Forecasting and Strategic Inventory Placement for Gas

advertisement
Forecasting and Strategic Inventory Placement for Gas
Turbine Engine Aftermarket Spares
By
Joshua T. Simmons
Bachelor of Science in Mechanical Engineering
Purdue University, 1999
Submitted to the Department of Mechanical Engineering and the MIT Sloan School of
Management on May 11, 2007 in Partial Fulfillment
of the Requirements for the Degrees of
Master of Science in Mechanical Engineering
and
Master of Business Administration
In Conjunction with the Leaders For Manufacturing Program at the
Massachusetts Institute of Technology
June 2007
@2007 Massachusetts Institute of Technology. All rights reserved.
Signature of Author
7)
Department of Mechanical Engineering
MIT Sloan School of Management
May 2007
Certified by
David E. Hardt, Thesis Supervisor
Ralph E. and Eloise F. Cross Professor of Mechanical Engineering
Certified by
Xoy E. Welsch, Thesis Supervisor
Professor, Statistics & Management Science
Accepted by
Debbie Berechman, Executive' Director of Masters Program
tMIT Sloan School of Management
Accepted by
MASSACHUSET-1S INSTfTUTE
OF TECHNOL-oY
JUL 02E20E
ILMA2
Lallit Andhnd, Chairman, Committeeon Graduate Students
Department of Mechanical Engineering
BARKER
I
MITLibraies
Document Services
Room 14-0551
77 Massachusetts Avenue
Cambridge, MA 02139
Ph: 617.253.2800
Email: docs@mit.edu
http://libraries.mit.eduldocs
DISCLAIMER OF QUALITY
Due to the condition of the original material, there are unavoidable
flaws in this reproduction. We have made every effort possible to
provide you with the best copy available. If you are dissatisfied with
this product and find it unusable, please contact Document Services as
soon as possible.
Thank you.
The images contained in this document are of
the best quality available.
2
Forecasting and Strategic Inventory Placement for Gas
Turbine Engine Aftermarket Spares
Joshua T. Simmons
Submitted to the Department of Mechanical Engineering and the Sloan School of
Management on May 11, 2007 in Partial Fulfillment
of the Requirements for the Degrees of
Master of Science in Mechanical Engineering
and
Master of Science in Management
In Conjunction with the Leaders for Manufacturing Program at the
Massachusetts Institute of Technology
June 2007
ABSTRACT
This thesis addresses the problem of forecasting demand for Life Limited Parts (LLPs) in
the gas turbine engine aftermarket industry. It is based on work performed at Pratt &
Whitney, a major producer of turbine engines.
The author worked in the Global Materials Solutions program, Pratt & Whitney's latest
business venture, in which they will provide OEM quality materials for the CFM56
engine manufactured by CFM International. The new business venture required a
forecasting method that did not rely heavily on customer input or historic demand. A
forecast was developed using publicly available aircraft utilization history for the LLPs in
the aircraft. In addition, a methodology was proposed for the remaining parts that will be
studied in a future LFM internship. In addition, an inventory placement analysis was
completed for the GMS LLP supply chain.
Thesis Advisor:
David E. Hardt, Ralph E. and Eloise F. Cross Professor of Mechanical Engineering
Thesis Advisor:
Roy E. Welsch, Professor of Statistics and Management Science
3
ACKNOWLEDGEMENTS
I would like to thank the MIT Leaders For Manufacturingprogramfor providingsupport
in this project andfor the tremendous educationalopportunityprovided over the course
of two years as an LFM student.
In addition, I would like to thank the United Technologies Corporationand Pratt&
Whitney for proposing the researchcontained herein. Without the active support of
Larry Hosey and Jim Pennito, this project would not have been successful. I would also
like to acknowledge Tony Danburgand Maury Castanguay,the team of materials
controllersI worked most closely with, for their support of and interest in the project.
They made my time at Pratt& Whitney enjoyable and memorable.
Finally, I would like to thank my wife, Rachel,for her understandingof my time away
from home, encouragementin my studies, and my inspirationfor hardwork. Without
Rachel, none of this would be possible.
4
TABLE OF CONTENTS
ABSTRACT........................................................................................................
3
ACKNOW LEDGEM ENTS.................................................................................
4
TABLE OF CONTENTS......................................................................................
5
TABLE OF FIGURES........................................................................................
6
TABLE OF TABLES ..........................................................................................
7
Chapter 1: Introduction ........................................................................................
8
Chapter 1: Introduction........................................................................................
8
Chapter 2: The Gas Turbine Engine Business ...................................................
10
2.1: The Business M odel...............................................................................
10
2.2 Life-Limited Parts...................................................................................
11
2.3 Comm ercial Airline Aircraft Utilization..................................................
13
2.4: Strategic Perspective.............................................................................
15
Chapter 3: Forecasting ........................................................................................
17
3.1: The Pratt & W hitney Forecasting Process.............................................
17
3.2: Gas Path Parts ........................................................................................
20
3.3: Life Limited Parts .................................................................................
23
Chapter 4: The Aircraft Utilization Based Forecast ........................................
26
4.1: The Hypothesis ......................................................................................
26
4.2 Application to the PW 2000 Fleet.............................................................
30
Chapter 5: GM S Forecast and Inventory Control.............................................
44
Chapter 6: Discussion, Conclusions, and Recomm endations...........................
58
6.1: General Recommendations ...................................................................
58
6.2: Project Specific Recommendations ......................................................
60
BIBLIOGRAPHY .............................................................................................
63
5
TABLE OF FIGURES
Figure 1: A nnual A ircraft Landings..................................................................
14
Figure 2: Inputs to the forecasting process ......................................................
18
Figure 3: Scrap rate as a Markov Process........................................................
21
Figure 4: Hypothetical Demand Profile...........................................................
24
Figure 5: Demand From a Single Engine for a 15000 Cycle Part ....................
27
Figure 6: PW2000
28
3 rd
Disk Monthly Demand History....................................
Figure 7: Aircraft Accumulated Cycles ..........................................................
31
Figure 8: Aircraft S/N 23998 Utilization History.............................................
32
Figure 9: Life Limited Part Demand: Expected and Actual .............................
35
Figure 10: Expected and Actual Demand with Forecast .................................
37
Figure 11: 15000 Cycle Part Forecast Comparison..........................................
38
Figure 12: 20000 Cycle Part Forecast Comparison ...........................................
39
Figure 13: 30000 Cycle Part Forecast Comparison ...........................................
39
Figure 14: Forecast and Business Case Volumes .............................................
48
Figure 15: Scrap R ates ......................................................................................
49
Figure 16: Forecast and Business Case Volumes with Error...........................
51
Figure 17: The Life Limited Part Supply Chain ...............................................
53
Figure 18:
55
4 th
D isk Poissonness Plot................................................................
Figure 19: MRP Simulation Results ...............................................................
56
Figure 20: Inventory vs. Fill Rate Trade Off..................................................
57
6
TABLE OF TABLES
Table 1: Mean Absolute Percent Error at Various Stub Times .......................
34
Table 2: Forecast A ccuracy ............................................................................
40
Table 3: Stub Tim e Profiling Tool...................................................................
44
Table 4: Results of Inventory Placement Analysis ..........................................
54
7
Chapter 1: Introduction
Pratt & Whitney is a manufacturer of gas turbine engines for commercial and
military airliners. This thesis is largely concerned with Pratt's most recent business
venture, Global Material Solutions, in which Pratt will compete for the aftermarket gas
turbine engine components for rival's engine designs. For engines where Pratt is the
original equipment manufacturer (OEM), forecasts are driven largely by customer input
and demand history. For this new business, these traditional inputs are unavailable. Thus
a new forecasting method is required. This thesis presents a method for Life Limited Part
forecasting developed in cooperation with Pratt & Whitney as part of an MIT Sloan
Leaders For Manufacturing internship.
In Chapter 2, I will discuss Pratt's historic product and current market position. I
will expand and discuss the historic business model followed by the new business model
developed for the Global Material Solutions (GMS) business. In addition, I will
introduce the strategic concerns including discussion of the likely competitive response
of General Electric, the major OEM of the launch product for GMS. In Chapter 3, I will
introduce the problem experienced by GMS in forecasting demand for the new business
venture. I'll discuss Pratt's forecasting methodologies for current products and
demonstrate its lack of applicability for the GMS products. Finally, I will explain the
new forecasting methodology proposed for GMS.
Chapter 4 will discuss the application and validation of the proposed forecasting
methodology by applying it to spare components for which Pratt has owned the complete
market demand for the life of the product. In addition, the forecasts will be contrasted
with forecasts made by Pratt & Whitney for these parts, highlighting the benefits and
limitations of each method.
