demand point j will access the facility that is closest to it, call this distance as d(j,Xk),
i.e. distance from j to set Xk
• Distance to a point on an arc is measured through the end points of the arc (i.e. all movements
are on the network of arcs)
• The weighted accessibility function is then ∑ ℎ# 𝑑(𝑗, 𝑋) )
• The k median problem on the network is to find the set Xk which minimizes the
weighted accessibility function
• Hakimi’s theorem says that a solution to the k median problem on a network can be
found among the nodes of the network
• Verify from earlier problem that the 1-median problem always has a solution at a node
• Our beach problem had continuous spatial demand (there was no concept of a node)
1 and 2 median problems on networks -
example
https://web.mit.edu/urban_or_book/www/book/chapter6/6.5.2.html

The open courseware notes of Odoni and Larson’s book “Urban

Operations Research” has a section on Location on networks, including
a statement of Hakimi’s theorem and an example of 1 and 2 median
location

Integer programming formulations of this problem are possible and

quite effective
Practice problem – 6.5.2 in Odoni and Larson
• Try the example of section 6.5.2 in detail, so that you understand
Hakimi’s theorem (e.g. try with some values on the arcs C-D or C-F
and see the impact on the solution)
• Try the heuristic approach in the book
• Start with an arbitrary set of 2 medians and see if improvement works
to get the final optimal solution
• Note that the best 2-median solution does not contain the best 1-
median solution
 From Odoni and Larson – section 6.5.2

Numbers on arcs are distances/costs

Numbers on nodes are weights (demands)

If p facilities are located somewhere,

the objective is to minimize weighted sum of
demand satisfaction at all nodes.

i.e. Every demand node is assigned a facility

which will serve it and we minimize the
weighted sum of costs of service.

e.g.
In the 1-median case, cost if facility is located
at D is 43

In the 2-median case, cost if facilities are

located at C and D is 34
Plant/warehouse location problems
• In design of logistics networks, such as locating warehouses on a
distribution network, the following decisions are taken
• Where on a network to locate some facilities (e.g. warehouse)?
• This has a fixed cost, depending on location
• What are the allocations to all the service points?
• This depends on the set of location decisions made (i.e. nearest among the
locations selected)
• Total cost minimization of location and allocation by considering both
the above decisions together
• Can be attempted by enumerating different location possibilities
• For larger problems, some heuristics and exact procedures available
Solution methods – for p-median (and can
sometimes be useful for many other problems)
• Exact approaches
• Enumeration
• Integer programming solutions
• Heuristics
• Cluster first locate later
• Locate first cluster later
• Constructive procedures – may not work too well
• Improvement procedures based on interchange
• Meta-heuristics such as Simulated annealing, Genetic algorithms and variants
Location – allocation models
and moving to layout
Incremental solutions
• The locations for a best 2 facility solution may not be part of a best 3
facility solution
• But may be OK as part of a best 10 facility solution
• Many physical networks (e.g. distribution networks for a large
company) do evolve incrementally
• Long term view not always applied
• If we take 725 districts in India, weighted by population, and
distances as crows fly along a spherical surface
• What would be the best 1 location to put a response facility?
• What would be the best 2, 3, 4 and 10 locations?
Consequences of incremental location
• Is the best p+1 facility location solution better for weighted distance
than the best p facility location solution
• Yes
• We have seen that best p solution may not be part of best p+1 solution
• What about the extent of improvement in objective?
• Will generally decrease, but not in a structured way
• Suppose facility itself has a cost
• Example: fixed cost of setting up facility (plant, warehouse)
• Objective now becomes sum of fixed plus variable cost
• How to model this?
Plant location
• Where to locate a new refinery?
• Such large impact questions are often multi-objective decisions and
take a long time to make (and execute)
• Suppose we have 5 possible locations, A, B, C, D and E
• A is best for fixed cost (low land price)
• B is best for variable cost (good access to markets)
• C is best for reliable energy supply and environment
• D is most convenient for reliable supply of imported crude
• E may be the best compromise location!
Multi criteria decision making, Analytic Hierarchy Process etc.
Exact formulations for network facility design
• Integer programming
• 𝑦𝑖 = 0 or 1, indicating do not locate or do locate at 𝑖
• 𝑥𝑖𝑗 = 1, indicating 𝑗 is served from 𝑖 or not
• Variants possible (e.g. 𝑥𝑖𝑗 = fraction of 𝑗’s demand served from 𝑖)
• Linear equations and inequalities can express many things, especially
if we allow binary variables
• Useful in polyhedral approaches to solution (e.g. Linear Programming)
or search based techniques
Exercises
Using 𝑥, 𝑦, 𝑧 notation and integer restrictions, if required, and any
other constants
• Model the condition that no more than 𝑝 facilities should be located
• Model the fixed cost of the facilities that are selected
• Model the condition that all demand should be satisfied
• Model the condition that 𝑖 can supply 𝑗 only if facility at 𝑖 is open
• Model the condition that no more than 𝐾𝑖 units can be supplied from
facility at 𝑖
• Model the condition that 𝑖 can supply no more than 𝑘 locations
Pure allocation
• If locations are fixed, then what remains is allocation – what goes
where?
• Demand has to be met
• Can be only coverage constraint (ignore specific amount of demand)
• Can be minimize total/average cost of meeting all demand
• Generally, we will allocate so as to minimize costs ( e.g. transport
costs)
• Exercise – if no capacity constraints, then what is the best allocation?
• Exercise – if capacity constraints at chosen facility locations à
transportation problem in Linear Programming (will study in IE 501,
self study for others)
Solution techniques
• For rectilinear distances: Linear programming formulation or median
location (constructive procedure to have equal weight on each side),
can be done separately for each co-ordinate
• For square distances: Centre of gravity solution (formula)
• For Euclidean distances: Iterative procedure for 2 and 3 dimension
problems (for single dimension, same as rectilinear)
• Network distances: Heuristics or (mixed) integer programming
techniques
• See additional notes for follow up
Facility Layout
Introduction
Layout
• Detailed location of facilities with respect to each other and also
considering the size of the facility and detailed constraints on
feasibility
• Relative locations of all facilities
• Bulk manufacturing with multiple phases of manufacturing involve so
much material transfer that successive processes have to be adjacent
to each other
• For discrete manufacturing, especially job shops, there could be
multiple options available
 Facility Layout
Types of
layouts

