Design of service

system
chapter 3
Prepared by: Dr.Suhad alnatoor
Near east university
• Services are a diverse group of economic activities that include high
technology, knowledge-intensive sub-sectors, as well as labour-
intensive, low skill areas.
• In many aspects, service sectors exhibit marked differences from
manufacturing – although these distinctions may be blurring
Design of service systems
• Integer programing (IP) models with binary variables, (0–1) variables
• Examples of service system design problems:
• Resource allocation problem
• scheduling of airline crews, and aircrafts
• Supply chain management
• location and distribution decisions in supply chains
• workforce planning
• setting of staffing at a call center by queuing models
Modeling with Binary Variables
• We use binary (0–1) variables in modeling real-world problems
• Linear programming (LP) models with binary decision variables are
called integer programming (IP) models.
• Of course, a general IP model may include regular integer variables
(non 0–1), as well as continuous variables.
• Such IP models are called mixed integer programming (MIP) models.
LP models with just binary and/or integer variables are called pure
integer programs.
Capital budgeting problem
• A company is planning its capital spending for the next T periods.
There are N projects that compete for the limited capital Bi , available
for investment in period i. Each project requires a certain investment
in each period once it is selected. Let aij be the required investment in
project j for period i. The value of the project is measured in terms of
the associated cash flows in each period discounted for inflation. This
is called the net present value (NPV). Let νj denote the NPV for
project j. The problem is to select the proper projects for investment
that will maximize the total value (NPV) of all the projects selected.
• $200,000 is the investment money, four potential projects
• The cost of each project, required land, and potential usage by county
residents are given in Table 3.1.
• The basketball court will be built inside the park. Hence, no additional
land space is needed. However, the basketball court cannot be built
unless the park project is selected for funding. The county has 15
acres of land. The commissioners want to select those projects that
will benefit the county residents the most (maximize daily usage)
subject to the budget and land constraints.
Capital budgeting problem
• Net present value is how much an investment is worth throughout its
lifetime discounted to todays value (NPV)
• It is the difference between the present value of cash inflows and
present value of cash outflows over a period of time and it is used in
capital budgeting and investment planning to analyze a projects
projected profitability
• Optimal solution: The optimal solution to this integer program was
obtained using Microsoft’s Excel Solver software and is given below:
• The optimal decision for the county is to build the swimming pool and
the park, with the basketball court. The facilities would be used by
1200 county residents daily.
Fixed charge problem
• Consider a production planning problem with N products such that
the jth product requires a fixed production or setup cost Kj ,
independent of the amount produced, and a variable cost Cj per unit,
proportional to the quantity produced. Assume that every unit of
product j requires aij units of resource i and there are M resources.
Given that the product j, whose sales potential is dj , sells for $pj per
unit and no more than bi units of resource i are available (i = 1, 2, …,
M), the problem is to determine the optimal product mix that
maximizes the net profit.
• Formulation
• Let the binary integer variable δj denote the decision to produce or
not to produce product j. In other words,
Note that xj
can be positive only when δj
= 1, in which case its production
is limited by dj
and the fixed production cost Kj is
included in the objective
function.
Quantity discounts (lower prices for large
volume purchase)

• 1. “All-unit” discount 2. “Graduated” discount

• “All-unit” discount
• Under this scenario, the entire purchase will be charged at a lower
price based on the order quantity
• The mathematical representation of the cost function f(X), where X is
the quantity purchased, is as follows:
• Equations 3.11 and 3.12 will force exactly one of the δi ’s to be one and the
others zero in any solution.
• Hence, only one of the Xi ’s will be positive in Equation 3.6. For example, if
δ2 = 1, (δ1 = δ3 = 0), then X2 will be positive between 201 and 500 by
Equation 3.9. Since δ1 and δ3 will be zero, X1 and X3 will be forced to zero
by Equations 3.8 and 3.10.
• Note that M is a large positive number in Equation 3.10.
• In case there is a capacity limit on the maximum order quantity, then M
can be replaced by that capacity limit.
• Note that all the constraints and the objective function are now linear
functions.
• All unit quantity discount models are commonly used by trucking
companies for the freight rate structure in road transportation
Graduated quantity discount
• the entire order quantity does not get the lower price; only the
additional amount over a price range gets the lower cost
• Under this model, the first 200 units are always charged at $7/unit,
the next 300 units are charged at a lower price of $5/unit, and only
the amount over 500 is charged at $3/unit. For example, if the order
quantity is 300 units, then the cost is given by
• Graduated quantity discount models are frequently used by utility
companies for energy usage
• Let δ1, δ2, and δ3 be the binary variables for each price range and X1,
X2, and X3 are the quantity purchased under price ranges 1, 2, and 3.
Then, the linear integer programming formulation becomes
• Unlike the all-unit discount model, one or more of the δi ’s can be one
in a solution. Hence, one or more of the Xi ’s can be positive and their
sum equals to the total order quantity X, as given by Equation 3.14.
• Note that X3 can be positive only if δ3 = 1 (Equation 3.19). When δ3 =
1, X2 ≥ 300 by Equation 3.18. Since X2 cannot be positive unless δ2 =
1, it will force δ2 = 1 by Equation 3.17. Now, both Equations 3.17 and
3.18 reduce to
• Hence, the only solution is X2 = 300, its maximum value under the
second price range.
• Similarly, when δ2 = 1, it will force δ1 = 1 and X1 = 200. Thus, for X3
to be positive (to buy at the lowest price), we have to reach the
maximum under the previous two price ranges, namely, X1 = 200 and
X2 = 300.
Set Covering and Set Partitioning
Models
• linear integer programs with binary variables.
• Applied to : warehouse location decisions, airline crew scheduling,
aircraft scheduling, supply chain network design, and package delivery
problems
• Consider an (m × n) matrix A, called the set covering matrix, whose
elements aij’s are either 0 or 1. If aij = 1, we say that column j “covers”
row i. If not, aij = 0.
• The set covering problem is to select the minimum number of
columns such that every row is covered by at least one column.
• Formulation:
• Binary variable for each column
• Xj = 1 if column j is selected and 0 otherwise, for j = 1, 2, 3, … n.
 guarantees that
the minimum
number of
columns are
selected to cover
all
the rows.