8
Chapter 5 of this thesis will begin with the application of the new forecast method
to the parts planned for manufacture by GMS. Based on this forecast, an inventory
analysis will be completed consisting of optimization of service level, strategic inventory
placement, and finished goods target stock levels. Finally, Chapter 6 will conclude with
recommendations for GMS.
9
Chapter 2: The Gas Turbine Engine Business
2.1: The Business Model
Pratt & Whitney, a United Technologies Company, is a manufacturer of gas
turbine engines for commercial and military aviation applications. For the purposes of
this thesis, I will concentrate on the commercial side of the business. Customers for these
gas turbine engines include airliner manufacturers and airliner operators. The largest
base of customers is the airliner operator. When an operator purchases a new aircraft,
they are generally presented with several powerplant options by the aircraft manufacturer.
The operator then selects the engine that best suits their needs in terms of performance,
reliability, initial cost, and lifetime operating costs. This has led to the common practice
among the gas turbine engine manufactures of selling engines as a loss leader, earning the
bulk of their profits not from the original sale of engines but from aftennarket sales of
parts and service. Pratt & Whitney generates about 60% of revenue from the aftermarket
business (Chenevert 3)
This loss leader practice has been successful for Pratt & Whitney in the past.
Since 1964, more than 13,000 JT8D engines and derivates were manufactured and sold
(JT8D 1). These engines powered the Boeing 727, Boeing 737 - 100/200, Boeing MD80, and McDonnell Douglas DC-9 aircraft. In 2005, there were nearly 10,000 of these
engines still in service and generating revenue through the need for spare parts and
services.
In the early years of any engine program, the engine manufacturer essentially
owns a monopoly on the sale of spare materials for its designs. As the installed base
grows, those aftermarket revenues become attractive to low cost competitors who reverse
10
engineer various parts and begin to compete for the aftermarket business. Historically,
the bolts, brackets, nuts, tubing, and other relatively simple parts are those first copied.
New entrants have lower costs than the OEM as their overhead costs are significantly
lower due to much lower engineering and research and development expenses.
Expansion into more complex parts follows some time later.
2.2 Life-Limited Parts
All the parts in a gas turbine engine operated in the United States must be
approved by the Federal Aviation Administration. Most other countries have adopted the
same or similar standards as those set out by the FAA. Two major classifications of parts
have been defined for gas turbine engines subject to this oversight. These classifications
define whether a part must be replaced on a usage basis, measured in hours or flights, or
on a condition basis.
Life-Limited Parts (LLPs) are those parts in the engine that must be replaced on a
preventive maintenance schedule. The parts defined as LLPs are typically the large
rotating elements in the engine. These parts are subject to fatigue failures and have
potential to take down the aircraft if they were to fail while the engine is rotating. In June
of 2006, a high pressure turbine disk on the number one engine of a Boeing 767 ruptured
during routine maintenance on the tarmac at LAX. The aircraft experienced significant
damage from debris including ruptured fuel lines and fire damage. The level of damage
experienced by the aircraft due to failures of this class of parts motivates the requirement
for replacement as a preventive measure rather than based on engine performance or
monitored condition.
11
Non life limited parts are those that wear at predictable rates and have very low
probability of failing in the catastrophic manner. For instance, the most frequently
replaced parts in gas turbine engines are turbine blades. Blade failure is rarely through
fatigue and fracture. Rather, the blades suffer material loss from the high temperatures
experienced during engine operations. The wear rates are predictable and performance
loss is measurable. The loss of performance motivates replacement by the operators long
before performance degradation would affect the safety of flying the aircraft.
Preventive maintenance schedules for LLPs are detennined by the original engine
manufacturer and define the useful life of the part in terms of number of cycles, or
landings, allowable before replacement is required. Specifications also include an
acceptable number of operating hours on a part acceptable before replacement is
required. However, cycles are usually the limiting factor. The cycle limitation is
determined based on the stress experienced by the part during the highest thrust periods
of operation, takeoffs and landings.
The parts that have been reverse engineered by competitors are subject to the
same Federal Aviation Administration oversight as those of the engine manufacturer.
The responsible companies are issued a Parts Manufacturer Approval (PMA) certification
which controls both product design and production processes (PMA 1). This is compared
to the Type Certificate held by the OEM which fully describes the engine (Design
Approval 1). Prior to Pratt & Whitney Global Materials Solution decision to enter into
the manufacture of LLPs for competitors engines, parts manufacturers had only attempted
to gain approval to make non-LLPs such as airfoils, brackets, and tubing.
This decision to enter the PMA marketplace represents a significant strategic shift
for Pratt & Whitney and a major event in the commercial airliner powerplant industry. It
12
took the aviation industry by surprise when Pratt announced in February, 2006, their
intention to begin manufacturing PMA parts for the CFM56-3. The -3 is one the worlds
most popular turbofans, powering much of the worldwide fleet of Boeing 737 airliners.
Under a tacit agreement, no OEM had entered the PMA space. In fact, they had publicly
decried the existing PMA companies, claiming there parts where not held to the same
standards of testing and manufacturing as OEM parts. (Flint 2007)
2.3 Commercial Airline Aircraft Utilization
The various business models of the world's airlines have significant impact on the
potential future revenues for the OEM engine manufacturers. For life-limited parts, the
highest price and highest margin products in an engine, cycle utilization (number of
landings per day) drives the sale of replacement parts. Thus, an airline with a business
model that requires frequent short flights will use a larger number of LLPs than an
operator flying a similar plane on longer routes for the same amount of time each day.
Figure 1 depicts the distribution of several aircraft annual rate of cycle accumulation. If
we consider a 15000 cycle limit part, a typical preventive maintenance requirement for
many Pratt & Whitney parts, on a CFM56-3 in a Boeing 737 operated by Southwest
airlines, making 6 trips per day and accumulating 2300 cycles per year, and an A320
operated by United Airlines which accumulates only 4 flights per day, its clear that the
requirement for new parts will be 50% higher for the -3 engine. Additionally, Pratt's new
engine sales are for aircraft such as the A380 that will most likely be used for long haul
flight and average only 2 or 3 cycles per day. It will be 15 years or more before sales of
Life Limited Parts are realized for these engines. The decision to focus on the larger
13
engines for larger aircraft has resulted in a much longer delay between the sale of new
engines and the realization of revenues from the aftermarket.
Aircraft Annual Landings
2000-
1800
1760
1600
1553
1400
1200
1000
-
800 -
600
67
-----
400
200
0
737
A320
757
Aircraft Model
767
747
Figure 1: Annual Aircraft Landings
In addition to long delay between the sale of new engines and revenue from sale
of spares, Pratt & Whitney's legacy products with large fleets of existing engines such as
the JT8D are in the twilight phase of the product lifecycle. That is, a large number of
these engines have been retired, additional engines are retired on a regular basis, and
those engines that are in regular use are not utilized as heavily as their modem, more
efficient replacements. As a result, sale of spare parts and service to those engines is in
decline. Thus, Pratt's motivation for entering the PMA business is clear. While Pratt's
future sales of new engines look promising, particularly when one includes the
developmental geared turbofan, targeted to power the next generation of single-aisle
aircraft, revenues from new engines will not be realized for more than a decade. The
14
PMA business offers Pratt & Whitney the opportunity to enter a market with an existing
installed base, capitalizing on the success of it rivals and enhancing its service and
support business with internally developed parts. If successful, these revenues will fill
the gap in over the next decade when sales from legacy products are depressed and sales
from new engines have not yet materialized.
2.4: Strategic Perspective
From a strategic perspective, of particular interest is the competitive response
from the other original equipment manufacturers. Will they enter into the PMA business
as well? Or will they continue to disparage the practice? What is Pratt & Whitney's
competitive advantage in this business? Low cost? High quality of service? Pratt &
Whitney's management did not disclose the answers to these questions during the course
of this project. However, one can expect General Electric's response to include several
avenues:
1.
Regulatory - GE will lobby the FAA to not approve Pratt & Whitney as
Parts Manufacturing Authority for Life Limited Parts on the CFM56
engines.
2.
Market response - GE will attempt to tie up customers with long term
contracts to provide materials. They will argue they cannot continue to
offer warranty or guarantee performance if engines are assembled with Pratt
made parts
3.
Competitive advantage - neither GE nor Pratt wants to compete on the basis
of price for this business. The impact of price competition will be to shift
value captured by the engine manufacturers to the airline operators. Instead,
15
GE will offer its parts as a package of service and materials, guaranteeing a
level of customer service to attract customers over Pratt's lower price points.