Batch Cellular Mass

Process layout Product layout
production Manufacturing production
 Two extremes and things in-between
Process layout
• Facilities are grouped together according to their functions
• Good for customized products
Product Layout
• Machines and auxiliary services located according to the
processing sequence of the (dominant) product
• Good for standardized products
Many intermediate options, trying to achieve benefits of both
 Process layout Turning

Packaging

Milling

Assembly

Drilling

• Perhaps multiple machines of each type

• No dominant product that would define a flow
• Different products could have different routes through the facility
 Product oriented layout
Small number of high volume products that would define a flow,
and where relevant facilities would need to be adjacent to each other

Line 1 for product 1

Turning Milling Drilling Assembly Testing Packaging

Line 2 for product 2

Turning Drilling Assembly Testing Packaging
Facility Layout
Introduction
Layout
• Detailed location of facilities with respect to each other and also
considering the size of the facility and detailed constraints on
feasibility
• Relative locations of all facilities
• Bulk manufacturing with multiple phases of manufacturing involve so
much material transfer that successive processes have to be adjacent
to each other
• For discrete manufacturing, especially job shops, there could be
multiple options available
 Facility Layout
Types of
layouts

Batch Cellular Mass

Process layout Product layout
production Manufacturing production
 Two extremes and things in-between
Product Layout
• Machines and auxiliary services located according to the
processing sequence of the (dominant) product
• Good for standardized products
Process layout
• Facilities are grouped together according to their functions
• Good for customized products
Many intermediate options, trying to achieve benefits of both
 Product oriented layout (example)
Small number of high volume products that would define a flow,
and where relevant facilities would need to be adjacent to each other

Line 1 for high volume product 1

Turning Milling Drilling Assembly Testing Packaging

Line 2 for high volume product 2

Turning Drilling Assembly Testing Packaging
 Process layout Turning

Packaging

Milling

Assembly

Drilling

• Perhaps multiple machines of each type (a ‘department’ for each type of

machine)
• Many products made. No dominant product that would define a flow
• Different products could have different routes through the facility, i.e.
through the various departments

Call these departments T, P, A, D, M

 Bus stop example is similar!
Each bus stop is a department and instead of
all departments possible in all locations, we
Where to locate departments T, P, A, D, M? have 2 feasible locations for each bus stop.
Same principle as this (ie. Relative locations
results in a cost).

Example 1 Example 2

T D T *
T P P M A P
P T A * D M
A M One possible
Available Another possible
D A area allocation allocation
M D For each allocation of departments, the associated
item/people cost can be computed using the movement
Available One possible Another possible information
area allocation allocation
Appropriate distance/cost measure needed

For specific example, pl see note Could be multiple arrangements where the same cost is
accompanying lecture slides incurred, based on relative positioning of departments
Linear and U-shaped layouts
• If we have n facilities to locate in a product oriented layout, the
relative location is decided by the product flow (i.e. the sequence of
operations will determine the relative location)
• So the default layout is linear, in order of the operation sequence
• If the available area has a particular shape, can lay out the facilities in
a U shaped or other convenient shape, respecting the available area
• U shaped cells are preferred in manual systems because there is
better interaction between workforce members and movement
between stations (in case of re-work and any trouble-shooting) is
more convenient
The following simple layout problem has to be modelled.

A four-way intersection has 2 bus routes, number 61 and number 62, which cross each other at
the intersection (61 goes east <—> west and 62 goes north <—> south at the intersection).
Assume that buses drive on the left of the road (as in India). There is space to create bus stops
either before or after the junction, so the 61 E-W bus can stop at location J before the junction
or M after the junction. We would like to locate the bus stops so as to maximise the benefit to
passengers who wish to change buses (for example, as few people as possible should cross the
road for bus change).

North
G H
F I
West East
M J
L K

\ South

The buses are every ten minutes and staggered so that transfers both ways can take place.
Assume we have data about how many passengers generally change from 61 E-W to 62 N-S and
all other combinations. Here is the estimate of number of people per bus who wish to change
(the traffic is symmetric, when viewed for the entire day, ie. N-S people who wish to change to
E-W will be the same as W-E changing to S-N, etc.):

N-S S-N E-W W-E
N-S -- -- 8 12
S-N -- -- 5 10
E-W 10 12 -- --
W-E 5 8 -- --

For example, 5 people per S-N bus wish to transfer to an E-W bus on average.

Propose a model to answer this question in general.
Layout problem : Linear assignment and Quadratic assignment

Below is a layout problem posed almost as a location problem (the decision of

what goes where determines the relative locations and thereby the layout).

Data: The owner of a job shop has to locate four machines: (1) a punch press, (2) a
grinder, (3) a lathe, and (4) and a welding machine. Four locations are available for
this purpose - labeled (A), (B), (C) and (D).

The foreman has estimated the costs of material movement to and from the new
machines from the existing facilities in Rs per hour as follows

To and From (1), if at A, B, C, D : [940, 130, 620, 710]


To and From (2), if at A, B, C, D : [620, 190, 840, 960]


To and From (3), if at A, B, C, D : [750, 880, 180, 800]


To and From (4), if at A, B, C, D : [110, *, 810, 210]
The * is because the welding

machine cannot be at location B as it is too big.

Questions:

a. Determine the best choice of locations for machines A, B, C and D. One

suggested solution is to try lowest cost locations first; i.e. start with (4) at A, then
(1) at B, then (3) and C and (2) at D. This gives a cost proportional to 1380. The
optimal one proportionately costs 1140. Find this optimal solution.

b. Write out a mathematical expression for the optimization problem that achieves
this.
See if you can solve this. How will the formulation change if you have one
more location E available as a possibility?

c. From an enumeration point of view, how many possible ways are there to assign
facilities to locations in this sort of problem?

d. How do you think the costs of locating a facility at a particular site (A, B, etc.
are computed)? Here the foreman has given you this data, but you should know
how to get this.
More difficult problem:

A variant is the following. In addition to the movement between (1), (2), (3) and
(4) and existing facilities, there is movement between (1), … (4) among the new
machines also (e.g. between (1) and (3) and other pairs). This also has to be
included in the decision.