guarantee that
every row is
“covered”
by at least one
column, , for row i,
when at least one
aij = 1, the
corresponding Xj
must be one
Example 3.3
• In the set partitioning problem, every row has to be covered by
exactly one column
Warehouse location –set covering
problem
• warehouse locations as “columns” and the customer regions as
“rows” of the set covering matrix A with aij elements
• By including the cost of building a warehouse at location j as Kj , we
will minimize the total cost of building warehouses such that every
customer region can be supplied by at least one warehouse.
Example 3.4
• A firm has four possible sites for locating its warehouses. The cost of
locating a warehouse at site i is $Ki .
• There are nine retail outlets, each of which must be supplied by at
least one warehouse. It is not possible for any one site to supply all
the retail outlets as shown in Figure 3.3.
• The problem is to determine the location of the warehouse such that
the total cost is minimized.
Solution
• define the set covering matrix (A) based on the network configuration
• The rows of the matrix will be the nine retail outlets and the columns
will be the four potential warehouse locations.
• The elements of matrix A, aij, will be set to 1 if retailer i (Ri ) can be
supplied by warehouse location j (Wj ); that is, there is a direct link
between Ri and Wj . Otherwise, we set aij = 0
• For example, W1 can supply R1, R2, R3, R4, and R5. Hence, we set a11
= a21 = a31 = a41 = a51 = 1 and a61 = a71 = a81 = a91= 0
 Note that
several of the
set covering
constraints are
redundant and
can be omitted
before solving
the integer
program.
Application to airline scheduling
• Two important airline scheduling applications that use set covering
and partitioning models are as follows:
• 1. Fleet Assignment: assigning aircrafts to specific flights
• 2. Crew Scheduling: assignment of crews to specific flight

• “flight segments” instead of flight numbers to denote specific flights.

• the fleet assignment problem is to determine which type of aircraft
should be assigned to a specific flight segment in order to minimize
the total cost of operations.
• The crew scheduling problem is to determine the optimal allocation
of crews to flight segments to minimize cost, taking into account
several operational constraints.
Set covering and partitioning model
• For both fleet assignment and crew scheduling, a set covering matrix
A has to be developed.
• The rows of the matrix will be flight segments and the columns are
different aircraft or crew schedules depending on the problem
• The complete set covering matrix A is then developed by generating
all possible feasible schedules for the crews. Once all feasible crew
schedules are developed (matrix A), the crew scheduling problem can
be formulated as a set covering/partitioning problem
Crew scheduling model
• Given a set of m flight segments and n crew schedules, the set covering
matrix is given by an (m × n) matrix, denoted by A = [aij], where
• aij = 1, if flight segment is covered by crew schedule j and 0 otherwise.
• Let Cj denote the cost of crew schedule j.
• Equation 3.27 guarantees that every flight segment has an assigned
crew
• more than one crew to be assigned to a flight segment, one of the
crews is working in the flight and the other crew is traveling as
passengers.
• A set Partitioning model is used in case one crew is used
• For fleet assignment problem one aircraft is used, so we use a set
partitioning model
• While developing the feasible crew schedules (columns of the A
matrix), other operational constraints can be imposed as follows
• Minimum connecting time between flights segments • Sufficient
layover time for crews • Hours of flying time • Total length of a crew
schedule • Crews returning to their home bases within 1–5 days •
Other union rules
• In addition, if schedules 1, 2, and 3 in Table 3.3 refer to the same crew
out of New York, then the following constraint can be added
• X 1 +X 2+ X3 = 1
• so that only one of the three schedules is selected
• if there are only k crews available (k < n), then we can add the
following constraint:
Work force planning
• (Nurse Scheduling Problem)
• A hospital administrator has minimal daily requirement for nursing
personnel at different time periods as shown in Table 3.4. Nurses can
report to the hospital wards at the beginning of each period and work
for 8 consecutive hours. The hospital wants to determine the minimal
number of nurses to employ so that there will be a sufficient number
of nursing personnel available for each period.
• The nurse scheduling model can also be applied to workforce
planning problems in other industries
• The workforce scheduling model has been applied in several
industries. A partial list is given below:
• Airlines • Fast-food restaurants • Call centers • Retailers • Hospitals
• United Airlines successfully applied this model to improve the
utilization of personnel at its reservation offices and airports by
matching work schedules to customer needs more closely (Holloran
and Byrn 1986).
• Taco Bell developed an integrated labor management system to
determine the number of workers needed in its restaurants at
different times (Hueter and Swart 1998). Demand varied widely
during the day, with 52% of sales occurring at the lunch period (11
a.m. to 2 p.m.). First, a forecast of lunch sales over 15-minute
intervals, by time of the day and day of the week, was developed
using the Moving Average method. The forecast was then used as an
input to an integer programming model for estimating the labor
needs. Taco Bell achieved labor savings of over $40 million over 3
years by using the integrated labor-management system.