4.
Respond in kind - It is unlikely that GE will engage in manufacturing parts
for Pratt & Whitney engines. The same lull in expected revenues that
motivated Pratt's entrance into this business will deter GE from responding
in kind. However, Pratt has now set a new course for the industry. By
entering this market as a PMA for Life Limited Parts, Pratt opens the door
for the PMA's who are not OEM's to begin working in this space. When
these competitors gain approval to begin manufacturing Life Limited Parts,
it will add another dimension to the competitive landscape. I believe that
this battle will be fought on a quality of service battlefield as more and more
airline operators outsource their maintenance activities to focus on their core
competencies, providing air transportation.
It is unclear what the outcome of this new competitive environment will be. If
Pratt and GE begin to compete on price for the aftermarket business, the loss leader
business model will suffer. If the companies are no longer able to get the expected return
from the development of new engines in the aftermarket, they will attempt to extract
more value in the service sector or in the original sale of the engine.
Chapter 3: Forecasting
3.1: The Pratt & Whitney Forecasting Process
Like many manufacturing organizations, Pratt & Whitney relies heavily on
demand forecasts in scheduling the production of spare parts. At Pratt, long lead times
for production of parts and relatively short lead times promised to customers results in a
particularly heavy dependency. Accordingly, Pratt & Whitney manufactures spare
materials to stock, according to a forecast, and carries a safety stock of inventory to
protect against forecast error.
Lead times in the aerospace industry have been increasing due to under capacity
in the supply chain for raw materials. Over the last decade, high demand for raw
materials such as steel, aluminum, and titanium from the developing world as well as
traditional industries in the west has overwhelmed the bottlenecks in the raw material
supply chain (Pinkham 3). This results in long waits at the mills for orders of alloys and
is exacerbated by the high degree of specification required for gas turbine engine parts.
At Pratt & Whitney, these delays have resulted in typical spares inventory material lead
times increasing from 60 to 90 to 120 days. Lead times of more than six months are not
unusual in the aerospace industry. For the Life Limited Parts on the CFM56-3, the GMS
programs first engine, estimated lead times based on vendor quotes and current state of
the art processing at Pratt & Whitney's manufacturing facilities average more than 400
days.
While lead times for raw materials and the production of components has been
increasing, quoted lead times to customers has not. Pratt & Whitney's standard promise
to customers is seven days. This combination of quick delivery requirements with a slow
17
A.A
supply chain has placed considerable burden on the Spares Organization at Pratt to
increase the accuracy of their forecasts to avoid increasing inventory levels or decreasing
service levels. Additionally, as with many manufacturing organizations, there is pressure
to continually reduce the amount of inventory carried.
Most forecasting processes at Pratt & Whitney begin with the Shop Visit
Forecast. In a sense, this is the master forecast, from which all other forecasts are
derived. It is measured in engine visits to repair facilities per unit time. Of typical
interest is the number of visits expected over the next year and the expected distribution
of those visits across the year.
Inputs to the Forecasting Process
Historic Demand
Industry
Fuel Prices
World Events
Chapter 5 Lives
Aircraft Utilization
Statistical Analysis and Business Judgment
Customer Data
Overhaul Visits
Scrap Rates
Build Standar
Route Changes
New Repairs
Budget/Financials
Figure 2: Inputs to the forecasting process'
Figure 2: Inputs to theforecastingprocess, was created by Pratt & Whitney
18
Shop visit forecasts are developed with several inputs but key among them is
surveys of customers. Controllers in the spares organization spend much of their time
communicating with customers, obtaining data from maintenance and operations on
which aircraft engines are expected to be removed for overhaul over the next time period.
This is coupled with Pratt's own knowledge of the fleet including total number of engines
on wing, new service bulletins specifying redesigned parts, and the financial state of their
customers, to estimate the number and timing of visits over some future period. From
these inputs, a rate of engine inductions over the next 18 months is projected that includes
the number and intensity of engine overhauls expected.
Generally, the shop visit forecast is fairly accurate for mature fleets; mature fleets
being those with a large installed base and a comparatively low rate of new engines being
manufactured. The military spares organization claims an accuracy of within two engine
inductions per month for the Fl 17, the military version of the PW2000. While the
specific engines that are serviced may not match exactly to those that arrive, the quantity
of engines services is accurate.
Pratt uses the Shop Visit Forecast (SVF) as the basis for part level forecasts by
applying a Scrap Ratio. The Scrap Ratio is the fraction of engines serviced that are
expected to require that part. For example, the SVF for PW2000 engines in 2005 was
298 engines. The historic Scrap Ratio for part number 1A8707, a compressor blade, is
0.315. Thus the Pratt forecast for part number 1A8707 for 2005 is 298 x 0.315 = 94
blades.
Clearly there are two sources of error in this forecast. First is the error in the
SVF. While the military spares organization claims a high degree of accuracy, the data
tells another story. In 2005, there were 391 actual inductions. The forecast was 24%
19
low. Unfortunately, at the time of this project, Pratt did not maintain a record of historic
forecasts with which to do further analysis. Their practice is to update forecasts
regularly, effectively erasing the previous data points. In light of this project, a new
policy of capturing a snapshot each month of the forecasts and maintaining that record
has been established for future work.
3.2: Gas Path Parts
The second source of error is the scrap ratio. Standard practice is to use the actual
scrap rate from the prior year. This historic scrap rate is assumed to be constant.
However, interviews with representatives from the spares organization and the Cheshire
Engine Center reveal that this is not actually the case. The scrap rate evolution they
described is a Markov process.
Consider a ring of 12 identical airfoils in a hypothetical engine. The airfoils are
not required to be replaced on a time basis but have standards for replacement and repair
based on condition. The engine is released new into the field and has 12 new airfoils.
After 3000 flights, a typical service interval for a PW2000, the engine is brought in for its
regularly scheduled inspection service. Three of the 12 airfoils are determined to be in
unrepairable condition due to excessive material loss, five are sent out for refurbishment,
and the remaining four are deemed fit for continued service. The scrap rate for this first
shop visit as Pratt & Whitney measures it is 0.250 scrapped parts per visit. Returning to
the field are 3 new pieces, and 8 refurbished or reused pieces.
The engine returns to the field and is flown an additional 3000 flights and before
another regularly scheduled inspection. Of the 3 new pieces, one is replaced and two are
reused. Of the 9 2"n run pieces, 4 are refurbished and 5 are replaced resulting in 6 new
20
pieces, 2 2 "d run pieces and 4 3 d run pieces returning to the field. As per Pratt's
measurements, the scrap ratio for this visit is 50%. We can continue this example and
find that the scrap rates changes based on the status of the engine.
This probability chain is commonly known as a Markov Process and is
graphically depicted in Figure 3. It is characterized by a chain of discrete states and
discrete times when those states can change. Each time the engine returns to the shop for
service, there is some probability that an airfoil will advance from one state to the next.
There is also a chance that the airfoil will require replacement, thus returning to the base
state, a 1 't run airfoil. The probability of advancing and the probability of replacement
must total 100% for each service event.
Anecdotal descriptions of the scrap rates for airfoils indicate that the probability
of an airfoil advancing to 4 th run from 3 rd is less than the probability of advancing to 3 rd
run from
2 d.
Based on the scrap rates described in interviews, approximate steady state
scrap rates would be reached by the 8t' engine shop visit. This is equivalent to about
24,000 flights or 20 years for an average PW2000 installed on a Boeing 757. Thus, over
time, the error from the scrap ratio should diminish in the total forecast error.
1st
Run
2nd
Run
3rd
Run
4th
Run
Figure 3: Scrap rate as a Markov Process
Unfortunately, data from the engine service centers that could be used to validate
this theory was not available. While data concerning the number of airfoils consumed
per shop visit is captured, there is not a ready connection between the service event
21
information and the general age of the engine in cycles or shop visits in Pratt &
Whitney's databases. As such, developing a forecasting method for these parts has been
deferred. Instead, Pratt uses the most recent 12 months of history to estimate the scrap
rate. If the Markov Process model accurately describes scrap rates, during the early years
of the engine program forecasts to be less accurate than later in the product life cycle.
Pratt & Whitney specifies forecasts for parts such as these on a per shop visit
basis. That is, the forecast specifies the number of parts expected to be used by each
shop visit captured through contract by Pratt & Whitney. Some contracts are written for
parts only and others for parts and service. Nearly all of the airline operators engage in a
long term contract with a provider for parts, service, or both.