Suppose the distance matrix is given between the locations A, B, C, D as

follows:
A to B,C,D [1,1,2]
B to C,D [2,3] C to D [1].

Now suppose costs of item movement between the new facilities (1), (2), (3) and
(4) are as follows (in Rs per unit distance per hour – this factors in the
requirements per hour of jobs moving to and fro these facilities):


(1) to (2), (3) and (4): [50, 60, 60]

(2) to (3) and (4): [70, 30]


(3) to (4): [10].


Given this and the earlier data, formulate an appropriate cost function and
optimization problem and then determine the optimum location of the new
facilities (1), (2), (3) and (4).

Sample problems for IE 503 : set 1

1. Rectilinear distance location: office

In an office, a new service booth is to be located to serve some departments located at

co-ordinate locations (10,2), (10,12), (7,6) and (12,4). The average number of trips to the 4
departments are 6,10, 8 and 4. What would be a good location for the new copier, assuming
costs proportional to rectilinear distance?

Answer: Median location : co-ordinates (10,6) - keep adjusting one of the co-ordinates at a time
till half the weight is exceeded for the first time.

Notes: Check that the problem is separately solvable in x and y co-ordinates. Check for
alternate solutions. Check that for some range of values of the extreme co-ordinates of the
existing departments, the median solution would remain the same.

2. Rectilinear distance location: Shop floor: Slide 8, lecture 2

Answer: median location : co-ordinates (4.4,4.7)

Note: The constructive approach to the 1-median problem works, but is not really extendable to
p medians.

3. 1-median problem on networks: Slide 15, lecture 2

Answer: Node B.

4. 1 and 2 median problems on networks: Slide 4, lecture 3

The obnoxious facility location problem is NOT guaranteed to have a solution at nodes.
Typically such problems are not solved as weighted objectives (all demands are equally
important, but weighted versions are also possible). Single arc network with d = 2 and equal
weights on demands will give the unique solution on the mid-point of the arc.

5. 1 and 2 median problem on networks: 6.5.2 of Odoni and Larson

Solution given in the notes of Odoni and Larson.

6. Modeling for location on networks: Lecture 4, slide 6 (these models are not unique
and could vary depend on the notation used). Please read if interested.

Use notation y_i = 1 if facility at i and y_i =0 if not, and x_ij = 1 if demand at j is (fully) supplied
from i and x_ij = 0 if not.
Can also try with 0 <= x_ij <= 1 representing the fraction of j’s demand met from i.

Answer :

Let y_i = 0 if facility located at i, 1 if facility not located at i

& x_ij ∈ [0,1], the fraction of demand of j supplied by i

● No more than p facilities should be located,

∑ 𝑦𝑖 ≤ 𝑝
𝑖
● Fixed cost of facilities that are selected,

∑ 𝑐𝑖𝑦𝑖 = 𝑓𝑖𝑥𝑒𝑑 𝑐𝑜𝑠𝑡

𝑖
c_i = fixed cost at location i if facility located

Let d_j be the demand at j’th location

● All demands must be satisfied,

∑ 𝑥𝑖𝑗 ≥ 1 ∀ j
𝑖

● i can supply j only if facility at i is open,

𝑥𝑖𝑗 ≤ 𝑦𝑖 ∀ i,j

● No more than K_i units can be supplied from i,

∑ 𝑑𝑗𝑥𝑖𝑗 ≤ 𝐾𝑖 ∀ i
𝑗

Let x_ij ∈ {0,1}, indicating j isserved from i or not

● i can supply no more than k locations,

∑ 𝑥𝑖𝑗 ≤ 𝑘 ∀ i
𝑗
 Layout Decisions as
Location/Assignment Problems​
 Example
Owner of a shop has to locate 4 machines:​
• (1) A Punch Machine​
• (2) A Grinder​
• (3) A Lathe​
• (4) A welding Machine​
​
• 4 locations are available for this purpose labelled as : A, B, C, D​

• There will be one machine at each location

• Locate machines so that the total cost of material transfer to and from
the new machines is as small as possible
 The Cost matrix for movement is

Machine | Location​ A​ B​ C​ D​

(1) Punch Machine​ 940​ 130​ 620​ 710​

(2) Grinder​ 620​ 190​ 840​ 960​

(3) Lathe​ 750​ 880​ 180​ 800​

(4) Welding Machine​ 110​ **​ 810​ 210​

** indicates that location B is not available for the Welding machine

How do we determine the best choice of locations for machines
A, B, C and D ?
• Welding Machine (4) can not be located at B

• Various possible combinations, for example

1-A, 2-B, 3-C, 4-D
1-B, 2-C, 3-A, 4-D
1-C, 2-B, 3-D, 4-A

• How many ways are there to assign facilities to

locations?
• Possible No. of Ways = 24 – 6 = 18

• Let us try assigning low cost assignments first: gives (4) at A, then
(1) at B, then (3) and C and (2) at D

• This gives a cost proportional to 1380

• Is this best possible? Try interchanges, at least!

Approach 1:

• Try all possible combinations to get the solution.

• Compare all, and the minimum cost solution will be your Optimal
solution

Better methods possible

Approach 2:

• Model it as a Linear Minimum Cost assignment Problem

(Linear Programme)

• Solution method: for example, the Hungarian Method

• Specialized algorithm based on network simplex

• The optimal solution is

1-B, 2-A, 3-C, 4-D and minimum cost = 1140
𝑀𝑎𝑡ℎ𝑚𝑎𝑡𝑖𝑐𝑎𝑙 𝑀𝑜𝑑𝑒𝑙 𝑓𝑜𝑟 𝐴𝑠𝑠𝑖𝑔𝑛𝑚𝑒𝑛𝑡 𝑃𝑟𝑜𝑏𝑙𝑒𝑚
(for layout decisions)