It is within the conditions of these contracts that Pratt is able to collect and
identify much of the information that it uses to develop forecasts. For instance, when
forecasting life limited part consumption, Pratt uses the current condition of the engines
which the airline operator expects to bring in for maintenance over a period of time to
estimate demand from that customer. That is for a given engine with 27 life limited parts,
the operator will provide a list with the current total accumulated cycles on the engines,
and the expected date and expected accumulated cycles on the engine parts at the time of
service. From these data, it is relatively easy to extract a forecast. Some assumptions
must be made regarding the operators minimum acceptable remaining life for any part in
an engine as it exits service, but these are typically shared by the operator with Pratt &
Whitney.
22
3.3: Life Limited Parts
Forecasts for Life Limited Parts are not so easily generated. Trending does not
yield adequate infornation for parts with long lives in the engines. Observation of recent
past might indicate that volume for a particular spare part is next to nothing and motivate
Pratt to carry only a small amount of inventory. However, as will be shown, this is
simply a matter of timing. By using what is known about the way operators are flying
aircraft, we can make some reasonable estimates of what the total volume of demand will
be over the lifetime of the aircraft and, based on current status of the aircraft, forecast the
timing of that demand.
Consider a 15000 cycle limit part in a fleet of 100 engines, introduced in one year.
The planes are flown, on average, 1000 cycles per year. With no variation from the
mean, the hundred engines will require 100 new parts in exactly 15 years. Introducing
random deviation from the mean annual cycles and cycles remaining at replacement
changes the demand profile as shown in Figure 4. Clearly, when working with a demand
scenario as depicted above, applying scrap information from the prior year will not
provide a good estimate of expected demand in the coming year.
23
Hypothetic Demand Scenario (15000 Cycle Limit Part)
70 -
60
- - -
-
50
-
-4-
40
-
20 -_-_-
10 -
0-
--
--
-
-
- --
-1
1
3
5
7
9
11
13
15
17
19
21
23
25
27
29
31
33
35
37
39
41
43
45
47
49
Year
Figure 4: Hypothetical Demand Profile
With the defined replacement conditions regulated for Life Limited Parts, one
would expect forecasting demand to be routine. If a part must be replaced every 15000
flights, one might think it is simply a matter of bookkeeping to estimate demand. This
would be true if the placement of engines and parts in the fleet were static. If an engine
were installed on an aircraft and that aircraft, engine, and parts combination remained
together throughout its life and the operators extracted all of the useful life from each
part, the forecast would only require estimating when the aircraft were to reach specified
intervals at which point a replacement would be required. However, the fleet is not static.
In the experience of managers at Pratt & Whitney, operators routinely make decisions to
pull parts from an engine based on build standards, strategic business concerns, and
unanticipated maintenance. In addition, spare engines are kept on hand. When an engine
is removed from an aircraft for service, spares are installed such that the plane is removed
24
-
from service only for the length of time required to remove and install an engine, and
complete any non-engine related service. The engine that was removed from the aircraft
for overhaul then becomes a spare engine after service is completed. Thus Pratt &
Whitney has difficulty knowing the current status of any particular engine.
To deal with this issue, Pratt relies heavily on customer input to estimate future
demand for life limited parts. Customers provide this information in a variety of ways.
Generally, at the beginning of a contract, current status of engines including the
accumulated cycles on all the LLP's is provided. Contracts are long term, typically 10
years or more and written for parts, service, or both. Over the course of the contract,
Pratt is not usually able to get updated engine condition reports. Instead they rely on the
customers anticipated maintenance plans including the engines they expect bring in and
their anticipated parts consumption. However, the data provided by customers is
typically limited to the next 12 months. With part lead times at in excess of one year,
Pratt becomes dependant on expediting and high levels of inventory to provide adequate
service levels.
The forecasting methods described above rely on two major data sources:
customer input and engine condition. This is a source of great difficulty for the GMS
program. This new engine program is intended to service customers who are currently
under contract with competitors and will provide parts to a mature fleet for which Pratt &
Whitney has no history. Thus the need for a new forecast methodology that relies only
on generally available industry information.
25
Chapter 4: The Aircraft Utilization Based Forecast
4.1: The Hypothesis
At the beginning of this project, we proposed the hypothesis that we could
estimate future demand based on aircraft status rather than engine status. The
accumulated cycles on an airframe is publicly available data. In addition, the preventive
maintenance schedules required for the engines that power the airframes is published in
Chapter 5 of the engine maintenance manuals as specified by the OEM in accordance
with the FAA. Pratt & Whitney has access to these engine manuals as a maintenance
service provider for competitor's products (most of the OEM's service each others
engines).
Airframe status data was collected through a subscription service to the Aircraft
Analytical Systems database provided by Flight (http://www.flightglobal.com). The
ACAS database captures data on a monthly basis and includes the monthly history of
cycle accumulation on all airframes manufactured in the west. The data includes the
engine model installed on the aircraft, current and previous aircraft operators, engine
utilization history in hours, and a variety of other airline market information.
Mathematically, the forecast model developed is very simple. If one assumes that
aircraft operators are basically rational and, as rational businesspeople, extract as much
useful life from life-limited parts as possible such that they are disposed of with near zero
remaining life, then the total fleet accumulated cycles divided by the cycle limits of a part
is the total expected historic demand for that part. The underlying driver for demand,
then, is the simply the number of cycles on the aircraft engine. That is to say, if we were
26
to follow the life of any single engine, and Life Limited Parts were replaced only at
expiry, the demand process for a 15000 cycle parts appears as in Figure 5.
Demand for a 15000 Cycle Limit Part
1.2
1
-
0.8
0.4
Accumulated Engine Cycles
Figure 5: Demand from a Single Engine for a 15000 Cycle Part
Summing a series of engines, which are introduced at varying times and
accumulate cycles at different rates results in the apparently complex demand pattern
seen by Pratt & Whitney material controllers (Figure 6). By identifying this underlying
driver of demand, or approximating it using aircraft accumulated cycles as a proxy for
accumulated cycles on a part, a model that will describe future demand is generated.
Forecasting models often use statistical analysis of historic data to estimate future
trends through various analyses. (Montgomery, Johnson, and Gardiner 8). Models are
used to simulate seasonality, product life cycles, linear trends, etc. However, the demand
model we will build is a transformation of another forecast. That is to say, the forecast
27
generated is of aircraft utilization, rather than demand. We then transform the forecast
mathematically as a causal model based on preventive maintenance requirements.
PW2000 3rd Disk Monthly Demand
E__
Year
Figure 6: PW2000 3 rd Disk Monthly Demand History
Of course, there are inefficiencies in the demand process that will yield error in
this forecast model. One such inefficiency is the early replacement of parts. Operators
are not allowed by regulation to fly with an aircraft that has a part beyond its preventive
maintenance limit. They schedule their maintenance with a buffer of time before the hard
expiration of the part. The life remaining on a part is referred to as the stub time. In
addition, parts do not have identical lives. To minimize total maintenance costs, an
operator might remove a part early when servicing the engine for other reasons, to avoid
bringing the engine in at a later date. This has the impact of bringing demand in early.
In addition to early replacement, there is cannibalization of parts from other
engines. After the terrorist attacks on September 1 1 th, world wide air travel fell by more
28
than 30% (Smallen 2002). Airline operators found themselves over capacity with more
available seat miles than the demand for air travel could support. They began a
rationalization program wherein many older, less efficient aircraft were retired before
their age would have necessarily dictated. By August 2002, they had successfully
reduced the available seat miles (avg number of seats x number of planes x number of
flights x avg miles per flight) by 6% compared to the prior year.
The MD80, powered by the Pratt & Whitney JT8D is one of the older, less
efficient aircraft that was affected by the rationalization program. Over the following six
years, Pratt has experienced significantly lower demand than expected. Operators were
using the engines and parts from retired aircraft to maintain the aircraft still in service. In
addition, a secondary market for these parts exists. When an aircraft is retired, if there is
remaining useful life in the components, they are removed and put up for sale. This has
the effect of delaying demand.
The secondary market for used components does not affect the net demand over
the lifetime of an engine fleet. Those components that are removed early from one
engine and used to satisfy a timed out replacement on another have potential to shift the
timing of demand. That is, the purchasing customer can delay the acquisition of a new
part indefinitely if enough used material is available. However, the supply of used
material is limited to either retired engines that have useful life remaining or engines that
are still in service. Clearly, the supply from retired engines is limited and will increase as
the fleet ages in a generally predictable manner. For engines that are still in service the
net effect of introducing components into the secondary market is to accelerate one
demand (the engine providing the part) and delaying another (the engine which accepts
the used component).
29
While the net effect of these errors might be to shift demand from one source to
another, or to delay demand through the use of components from retired engines, and
might be small when applied to an entire fleet of engines, there is potential for very large
errors if applied to small subsets. For instance, if we were to consider the behavior of a
major airline in its efforts to emerge from bankruptcy over the past several years, we
would find clear evidence of this difficulty. This airline made significant efforts to cut
costs immediately and retain as much cash as possible during its emergence from
bankruptcy. One of the sources of cash that they were able to utilize in this effort was
purchasing used parts from the secondary market for use in their engine overhauls. As
reported to Pratt & Whitney, the airline changed its maintenance schedules, reducing the
time between services significantly. By shortening these maintenance intervals, they
were able to use materials from the secondary market with low remaining cycles till
required replacement. The prices for these used materials are typically prorated from the
original purchase price based on the remaining life thus allowing the airline to conserve
cash. The consequence of using this decision is higher lifetime maintenance costs due to
an increased number of engine overhauls. Survival in a competitive environment at
sometimes necessitates suboptimal decisions. When transforning the utilization forecast
to a demand forecast, we will attempt to modify the transform to correct for these
inefficiencies.
4.2 Application to the PW2000 Fleet
As stated in Chapter 3, the goal of this project was to develop a method to
forecast Life Limited Part demand based on commonly available industry data rather than
the high labor content customer obtained data typically used by Pratt & Whitney in
30
forecasting. Key among the available data is aircraft utilization as measured in cycles per
time period. A cycle is defined as an aircraft landing. In 2002, United Airlines
scheduled 650,000 flights for its fleet of 554 aircraft, targeting an aircraft cycle utilization
of 1174 cycles per year.
100
-
908070 6050 -
50
30
20
10
0
-
Aircraft Accumulated Cycles
August 2006
Figure 7: Aircraft Accumulated Cycles
Through the ACAS database, utilization of individual aircraft is available. Figure
7 is a snapshot of the age of the PW2000 fleet measured in cumulative engine cycles in
August of 2006. Using the utilization history, one can estimate the future usage based on
the history for individual aircraft. The utilization history for a typical PW2000 powered
aircraft is shown in Figure 8. This Boeing 757-200 is owned and operated by Delta
Airlines, and is powered by twin PW2037's, the most common PW2000 derivative. The
usage pattern, xt, is modeled well with a basic 5 year moving average. That is to say, our
best estimate of xt is given by the equation x, = b + e, where our estimate of b is
31
iT
b
L x, and c, is normally disturbed, random error term at time t. There is a clear
T j
shift in the utilization of aircraft S/N 23998 likely resulting from the terrorist attacks of
September 11
.
Our moving average will over predict utilization following a shift such
as this. However, over predicting utilization by several hundred cycles per year will not
greatly affect the demand forecast because this over prediction is such a small fraction of
the accumulated cycles necessary to trigger a demand point. More sophisticated models
such as exponential smoothing were explored and can better detect a shift in utilization.
However, the simple moving average is computationally efficient and is effective for our
purposes.
Aircraft Utilization S/N 23998
2000
1800
30000
-_-_--
25000
1600
-
1400
U)
-20000
2
1200
-+- Utilization
1000
-5 Year Average
-*- Cumulative Cycles
15000 U
800
4
-_-
-
E
10000 E
U 600
-
400
-
-
--
5000
200
0
1990
0
1992
1994
1996
1998
2000
2002
2004
Year
Figure 8: Aircraft S/N 23998 Utilization History
With the expected future utilization of aircraft established, it is a simple matter to
generate the expected replacement of Life Limited Parts. As stated in Chapter 2, the
32
starting assumption is that parts are used completely. In addition, it is assumed that all
accumulated aircraft cycles must be applied to a life limited part and that parts are
replaced according to their regulated life limits. So, one would expect Aircraft S/N
23998 to have required a 15,000 cycle limit part in 1996 and a 20,000 cycle limit part in
2000. In addition, one expects an additional 15,000 cycle limit part and the first 30,000
cycle limit part to be required in 2007.
An expected demand history for the Life Limited Parts on the PW2000 for all
PW2000 powered aircraft was generated in this fashion and the results compared to the
actual demand history. The error remaining after this best fit between historic and actual
usage is error for which the source cannot be identified with current data. This error
likely comes from several sources, key among them the actual remaining life at time of
replacement. This in turn is driven by many of the factors addressed in Chapter Two
including the ability of operators to move engines from aircraft to aircraft, the presence of
spare engines in the fleet, and the management of maintenance operations. Some of these
sources tend to delay demand and others tend to accelerate it.
Completing the best fit of historic demand and expected demand based on aircraft
utilization history yielded the following results for Life Limited Parts on the PW2000.
33
Cycle Limit
Part Description
0
Estimated Stub Time
500
15,000
15,000
20,000
20,000
20,000
20,000
1 stg Disk
2nd Hub
16 stg disk
17 stg Disk
6th stg front Hub
7-15 Stg Drum
33%
61%
43%
41%
51%
44%
35%
86%
38%
46%
57%
46%
Fan Hub
3rd disk
4th disk
5th Disk
6th Disk
7th Disk
28%
28%
43%
24%
34%
25%
1000
40%
121%
41%
46%
42%
27%
709%4
v U
20,000
20,000
20,000
20,000
20,000
20,000
!L ~
Rear Hub
30,000
91%
L
76%
31%
44%
30%
28%
56%
30%
27%
32%
35%
26%
43%
29%
~~
41%
90%
Mean Absolute Percent Error (MAPE)
Table 1: Mean Absolute Percent Error at Various Stub Times
It's clear that some of the parts reviewed in this study are not well represented by
the expected historic demand model. The most striking examples are the 2-5 stage hub
and the Low Compressor Long Shaft. As it turns out, both of these parts have
experienced a significant service event in their history. The parts were showing signs of
failure earlier than expected and were redesigned by Pratt & Whitney. The entire fleet
that had been originally manufactured with the problem design was refitted with new
parts, and could not be introduced into the serviceable market. Therefore, the
assumptions regarding that efficient secondary market place that results in most life being
consumed before final disposition of the part, does not hold for these cases. Thus, these
parts have been dropped from further analysis. In addition, Life Limited Parts that were
not similar to those on CFM56-3 that Pratt & Whitney intended to manufacture were not
included in the analysis.
The total error expected for forecasts of the Pratt & Whitney parts will include
error from the sources estimated above as well as error in the estimation of future annual
34
cycle accumulation. This second source of error is quite insignificant compared to the
first and will be ignored in the application of the forecast model. The average annual
landings experienced by at PW2000 engine over the past 5 years is 1030 cycles with a
standard deviation of 6 cycles. Clearly, the 1% error from our estimate of the future rate
of cycle accumulation is insignificant compared to other sources of error.
There are two engine models within the PW2000 engine family, each specified by
the maximum thrust the engine is capable of producing. In the interest of simplicity,
clarity, and to protect the interests of Pratt & Whitney, I have aggregated the demand and
the parts for each of the separate models. To validate the forecast model, I applied the
forecast model to the PW2000 engine fleet accumulated cycle history and compared the
results to the demand history for those parts. As Pratt is the sole provider of Life Limited
Parts for the PW2000, they own the complete demand history.
All PW2000 LLP Demand
Expected and Actual
700
--
-_
-- Actual Demand
-G- Expected Demand
600
500
-
Correlation Factor = 0.91
Total Error = -15%
Mean Absolute Percent Error (MAPE) =
29%
Mean Absolute Deviation (MAD) 75 units
400
-
E
4)
0
U 300
-
-
200 - --
100 -
-
- -
--
0
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
Figure 9: Life Limited Part Demand: Expected and Actual
35
Figure 9 depicts the results of the application of the forecast method to historic
aircraft utilization. However, this is not a forecast. This is expected demand based on
the actual aircraft utilization rather than projected forecast utilization. The visual
correlation between the expected demand based on the aircraft utilization history and the
actual demand history is significant. The coefficient of correlation is calculated as 0.91.
Of particular concern are the large errors noted in 1998, 1999, and 2001. We will attempt
to identify the sources of these errors and, if possible, correct the model to account for
them.
In addition to the difference between the expected and actual demand, the overall
shape of the demand curve is of interest. A question that might occur is what caused the
decline in demand for Life Limited Parts from the PW2000 fleet beginning in 2001.