• 𝐼𝑛𝑑𝑒𝑥 𝑖 𝑓𝑜𝑟 𝑚𝑎𝑐ℎ𝑖𝑛𝑒𝑠 ; 𝑀 = 1,2,3,4 𝑀𝑎𝑐ℎ𝑖𝑛𝑒 𝑠𝑒𝑡

• 𝐼𝑛𝑑𝑒𝑥 𝑗 𝑓𝑜𝑟 𝑙𝑜𝑐𝑎𝑡𝑖𝑜𝑛𝑠 ; 𝐿 = 𝐴, 𝐵, 𝐶, 𝐷 𝐿𝑜𝑐𝑎𝑡𝑖𝑜𝑛 𝑠𝑒𝑡

1 𝑖𝑓 𝑚𝑎𝑐ℎ𝑖𝑛𝑒 𝑖 𝑖𝑠 𝑙𝑜𝑐𝑎𝑡𝑒𝑑 𝑎𝑡 𝑙𝑜𝑐𝑎𝑡𝑖𝑜𝑛 𝑗

𝑥CD = E
0 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒

𝑎CD 𝑖𝑠 𝑐𝑜𝑠𝑡 𝑜𝑓 𝑚𝑎𝑡𝑒𝑟𝑖𝑎𝑙 𝑚𝑜𝑣𝑒𝑚𝑒𝑛𝑡 𝑓𝑟𝑜𝑚 𝑚𝑎𝑐ℎ𝑖𝑛𝑒 𝑖 𝑙𝑜𝑐𝑎𝑡𝑒𝑑

𝑎𝑡 𝑙𝑜𝑐𝑎𝑡𝑖𝑜𝑛 𝑗
 𝑚𝑖𝑛𝑖𝑚𝑖𝑧𝑒 𝑐𝑜𝑠𝑡 ∑C∈N ∑D∈M 𝑎CD ∗ 𝑥CD

𝑠𝑢𝑏𝑗𝑒𝑐𝑡 𝑡𝑜 𝑐𝑜𝑛𝑠𝑡𝑟𝑎𝑖𝑛𝑡𝑠

∑C∈N 𝑥CD = 1 ∀ j ∈ L

∑D∈M 𝑥CD = 1 ∀ i ∈ 𝑀

𝑥PQ = 0 – can also be handled by suitable cost aij

𝑥CD ∈ {0,1}
Relaxing the constraints 𝑥𝑖𝑗 ∈ {0,1} to 0 ≤ 𝑥𝑖𝑗 ≤ 1 and solving
as a Linear Programme (which can be done efficiently) works
in this case

How are the costs of locating a facility at a particular site (A,

B, etc.) computed?
In this case, it is because of item movement to and from
new facility to existing facilities in the factory

Here the foreman has given you this data, but you should
know how to get this
More general layout problems
• Suppose the distance between two locations is given

• Cost of transferring material per unit distance is also given

between machines (including the ones to be located now)
• We now have a quadratic objective
• Objective is to locate one machine at each location
considering minimum total material movement cost
• Distance between Locations in suitable units

• Cost of transferring material between machines (Rs/unit distance/hr)

𝑀𝑎𝑡ℎ𝑚𝑎𝑡𝑖𝑐𝑎𝑙 𝑀𝑜𝑑𝑒𝑙 𝑓𝑜𝑟 𝑞𝑢𝑎𝑑𝑟𝑎𝑡𝑖𝑐 𝑎𝑠𝑠𝑖𝑔𝑛𝑚𝑒𝑛𝑡
𝑓𝑜𝑟 𝑙𝑎𝑦𝑜𝑢𝑡 𝑝𝑟𝑜𝑏𝑙𝑒𝑚𝑠
• 𝐼𝑛𝑑𝑒𝑥 𝑘, 𝑙 𝑓𝑜𝑟 𝑚𝑎𝑐ℎ𝑖𝑛𝑒𝑠 ; 𝑀 = 1,2,3,4 𝑀𝑎𝑐ℎ𝑖𝑛𝑒 𝑠𝑒𝑡
• 𝐼𝑛𝑑𝑒𝑥 𝑖, 𝑗 𝑓𝑜𝑟 𝑙𝑜𝑐𝑎𝑡𝑖𝑜𝑛𝑠 ; 𝐿 = 𝐴, 𝐵, 𝐶, 𝐷 𝐿𝑜𝑐𝑎𝑡𝑖𝑜𝑛 𝑠𝑒𝑡

1 𝑖𝑓 𝑚𝑎𝑐ℎ𝑖𝑛𝑒 𝑘 𝑖𝑠 𝑙𝑜𝑐𝑎𝑡𝑒𝑑 𝑎𝑡 𝑙𝑜𝑐𝑎𝑡𝑖𝑜𝑛 𝑖

𝑥CY = E
0 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒

𝐶YZ 𝑖𝑠 𝑐𝑜𝑠𝑡 𝑝𝑒𝑟 𝑢𝑛𝑖𝑡 𝑑𝑖𝑠𝑡𝑎𝑛𝑐𝑒 𝑜𝑓 𝑚𝑎𝑡𝑒𝑟𝑖𝑎𝑙 𝑚𝑜𝑣𝑒𝑚𝑒𝑛𝑡

𝑓𝑟𝑜𝑚 𝑚𝑎𝑐ℎ𝑖𝑛𝑒 𝑘 𝑡𝑜 𝑚𝑎𝑐ℎ𝑖𝑛𝑒 𝑙

If k and l are at locations i and j, then the relevant distance is dij

 𝑚𝑖𝑛𝑖𝑚𝑖𝑧𝑒 𝑐𝑜𝑠𝑡 ∑C,D∈M ∑Y,Z∈N 𝑑CD ∗ 𝐶YZ ∗ 𝑥CY ∗ 𝑥DZ

𝑠𝑢𝑏𝑗𝑒𝑐𝑡 𝑡𝑜 𝑐𝑜𝑛𝑠𝑡𝑟𝑎𝑖𝑛𝑡𝑠

∑C∈M 𝑥CY = 1 ∀ 𝑘 ∈ 𝑀
∑Y∈N 𝑥CY = 1 ∀ i ∈ 𝐿
𝑥CY ∈ 0,1

This is much more difficult to solve to optimality than

the Linear Assignment problem
 Heuristic algorithms for quadratic assignment
• Constructive:
• Order the facilities in some way (e.g. the one which has most interaction
with others first)
• By some method, decide on a good location for each facility in list
• Improvement based:
• Start with some layout (assignment of facilities)
• Check on neighbouring solutions (e.g. interchanging the location of two
facilities)
• Accept improvements and continue till no more improvement possible
• More sophisticated methods possible
• Simulated Annealing
• Genetic Algorithms etc.
 Summary
• Layout can sometimes be thought of as assignment (of facilities to
specific location), which will determine relative location
• If costs of location of facility are independent of where other
facilities are, then it is a Linear Assignment problem – easy to solve
exactly (through Linear Programming and Network Flow models)
• Example: adding new facilities to existing layout, where new facilities do
not interact with each other
• If costs of location of facility depend on relative location of new
facilities, then it is a Quadratic Assignment problem – tough to solve
(for large instances) – heuristics allowed
• Example: De novo layout where all interactions are to be considered
• Techniques: Interchange and other heuristics
Planning problems in
layout
 Comparison of Different facility layouts
Characteristic Process Oriented Batch Production Mass Production Product Oriented