When the product managers at Pratt were interviewed regarding this curve, they stated
that this reflected the life cycle of the fleet and events in the industry. The fleet was new
and growing in the early 90's and demand grew with it till it reached maturity in
1999/2000. At this time, production of new PW2000 was reduced as production of the
Boeing 757 tapered off. 2001, 2002, and 2003 experienced depressed demand due to
world events including the terrorist attacks of September 11, 2001 and the medical scares
of SARS and the avian flu. The question that we posed was whether our model, which
reflected depressed demand based on the actual aircraft utilization during these years,
have predicted reduced demand given data available only through the years 2000.
To test the robustness of the methodology and to determine if the manager was
correct in believing that the changes in demand were due to events in the airline industry
that could not be anticipated, or if the depression was predictable and the result of the
status of the fleet, data from December 2000 was loaded into the model. Aircraft
36
accumulated cycles as of the end of December were used as a starting point. The rate of
accumulation of cycles was modeled as the average of the previous 5 years of utilization
for each aircraft. The resulting output can be seen in comparison with the actual demand
and the expected demand based on actual utilization in Figure 10.
All PW2000 LLP Demand
Expected and Actual
700 --
--
- -600
500-
--4-*Actual Demand
-- Expected Demand
Dec. 2000 FCST
Correlation Factor = 0.91
Total Error = -15%
Mean Absolute Percent Error (MAPE) = 29%
Mean Absolute Deviation (MAD) = 75 units
400
E
0
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
Figure 10: Expected and Actual Demand with Forecast
The yellow line in the Figure represents the forecast generated through the aircraft
utilization model for total LLP demand for the PW2000 made using December 2000 data.
The model predicts significantly lower demand in 2001 than in 2000 indicating that the
depressed demand experienced during this time was not a result of industry events and
decreased aircraft utilization but was a product of the status of fleet at the time. Of
particular interest is the success in forecasting the lowest demand year in 2004. In the
words of the PW2000 spares manager, "We're pretty good at forecasting when we're
37
following a trend, but we get in trouble when we have to turn a corner." The value to this
manager in forecasting the timing of swings in future demand is significant.
This is one example of an apparent success in forecasting using aircraft utilization
and a generic profile of parts replacement. It is important at this point to develop a frame
of reference for success or failure in forecasting for these parts. Figures 11 - 13 compare
forecasts made using the aircraft utilization based method with forecasts generated for the
same classes of parts by Pratt & Whitney at the same point in time. Unfortunately, Pratt
has not made a practice of maintaining forecast records for posterity so there are only a
few points in time that can be compared. However, the results indicate that the utilization
based forecast is comparable in quality to those forecasts generated with heavy customer
input.
PW2000 15000 Cycle Limit Parts
120
-
+15000
cycle part average demand
-E- Utilization Based Fcst 2003
-A Pratt Forecast - 2003
100 -
-0- Pratt Forecast - 2004
80
-N-
Pratt Forecast - 2005
-*-
Pratt Forecast - 2006
E 060-
40
202Forecast Based on 2003 Data
0
1999
2000
2001
2002
2003
2004
2005
2006
2007
Figure 11: 15000 Cycle Part Forecast Comparison
38
2008
2009
PW2000 20000 Cycle Limit Parts
100
-4--Actual Demand
Pratt Forecast - 2003
Pratt Forecast - 2004
Pratt Forecast - 2005
Pratt Forecast - 2006
Utilization Based 2003
90
80
-
70 -
-E
60
-
50
-
'U
E
0
40 30 20 -
IForecast
10 -
Based on 2003 Data
C::
0 1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
Figure 12: 20000 Cycle Part Forecast Comparison
PW2000 30000 Cycle Parts
454035 -
-'4-30000 cycle part average dem and
-UUtilization Based Fcst 2003
--Pratt Forecast - 2003
-Pratt Forecast - 2004
-*- Pratt Forecast - 2005
-#-- Pratt Forecast - 2006
30-
E
250-
U) 2015105Forecast Based on 2003 Data
.
.
0-1999
2000
2001
2002
2003
.2
2004
2005
.2
2006
2007
2008
2009
2010
2011
Figure 13: 30000 Cycle Part Forecast Comparison
39
Pratt & Whitney uses two measures of forecast accuracy that will be familiar to
those readers involved in forecasting. First is Mean Absolute Deviation (MAD). This is
the average difference between the forecasted value and the realized value for the period
of interest. Second is Mean Absolute Percent Error (MAPE). This is related to MAD but
presented as a percentage with the actual value as the basis for a percentage measure,
making comparison of accuracy between parts easier. The following Table compares the
accuracy of the utilization based forecast with the forecasts generated by Pratt & Whitney
from the previous figures. In the figures above, only a single utilization based forecast is
presented for clarity. In the calculation of MAPE and MAD, forecasts were made using
data available through 2003, 2004, and 2005 to ensure that the comparisons are well
founded.
iparison,,
nnt Tools,
15000
20000
36%
30000
40%
, ,
15000
20000
30000
26%
25%
44%
15000o
20%
35000
30%
10
8.4
9
19%
17%
34%
14%
14%
13%
1
36%
1
15
5
5
atdtb
ddo ,"
,M
MADutiliz.tior
16
7
9
4
12
9
2%
14
29%
14
12
Table 2: Forecast Accuracy
One can see from the table that the aircraft utilization based is neither better nor
worse than the forecasts made by Pratt & Whitney in terms of the raw accuracy numbers.
However, there are several things to note in observing the results of the two methods.
40
First is the general shape of the utilization based forecasts (and historic expectation) as
compared to realized demand. When compared to the Pratt & Whitney forecasts, it
seems that the UB forecasts reflect the shape of actual demand better. From the position
of a materials manager for a part a good forecast of the shape can be very valuable. As
the materials manager at Pratt said, "we're not very good at turning corners." A forecast
that predicts major inflection points in the demand profile can go a long way to
improving Pratt's performance in this area.
In addition, you will notice that the error terms in Table 2 are lower than those
estimated based on the historic best fit. This is a result of the aggregation of parts with
the same life limits for purposes of protecting Pratt & Whitney's private data. By
aggregating, the errors tend to cancel out yielding a more accurate, albeit less useful,
forecast.
The next observation concerns the falling nature of Pratt & Whitney's forecasts.
In almost every case, Pratt's year over year forecasts decline. That is, the forecast for
two years out is less than the forecast for next year. This is an artifact of the forecasting
system. The MRP system at Pratt is set up to automatically make a projection for
demand based on time weighted average of historic demand. The materials managers
then update this forecast based on their own research and knowledge of the market for
particular parts. As one would expect, the materials managers ability to forecast based on
customer input improves as time goes on such that next years forecast is the most up to
date and most accurate in the three year forecast. In a time period of generally increasing
demand, the years further from the forecast point that were populated based on historic
demand will lag the current trend. One would expect that if Pratt were to enter a period
when demand is generally falling, those further out years would also lag and show a
41
rising forecast when the trend is actually declining. Clearly this artifact of the system is
detrimental to capacity planning, raw material procurement, and supply chain planning,
and places pressure on the manufacturing organization to respond to sometimes dramatic
year over year changes in the forecast.
In summary, the utilization based forecast methodology provides similar accuracy
to the more traditional customer survey based forecasting that Pratt & Whitney has been
engaging in. However, it does with much less effort than the hundreds of hours spent in
phone calls and data collection than is currently required. The results presented in this
chapter are all aggregated results. It is possible to disaggregate on two parameters of
interest, namely specific customers and specific part numbers. As one would expect,
disaggregation yields less accurate results. Pratt did not have records of customer
specific forecasts or demand history to investigate the degree of this decline in accuracy
on this dimension. However, it is expected to be significant. The assumption in
generating these forecasts is that the efficiency of the market makes the many customer
specific decisions irrelevant as in the case of build standards.
Part specific forecasts, on the other hand, are a necessary part of doing business.
While an aggregate market forecast is interesting to the materials managers, a distribution
across specific part numbers is necessary for making planning decisions. These are not
presented here in order to protect Pratt's interests. However, developing part specific
forecasts yielded one key learning regarding the utilization based forecast. Parts that
have experienced a major service issue, for which the response was to replace all parts in
the fleet, cannot be forecasted using the utilization based methodology without resetting
the replacement profile to account for the shift. Unfortunately, this can be quite
42
cumbersome due to the timing of a service issue in the maturity of the fleet and the
specific part revisions affected.
In addition, the part specific forecasts were been adjusted based on historic
correlation of actual and expected demand to yield the current forecasts. The number of
cycles to replacement has been adjusted from the published Chapter 5 preventive
maintenance limits to give a best fit with historic demand. The parameter in the fleet
reflected in this adjustment are the build standard of customers. In other words, parts are
not used up completely, thus requiring replacement before one would expect resulting in
shorter average part lives. Data is not available to calculate an actual average part life so
we use the correlation of actual and expected demand to approximate the part life going
forward. The remaining error after this adjustment, plus error introduced by the
uncertainty in the future rate of aircraft cycle accumulation is then applied to the
projection, yielding a forecast with error. Forecasts with error were an unfamiliar idea at
Pratt & Whitney when we began this project. By the end of the six month project, some
early adoption of the inclusion of an error in forecasts, yielding a range of expectations
for the manufacturing organization, had begun.
43
Chapter 5: GMS Forecast and Inventory Control
The application of the forecast model to GMS parts is similar to that of the Pratt
& Whitney parts with one exception. Pratt does not have data on the demand history for
these parts. Therefore, a best fit of historic data to identify the estimates of remaining
stub life at replacement cannot be completed. Two alternatives were proposed to
compensate for this lack of data. One proposal is to create a profiling tool that
incorporates operator's typical maintenance intervals, the Chapter 5 Limits for the life
limited parts. An example of this profiling tool is shown below.
LLP Life
Fan Disk
Booster Spool
Fan Shaft
HPC Forward Shaft
HPC 1-2 Spool
HPC Stage 3 Disk
HPC Stage 4-9 Spool
HPC Rear CDP Seal
HPT Front Shaft
HPT Front Air Seal
HPT Disk
HPT Rear Shaft
LPT Stage I Disk
LPT Stage 2 Disk
LPT Stage 3 Disk
LPT Stage 4 Disk
LPTShaft
LPT Rotor Support
LPT Stub Shaft
30000
30000
30000
20000
20000
20000
18824
17980
18323
18000
20000
20000
25000
25000
25000
25000
30000
25000
25000
Maintenance Invervalsj
Accumulated Engine Cycles
10000
10000
Build Standard
1
1
-
-
-
-
1
2
-
2500
2500
2500
1324
480
823
500
2500
2500
-
1
3
|
4
-
0
-
-
-
-
-
-
Stub Life Plot
|
5
1
0
0
-
6
1
7
8
9
1
0
0
-
2100
2100
2100
924
80
423
100
2100
2100
-
_
_-
-
4224
3380
3723
3400
-
-
-
-
-
-
-
-
-
0
0
0
0
0
-
-
0
-
-
-
-
0
-
-
0
-
-
-
-
0
-
-
0
0
0
-
42900
7100 1
1 4600
50000
55400
60000
1 7500
17500
0
0
1 7500 1 5000
25000
30000
5400 J
35400
-
-
-
1
1 7500
67500
7500
Table 3: Stub Time Profiling Tool
Three inputs are required to generate such a profile. First are the LLP Chapter 5
limits. Second is the build standard to which the engines will be maintained or the
minimum acceptable remaining cycles on a part after maintenance. Lastly are the
expected maintenance intervals of the engine during its lifetime. In practice, these inputs
44
10
0
-
-
|
75000
are very operator specific. Some operators manage this activity to minimize the lifetime
operating costs of the asset. Others focus on minimizing current costs. Some operators
manage this process with little sophistication and simply define a maintenance interval
that is convenient for their operations. In aggregate though, the varying standard and the
presence of an efficient secondary market yield few parts that are disposed of with
remaining life in excess of 2000 cycles. This profiling tool was expanded to include the
revenue generated by the sale of parts into the secondary marketplace when they are
removed with significant life remaining and optimized to provide the operator the lowest
lifetime operating costs.
The second proposal for estimating the remaining stub life for LLP's on the
CFM56-3 engine was to use a fixed percentage of the life limit of the part. Five percent
is the average observed stub life experienced at the Pratt & Whitney engine repair
centers. However, both these proposals fail to recognize that the forecast model stub life
term is not just a reflection of the life remaining on parts when they are scrapped. Rather,
it reflects this remaining life as well as the impact of other inefficiencies in the system
such as the impact of spare engines in the fleet. Instead, we elected to use a zero stub life
in our forecasts. Most of the PW2000 parts showed a best fit between expected demand
and actual demand over the course of the engines history with a using zero stub life. In
addition, we know that in reality, no parts will be consumed completely. By assuming
that they will be, we generate a more conservative forecast. Given the uncertainty in the
business (at the time of this project, Pratt had not yet gained FAA approval to
manufacture these parts) a conservative forecast seemed prudent.
The resulting forecast is shown below in Figure 14. Again, to avoid released of
specifics of Pratt's business plans, I have aggregated the data based on some common
45
characteristics. In this case, they are aggregated based on the engine module of the parts.
The four modules of the CFM56-3 engine (and most gas turbine engines for commercial
aircraft) are the Low Pressure Compressor (LPC) or Fan Module, the High Pressure
Compressor (HPC), the High Pressure Turbine (HPT) and the Low Pressure Turbine
(LPT).
The estimate of error included in the forecast is based on the observed error for
the Pratt & Whitney parts. The average error for individual parts was about 40%. This
was included in the individual part forecasts and is aggregated in the module forecasts.
The charts show the forecast applied over the course of the business case developed by
Pratt & Whitney when making the decision to enter the market. This business case was
developed with a ten year planning horizon.
LPC Volumes Forecast and Business Case
El Utiliz ation Based Forecast
0 Busi ness Case
00
.00
-0"0
U)
.00 00
.10 04
D
00
.1 0,
00
oh
0
0/."
1
/Je
T
//R
.10
410
2008
46
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
HPC Volumes Forecast and Business Case
O2 Utilizaiton Based Forecast
o Business Case
U0
~
7
4.0
0
UtlztoAasdFrcs
AO
AO
00/
/W
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2017
2018
HPT Volumes Forecast and Business Case
E3Utilization Based Forecast
M Business Case
U)
210
2008
2009
2010
2011
2012
2013
2014
2015
2016
47
LPT Volumes Forecast and Business Case
0Utilization Based
Forecast
0Business Case
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
Figure 14: Forecast and Business Case Volumes
The aircraft utilization based forecast is displayed side by side with the volumes used to
justify entering the business. For the high and low pressure compressors, the business
case is within the expected error of the forecast. This is not true for the turbine side of
the engine. For these modules, the forecast comes in significantly lower than the
business case, albeit not significantly until 2010.
The business case was developed by Pratt & Whitney using information from
their engine repair facilities. Much like the forecasts used regularly for Pratt parts, scrap
rate data was collected from engine overhauls. Then this scrap rate was applied to an
expected number of shop visits that Pratt forecasted market wide for the CFM56-3
yielding an expected number of parts consumed. This was then adjusted based on the
fraction of the market Pratt expects to capture over the next decade. The business case
forecast has the same problem described in Chapter 2 regarding using scrap rates to
48
project Life Limited Part demand; namely that this method does not reflect the periodic
nature of demand for LLPs. In addition to this inherent problem, the data set was limited
to 33 engine overhauls. With this limited data set, it is impossible to make projections
without a significant error term.
Consider the following randomly generated data set generated with a probability
of encountering a part that needs replacement of 30%. Three samples of 11 engines are
taken. We calculate a scrap rate for each sample and find the average and standard
deviation of the samples. The resulting scrap rate is normally distributed with mean and
standard deviation of 33% and 10% respectively.
0
0
0
0
1
0
1
1
1
0
0
1
0
0
1
1
0
0
1
0
0
Scrap Rate
45%
0
27%
Mean Scrap Rate
Standard Deviation
0
0
1
0
0
0
1
0
0
0
1
27%
33%1
10%1
Figure 15: Scrap Rates
Based on these figures, we can only say with confidence that the true mean is between
13% and 53% (a = 0.05). Including an error term estimated based on this hypothesis of
approximately three standard deviations in the business case brings the aircraft utilization
based forecast and the business case into much closer agreement. The failure of the
business case to include the dynamic nature of the demand process accounts for the
residual differences.
49
LPC Volumes Forecast and Business Case
O Utilization Based Forecast
O Business Case
U)
400
0go
2008
2009
2010
A0
2011
2012
2013
2014
2015
2016
/
2017
0
2018
HPC Volumes Forecast and Business Case
5 Utilization Based Forecast
O Business Case
00
:0,
2008
50
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
HPT Volumes Forecast and Business Case
U Utilization
Based Forecast
* Business Case
L
Wd
U)
T
Ill
TIT
K
/
I-
/
/
2008
2009
2010
2013
2012
2011
2014
2015
2016
2017
2018
LPT Volumes Forecast and Business Case
13 Utilization Based Forecast
II Business Case
U,
iiI~
I~l~l-
F
I
_
C
/i
-ii
#1
0
I-
/Z
/
'I I
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
Figure 16: Forecast and Business Case Volumes with Error
51
With forecast in hand, we now turn to the issue of inventory control policy. The
aerospace industry has suffered from increasing lead times. With expected lead times of
more than a year for the Life Limited Parts Pratt & Whitney will be manufacturing for the
CFM56-3 and forecast errors on the order of 40%, high levels of inventory are to be
expected. However, looking into Pratt's supply chain points to opportunities to reduce
inventories beyond what would be necessary if Pratt applies its traditional inventory
control methods to these new parts.