Type of Work Specialized Contracts Fabrication Assembly Mixing, Treating,

Refining

Type of Product Unique Made-to-order Made-to-stock Commodity

(Customized) (Standardized)

Product Variety Large variety High variety Low variety Very low or no variety

Flow of Product Unique for every Unique for every batch Sequenced Sequenced
product

Type of Customer One at a time Few individual customers Mass market Mass market

Product demand Infrequent Fluctuates Stable Highly Stable

Worker skills required Experts, Craftsmen Wide range of skills Limited range of skills Equipment Monitors

Task Type Non repetitive Non routine Routine Repetitive

 Characteristic Process Oriented Batch Production Mass Production Product Oriented

Production System Long term project Intermittent job shops Flow lines, Assembly Process Industry
lines

Production Equipment Varied General purpose Special purpose Highly automated

Flexibility Very High High Low Very Low

Adjacent Facilities Decided by machines Decided by the

that need to be close to commodity or product
each other

Material Handling Very Low Low High Very High

requirement

Production Control Difficult Difficult Moderately Difficult Easy

Inventory Control Difficult Moderately Difficult Moderately Easy Easy

Advantages Custom work, Latest Flexibility, Quality Efficiency, Speed, Low cost Highly efficient, Large
Technology Capacity, Ease of Control

Examples Construction, ship Machine Shops, Print Automobiles, Paint, Chemicals, Food
building, Aircraft Shops, Bakeries, Televisions, Computers, stuff
Manufacturing Education Fast food
 Main issue in product oriented layout
● Relative locations of facilities already determined (exact orientation can be
decided – often U shaped)
● Most often used for high volume production of standardized items
● Main issue is to balance the throughput of each stage
● Idea of bottleneck resource
● Smooth material flow through every stage, no accumulation between stages
● With multiple operations at each stage/machine/work station, for example,
assembly lines, then we need to do line balancing
 Main issue in process layout
● Many different technologies/departments/facilities
● Relative locations to be determined
● Multiple products each with a routing through the factory
● Depending on volumes of each product, we may decide on relative locations
to minimize total cost of item movement
● With several possible locations, we may think of it as an Assignment problem,
i.e. the decision is “what goes where”
○ If relevant costs can be computed once we decide on the location of one facility, then it is a
linear assignment problem – easy to solve
○ If we need to compute costs only after deciding on pairs of locations, it is a quadratic
assignment problem – much more difficult to solve, for large cases
Automobile - robotic assembly line

Source: indiamart.com
Automobile - manual assembly line

Source: toyotavn.com
A scooter production line
 Boeing assembly line

Boeing introduced a faster layer of automation into its 737 factory, adding a new
robotic system to drill holes in the main beams inside each wing

Source: assemblymag.com
Planning of Assembly Lines
Instance of Product-
oriented Layouts
 Flows and bottlenecks
Machine A Machine B Machine C Machine D

Processing capacity : Processing capacity: Processing capacity: Processing capacity:

24 items an hour 40 items an hour 12 items an hour 30 items an hour

Bottleneck
facility

Flow
 Shifting bottlenecks
• Bottleneck is a process that limits (for now) the output of a system
• Could be machines, departments or other facilities

• Improving system capacity requires us to find the bottleneck

• Bottleneck detection is tricky
• Bottlenecks are not static but may move around (as new facilities are added or
due to temporary changes in processing)

• Why is there an imbalance in the first place?!

• Because facility addition is sometimes chunky and because everyone likes to have more
capacity to work with
 Shifting bottlenecks
24 items an 40 items an 12 items an 30 items an
hour hour hour hour

Bottle neck
Flow

12 items an
24 items an 40 items an hour 30 items an
hour hour 20 items an hour
hour

Bottle neck

Flow
Assembly lines – an important example of
product oriented layouts
• Many operations in assembly
• Subassemblies
• Components
• Manual/semi-automated/automated/robotic
• Sequence of operations is important
• Example 1: Some fastening of internal components has to be done before
external cover attached
• Example 2: All operations on one side to be done before turning around
• Want to have unidirectional flow of the assembled product
• Save time
• Avoid confusion (what operations are done and what not done)
Work station
• A physical location where one or more operations are performed
• Fixtures have to be available to do all these operations
• All components/sub-assemblies have to be stored and made available
• If manual, then workers are trained and able to perform all the
operations
• Even within a workstation, there could be a sequence of operations
• All predecessor operations should have been done in prior
workstations – so that there is no back-flow of items
• Total work content of a work station is the sum of all operation times
Planning of assembly lines

Layout follows the processing sequence of the operations

• Operations have:
• Processing time
• Predecessor activities which need to be completed before
that operation

• Predecessor operations need to be done before but not

necessarily immediately before the operation
Throughput of an assembly line
• Think of each assembly as a project – set of linked tasks
• Operations are considered indivisible
• Predecessor operations means that certain things can happen
sequentially only – in one or more workstations
• When can the project be completed (by one person)? This is called
the makespan
• If we have multiple workstations, what is the maximum throughput
we can achieve?
• If the desired number of products per day is N and we have 8
hours (480 mins) to manufacture them, then we need a product
every 480/N minutes (call this number L).

• Version 1: L is given. The total work content of a workstation

cannot exceed L mins. We need a certain number of workstations
(arranged in sequence) with operations assigned to each
workstation – the number of workstations needs to be minimized

• Version 2: Number of workstations is given, say k. Find the

maximum throughput achievable for a given number of
workstations
For version 1

•The two main considerations in assigning operations to tasks are:

•Total operation time should not exceed assigned cycle time of

L minutes

•No operation assigned to workstation k should have

predecessor operations in workstation k+1 onwards

•One suggested heuristic for this is by using the Ranked Positional

Weights (RPW)
• RPW is the sum of the operation processing time plus ALL operation times for
which this operation is a predecessor (not just immediate predecessor).

• Ordering operations in decreasing order of RPW and then assign operations

to the lowest number workstation possible. This attempts to minimize the
number of workstations that are used.
• This is a variant or extension of the First-Fit-Decreasing (FFD) algorithm for
Bin Packing problems, which tries to find the smallest number of bins of
capacity C to accommodate n items with integer weights
• In the FFD, items are ordered in decreasing order of weights and then
allotted the first available bin

• Exercise: The same procedure can be used to find the smallest cycle time for a
given number of workstations, by successively trying different cycle times.
Example from Nahmias – 12 operations, total work content = 70 mins
Operation Immediate Processing time Ranked positional
predecessor weight
1 - 12 70
2 1 6 58
3 2 6 31
4 2 2 27
5 2 2 20
6 2 12 29
7 3,4 7 25
8 7 5 18
9 5 1 18
10 9,6 4 17
11 8,10 6 13
12 11 7 7
In the example, suppose the desired number of
products/day manufactured is 32

Total time available/number of products = 15 mins/product

i.e. The allowable cycle time is then L = 15 mins

Minimum no. of workstations required =

Total work content (mins)/cycle time (mins) =
70/15 = 4.67, rounded up to 5
A solution with 6 work stations is as below, using the RPW heuristic

Check that (for example), operation 6 is after 2 and 3 in RPW, but

cannot be added to station 2, because of overshooting the cycle time
and that all precedence constraints are satisfied

This also means that within a workstation, operations can be suitably

sequenced (e.g. in workstation 4, operations will be done in the order
7 – 8)

Is a solution with 5 workstations possible??

If the allowable cycle time is increased to L = 16 min
Minimum no. of workstations required =
= Total work content/cycle time = 70/16 = 4.37 (rounded up to 5)

A solution with 5 workstations is as below, using the RPW heuristic

• The quality of the solution is measured by the total idle time for
all the workstations in every cycle.
• For the given operation times, the smallest theoretical number
of workstations may or may not be achievable
Actual operation
• The smoothest operation of such an assembly line is a
transfer line, with products moving from one station
to the other every cycle time minutes.
• In steady state, there is no pile up of inventory
between workstations and material is ideally pulled
into the system at the first station when needed
• Within a workstation, operations can be done by the
operator in any sequence that respects
the precedence constraints
Example of Assembly Line
Balancing (from Nahmias)
 Operation list for (old) computer assembly
1. Drill holes in metal casing and mount brackets to 7. Mount hard disk drive. Attach hard disk controller
hold disk drives (12) and power supply to drives (7)
2. Attach motherboard to casing (6) 8. Set switch settings on the motherboard for specific
configuration of system (5)
3. Mount power supply and attach to motherboard (6) 9. Attach monitor to graphics board prior to running
system diagnostics (1)
4. Place main processor and memory chips on 10. Run system diagnostics (4)
motherboard (2)
5. Plug in graphics card (2) 11. Seal the casing (6)
6. Mount disk drives and attach drive controller and 12. Attach company logo and pack system for shipping
power supply to drives (12) to customer (7)

Operation time in minutes given

Try to construct precedence table and also network of operations – to see it as a project
Heuristics for Assembly Line Balancing
• Suppose cycle time is given and we want to minimize the number of
workstations
• Ranked Positional Weight heuristic
• Compute RPW (sum of operation time and time of all successive operations)
• Assign operations to first possible workstation so that cycle time is not
violated and precedence conditions not violated
• Many other heuristics possible
 Practice problem
Task Time Immediate • An assembly requires tasks with
(mins) predecessor times in minutes and immediate
A 10 -
predecessors given on the left.
B 11 A
• Draw the assembly graph for this
data.
C 5 B
• What is the minimum number of
D 4 B workstations required to get an
E 12 A output of 40 units per day
(assume 8 hour shift)? What
F 3 C,D about output of 32 units a day?
G 7 F • What is the maximum output
H 11 E one can get with 5 workstations?
With 6 work stations?
I 3 G,H
 Try first fit increasing and first fit
Bin packing example increasing algorithms on the
following problem

Generic first fit algorithm for n Pipes of length 12 available.

items, bin size = B
We need lengths of various sizes: 2
• Arrange n items in some order
of length 2, 4 of length 3, 3 of
• Go down the list and put item length 4, 1 of length 6 and 2 of
into first bin that can length 7.
accommodate it (i.e. capacity
has not exceeded B) – stop How many units of the size 12 pipe
when items are exhausted do we need? Try first fit increasing
and first fit decreasing – and any
other algorithm you can think of.
A network view of the design of
assembly lines
Additional points
Note that in steady state, all workstations will be operating simultaneously,
with different operations begin carried out on each of the assemblies

For example, when operations 5,6 and 9 (in some order) are being carried
out on assembly C in the third workstation, operations 2,3 and 4 (in some
order) will be carried out on assembly B in the second workstation

Verify that there is no back–flow of product

Within a workstation, operations can be scheduled in whatever feasible

order is convenient (this will always be feasible to do)
 IE503 - Solutions/Hints

Location Problem

The below 10 by 10 square grid resembles the town of Mannheim in Germany.

Restaurant index Restaurant name Demand Location

A Lenok’s Burger 22 (8,2)

B Royal Donuts 13 (8,5)

C Little India 9 (7,9)

D Subway 25 (4,1)

E Surin Kuche 7 (3,7)

F Ponte Pizzeria 18 (0,2)

1a) Objective is to minimize the total distance (rectilinear distance in this case) that needs to
be travelled to fulfil all the demand. Therefore,

min wi (|xi - h| + |yi - k|)

where, wi is the demand from restaurant i

xi is the x coordinate of restaurant i
yi is the y coordinate of restaurant i
(h, k) is the operator coordinates
However, there is a direct approach for the rectilinear distance case. It is to find the median
of x and y coordinates of supply location separately considering the demand as weights
(Ref. Production and Operations Analysis – S A Nahmias).

i) Rank the coordinates of supply locations in ascending order and accumulate the weights.

ii) The optimal value of the coordinate is found by dividing the total cumulative weight by 2
and identifying the first location at which the cumulative weight exceeds this value.

The optimal location for the operator to locate is (4, 2).

1b) Centre of gravity solution

∑𝑤𝑖*𝑥𝑖 ∑𝑤𝑖*𝑦𝑖
ℎ = 𝑖
= 4. 94 , 𝑘 = 𝑖
= 3. 19
∑𝑤𝑖 ∑𝑤𝑖
𝑖 𝑖

2) Apart from formulating this as an optimization model and solving it computationally,

‘cluster first locate later’ heuristic can be used to find the solution with reasonable efforts.

i) Divide the demand location into clusters (equal to no. of operators, 2 in this case), where
each cluster is served by one of the operators.

ii) The best possible location for both the operators can be calculated separately using the
same approach used in part (a).

iii) Check the feasibility i.e., whether the restaurant is being served by the nearest operator
or not. Calculate the total distance.

iv) The process is repeated by making changes in the cluster formation until we find the
optimal solution i.e., the solution with least distance.

Optimal solution:

Cluster 1: Restaurant A, B and C

Operator location: (8,2)

Cluster 2: Restaurant D, E and F

Operator location: (4,2)
Bus Stop Layout Problem

We aim to minimize the number of road crossings by the people who wish to change the bus
Objective Function Value : Total Number of Road crossings

We will use the heuristic based approach to solve this

So, First, we consider all the stops before the intersection, and from there we will try to
improve our objective value
So we have stops at L, F, H, J
● Some of the people in the bus at stop L (going in direction S-N) wish to go to either
West (stop at J) or East (stop at F)
Number of road crossings for the people who wish to go to West = 1*5
Number of road crossings for the people who wish to go to East = 1*10
● People in the bus at stop F
Number of road crossings for the people who wish to go to North (stop at L ) = 1*8
Number of road crossings for the people who wish to go to South( stop at H) = 1*5
● People in the bus at stop H
Number of road crossings for the people who wish to go to West (stop at J) = 1*8
Number of road crossings for the people who wish to go to East (stop at F) = 1*12
● People in the bus at stop J
Number of road crossings for the people who wish to go to North (stop at L )=1*12
Number of road crossings for the people who wish to go to South (stop at H)=1*10
Total Number of Road Crossings when stops are at L, F, H, J = 2(1*5+1*10+1*8+1*12) = 70
“Now at a time, we will change the position of the bus stop on one route and keep fix
for other routes, we will do this for all routes and we will see which change give us the
best improvement in our objective”
1. Change L to G (route S-N), Stops at G, F, H, J
Total number of road crossings : 69
2. Change F to I (route W-E), Stops at L, I, H, J
Total number of road crossings : 71
3. Change H to K (route N-S), Stops at L, F, K, J
 Total number of road crossings : 69
4. Change J to M (route E-W), Stops at L, F, H, M
Total number of road crossings : 71
From 1) and 2) we have better solution, so our current locations for bus stops are either G, F,
H, j or L, F, K, J, let's go with G, F, H , J
Now, we fixed the location of stop at G on route S-W and change location of stops on
other routes and if improvement happens we will take it as updates solution
1. Change F to I (route W-E), Stops at G, I, H, J
Total number of road crossings = 70
2. Change H to K (route N-S), Stops at G, F, K, J
Total number of road crossings = 68
3. Change J to M (route E-W), Stops at G, F, H, M
Total number of road crossings = 70
So from 2) we get a better solution so our new locations are at G, F, K, J
Now, we will fix the locations of stops at G and K and change the locations of the
other two routes
1. Change F to I (route W-E), Stops at G, I, K, J
Total number of road crossings = 69
2. Change J to M (route E-W), Stops at G, F, K, M
Total number of road crossings = 69

So, no improvement in objective value in this step so our final Locations for bus
stops are at G,F,K,J with total number of road crossings = 68

Machine Layout problem

1st part: where a location is to be assign to a particular machine

a) Using Hungarian method we can get optimal cost 1140

1-B,2-A,3-C,4-D

b) i) Mathematical model for the problem

Index i for machines ; M= {1,2,3,4} Machine Set
Index j for locations ; L= {A,B,C,D} Location Set

𝑥𝑖𝑗 = 1 𝑖𝑓 𝑚𝑎𝑐ℎ𝑖𝑛𝑒 𝑖 𝑖𝑠 𝑎𝑠𝑠𝑖𝑔𝑛𝑒𝑑 𝑡𝑜 𝑙𝑜𝑐𝑎𝑡𝑖𝑜𝑛 𝑗 ; 𝑒𝑙𝑠𝑒 𝑥𝑖𝑗 = 0

𝑎𝑖𝑗𝑖𝑠 𝑐𝑜𝑠𝑡 𝑓𝑜𝑟 𝑚𝑜𝑣𝑖𝑛𝑔 𝑚𝑎𝑐ℎ𝑖𝑛𝑒 𝑖 𝑡𝑜 𝑙𝑜𝑐𝑎𝑡𝑖𝑜𝑛 𝑗

𝑂𝑏𝑗𝑒𝑐𝑡𝑖𝑣𝑒: 𝑚𝑖𝑛 ∑ ∑ 𝑎𝑖𝑗𝑥𝑖𝑗

𝑖 𝑗
Subjected to constraints:
 ∑ 𝑥𝑖𝑗 = 1 𝑓𝑜𝑟 𝑎𝑙𝑙 𝑗 ∀ 𝐿
𝑖ϵ𝑀

∑ 𝑥𝑖𝑗 = 1 𝑓𝑜𝑟 𝑎𝑙𝑙 𝑖 ∀ 𝑀

𝑗ϵ𝐿

ii) When an extra E location is also available:

𝑥𝑖𝑗 = 1 𝑖𝑓 𝑚𝑎𝑐ℎ𝑖𝑛𝑒 𝑖 𝑖𝑠 𝑎𝑠𝑠𝑖𝑔𝑛𝑒𝑑 𝑡𝑜 𝑙𝑜𝑐𝑎𝑡𝑖𝑜𝑛 𝑗 ; 𝑒𝑙𝑠𝑒 𝑥𝑖𝑗 = 0

𝑎𝑖𝑗𝑖𝑠 𝑐𝑜𝑠𝑡 𝑓𝑜𝑟 𝑚𝑜𝑣𝑖𝑛𝑔 𝑚𝑎𝑐ℎ𝑖𝑛𝑒 𝑖 𝑡𝑜 𝑙𝑜𝑐𝑎𝑡𝑖𝑜𝑛 𝑗

𝑂𝑏𝑗𝑒𝑐𝑡𝑖𝑣𝑒: 𝑚𝑖𝑛 ∑ ∑ 𝑎𝑖𝑗𝑥𝑖𝑗

𝑖 𝑗

Subjected to constraints:

∑ 𝑥𝑖𝑗 ≤ 1 𝑓𝑜𝑟 𝑎𝑙𝑙 𝑗 ∀ 𝐿

𝑖ϵ𝑀

∑ 𝑥𝑖𝑗 = 1 𝑓𝑜𝑟 𝑎𝑙𝑙 𝑖 ∀ 𝑀

𝑗ϵ𝐿

∑ ∑ 𝑥𝑖𝑗 = 4 𝑓𝑜𝑟 𝑎𝑙𝑙 𝑖 ∀ 𝑀 𝑎𝑛𝑑 𝑗∀ 𝐿

𝑖ϵ𝑀 𝑗ϵ𝐿

c) No. of ways to assign are = 24 - 6 = 18

d) Machines are involved in machining of various products. To calculate the cost for
installation of a particular machine at a location, we first see for how many products
this machine is used, then we see the if the machine is at a particular location what is
its distance from all other machines, then total cost of machine for one product type
is calculated by multiplying quantity with the sum of distance from which product is
coming and to which it is going after getting machined from the particular machine.

2nd part: where cost of material movement from one machine to another machine is also
considered.

- Mathematical model for this: Similar to above one, but some additional cost that to be
considered is as follows:

𝑖𝑛𝑑𝑒𝑥 𝑘 𝑎𝑛𝑑 𝑙 𝑎𝑟𝑒 𝑢𝑠𝑒𝑑 𝑓𝑜𝑟 𝑚𝑎𝑐ℎ𝑖𝑛𝑒𝑠

𝑖𝑛𝑑𝑒𝑥 𝑖 𝑎𝑛𝑑 𝑗 𝑎𝑟𝑒 𝑢𝑠𝑒𝑑 𝑓𝑜𝑟 𝑙𝑜𝑐𝑎𝑡𝑖𝑜𝑛𝑠
 𝑐𝑘𝑙 𝑖𝑠 𝑐𝑜𝑠𝑡 𝑓𝑜𝑟 𝑚𝑜𝑣𝑖𝑛𝑔 𝑚𝑎𝑡𝑒𝑟𝑖𝑎𝑙 𝑓𝑟𝑜𝑚 𝑚𝑎𝑐ℎ𝑖𝑛𝑒 𝑘 𝑡𝑜 𝑚𝑎𝑐ℎ𝑖𝑛𝑒 𝑙
𝑑𝑖𝑗 𝑖𝑠 𝑑𝑖𝑠𝑡𝑎𝑛𝑐𝑒 𝑏𝑒𝑡𝑤𝑒𝑒𝑛 𝑙𝑜𝑐𝑎𝑡𝑖𝑜𝑛 𝑖 𝑎𝑛𝑑 𝑙𝑜𝑐𝑎𝑡𝑖𝑜𝑛 𝑗

Objective: 𝑚𝑖𝑛 ∑∑ 𝑑 𝑐 𝑥 𝑥 + ∑ ∑ 𝑎𝑘𝑖𝑥𝑘𝑖

𝑖𝑗 𝑘𝑙 𝑘𝑖 𝑗𝑙
𝑖 𝑘
𝑥𝑘𝑖 𝑖𝑠 1 𝑤ℎ𝑒𝑛 𝑚𝑎𝑐ℎ𝑖𝑛𝑒 𝑘 𝑖𝑠 𝑎𝑠𝑠𝑖𝑔𝑛𝑒𝑑 𝑡𝑜 𝑙𝑜𝑐𝑎𝑡𝑖𝑜𝑛 𝑖
𝑥𝑗𝑙 𝑖𝑠 1 𝑤ℎ𝑒𝑛 𝑚𝑎𝑐ℎ𝑖𝑛𝑒 𝑙 𝑖𝑠 𝑎𝑠𝑠𝑖𝑔𝑛𝑒𝑑 𝑡𝑜 𝑙𝑜𝑐𝑎𝑡𝑖𝑜𝑛 𝑗
Constraints will be same as above

As the objective function has two variables in the same expression, therefore it becomes a
quadratic mathematical model which is hard to solve.

Approach to solve: heuristic based solution can be used to solve this problem.
Improvement based heuristics :
➔ Start with some layout (assignment of facilities)
➔ Check on neighbouring solutions (e.g. interchanging the location of two
facilities)
➔ Accept improvements and continue till no more improvement possible

Bin packing Example

Quantity Length

2 2

4 3

3 4

1 6

2 7

We have pipe of length 12 available i.e. bin of size 12

First Fit Decreasing Method : Fit the item of large size first

Bin 1 7,4

Bin 2 7, 4

Bin 3 6, 4

Bin 4 3, 3, 3, 3

Bin 5 2, 2
First Fit Increasing Method: Fit the item of small size first

Bin 1 2, 2, 3, 3

Bin 2 3, 3, 4

Bin 3 4, 4

Bin 4 6

Bin 5 7

Bin 6 7

Optimal Solution

Bin 1 7 , 3, 2

Bin 2 7, 3, 2

Bin 3 4, 4, 4

Bin 4 6, 3, 3

● By FFD Method we need 5 pipes of length 12

● By FFI method we need 6 pipes of length 12
● 4 pipes of length 12 are sufficient to fill our demand, hence FFD and FFI are just the
heuristics, do not guarantee the optimal solution