The primary opportunity for inventory reduction comes in the fonn of strategic
inventory placement. Pratt's supply chain for Life Limited Parts consists of four stages
from raw material through finished machinings. Some parts require assembly after
machining adding a fifth stage to the chain. The first stage is the procurement of raw
material or ingot. Pratt places orders for raw material and enters a queue for fulfillment.
Lead times for this stage are commonly measured in months. The average for the GMS
LLP's is nearly six months. Next stage is billet where the raw ingot is transformed into
standard size stock. This process averages two months. The billet material is sent on for
forging with a lead time of two and a half months. Finally, the forging is machined into a
finished part. The total lead time from beginning to end of this supply chain is more than
13 months.
Most Life Limited Parts at Pratt & Whitney are ordered as rough forgings. That
is, Pratt places an order with a supplier for a rough forging, receives the forging, and then
machines the part internally. With the GMS program, existing relationships and the
current business climate motivated Pratt to use smaller suppliers.. Many of the first tier
suppliers were hesitant to work with Pratt on equivalent parts to those they were currently
manufacturing for GE. The second tier aerospace suppliers selected were not in the
52
position to purchase raw material for Pratt parts, therefore, Pratt owns the raw material
much further back in the supply chain than is normally the case. This presents Pratt an
opportunity to take a global look at inventory placement within the supply chain, and
evaluate where in the process it is optimal to keep inventory to provide the desired
service level.
To consider this opportunity, the inventory demand process was modeled as
normally distributed with the demand rate equal to the forecasted demand in 2010
variance equal to the expected forecast error of 40%. We assumed that the inventory
control process would a continuous review policy and the desired service level to be
95%. While the assumption of normally distributed demand does not hold when we look
at the demand process over a period of a number of years is not accurate, it is
computationally efficient. Additionally, this model is not intended provide exact results,
simply to identify where in the supply chain safety stock might be positioned to minimize
costs.
i
Ingot
Buffer
Machining
Forging
Billet
Buffer
Buffer
Finished Goods
Figure 17: The Life Limited Part Supply Chain
In order to find the optimal positioning of safety stock, the stock necessary to
satisfy a 95% service level was calculated under a number of scenarios. In all, it was
assumed that finished goods inventory would be held as the lead time for the final
echelon of the supply chain is longer than Pratt's seven day delivery commitments. The
scenarios are as follows. In addition to finished goods, inventory is held:
1.
nowhere
53
2. as forgings
3. as billet and forgings
4. as ingot billet and forgings
5. as billet
6. as ingot
7. as ingot and billet
8. as ingot and forgings
Table 4 shows the results of this analysis for a representative part. The total value
of safety inventory is minimized when material is held as billet in addition to finished
goods. The total value of safety stock carried is 23% lower when maintaining a buffer as
billet when compared to than holding finished goods only. Cases 6 and 7 also achieve
significant reductions in the total value of safety stock.
Inventory Units Held After Piocessing Stage
Stage LT
Net Cost
(weeks) after Stage
Process
Stage
Ingot
Billet
Foige
Machine
28
10
12
18
Profile 1
(0001)
$ 5,500
0
0
$ 7,500
0
$ 21,500
10
$ 30,000
Safety Stock $ $ 300,000
Profile 7
(1101)
Profile 2
(0011)
Profile 3
(0111)
Profile 4
(1111)
Profile 5
10i011
Profile 6
(1001)
0
0
8
5
$ 322,000
0
7
4
5
$ 288,500
6
4
4
5
$ 299,000
0
7
0
6
$ 232,500
6
6
4
0
0
0
6
7
$ 243,000 $ 243,000
Table 4: Results of Inventory Placement Analysis
The next task to be completed with a demand forecast was to determine
appropriate inventory targets. Of particular interest to the Pratt & Whitney managers are
the expected average inventory levels for the GMS program as this metric is included in
their annual performance targets. Because the program is new, there are no established
metrics. The estimates we generated as part of this project will become the targets for the
first year of production.
54
Profile 8
(1011)
6
0
6
5
$ 312,000
To estimate the required inventory, Pratt & Whitney's typical target service level
of 95% was used. This fill rate target is based on Pratt's experience with customers. As
one Pratt manager put it, "when we drop below 95%, that's when the pain begins."
These difficult times stem from the efforts of expediting, putting customers on allocation,
repurchasing materials from customers that have material in their own inventory, and
other efforts made to satisfy new orders. We then simulated demand as a Poisson process
for three years. The rate of demand arrivals was set at the forecasted annual demand rate.
The Poisson assumption was based on examination of Pratt & Whitney demand history
for the PW2000. Using a Poissonness Plot developed by Hoaglin (1980) shows that
demand for LLP's is reasonably well described by the Poisson process. Figure 18 shows
such a Poissonness Plot for the PW2000 4h disk. The straightness of the line
characterizes the degree to which the sample data fits the Poisson.
4th Disk Poissoness Plot
14
12 -
10
8
0
4
2
0
0
1
2
3
4
5
6
7
8
9
k (monthly demand)
Figure 18: 4th Disk Poissonness Plot
55
The inventory control process used for the simulation is an adaptive base stock
policy as described by Graves (1999). This policy is similar to the policy employed by
Pratt & Whitney in their MRP system. We simulated the response of the system over the
three year period during which Pratt expects to build up its capture of the CFM56-3
aftermarket. The base stock level adapts to reflect the demand forecast over the lead time
and the current level of inventory. In addition, the level of safety stock changes over time
to reflect an expected increase in volatility as the level of demand increases. The results
for a typical part are shown in Figure 19. The horizontal column in measured in
multiples of the safety stock calculated above in the inventory placement calculations.
LPT Stub Shaft Simulation Results
99.5%
- -
$300,000
-1-99.0%
1
$250,000
98.5%
$200,000
98.0%
$150,000
0)
ii
97.5% U-
-4-Average Inv (2008 - 2010)
- Lost Sales
$100,000
97.0%
-A- Fill Rate
$50,000
96.5%
I96.0%
1.00
1.10
1.20
1.30
1.40
1.50
1.60
1.70
1.80
1.90
2.00
Multiples of Calculated Safety Stock
Figure 19: MRP Simulation Results
The results shown demonstrate the impact of increasing safety stock, and thus
average inventory has on Fill Rate and Lost Sales. For the LPT 1" Stage Disk, increasing
Fill Rate from 96.5% to 99% requires an increase in safety stock by 70% and average
56
inventory by 40% (~$200,000). The increase in fill rate yields an increase in expected
annual revenue by only $12,000 for this part. This tool will be useful for managers to
evaluate the financial costs and benefits of changes in target fill rate. In summary, we
found that to achieve the targeted fill rate, Pratt & Whitney would need to carry
approximately $5.8 million. Figure 20 shows the trade off for the sum of all the GMS
Life Limited Parts.
LLP Inventory vs Service Level
$10,000,000
$9,000,000
$8,000,000
$7,000,000
$6,000,000
$5,000,000
$4,000,000
$3,000,000
$2,000,000
$1,000,000
40%
50%
60%
70%
80%
90%
110%
100%
Fill Rate (Fraction of demand satisfied)
Figure 20: Inventory vs. Fill Rate Trade Off
57
Chapter 6: Discussion, Conclusions, and
Recommendations
The results of this project completed at Pratt & Whitney's East Hartford, CT
campus include a novel forecasting methodology for Life Limited gas turbine engine
parts based on aircraft utilization, application of that methodology to Pratt & Whitney's
Global Material Solutions business, a strategic inventory placement analysis for GMS,
and target inventory requirements to achieve desired fill rate. Over the course of the
project, the author had the opportunity to work in partnership with Pratt & Whitney
employees and spent a great deal of time discussing the potential implementation and
future use of the work as well as the GMS business in general. These discussions are also
relevant to these conclusions and recommendations.
6.1: General Recommendations
Six months on site with Pratt & Whitney provided the opportunity to experience
the benefits and problems inherent in organizational design. Pratt operates as a matrix
organization with the goal being a workforce that is focused on globally optimal decision
making. That is, if reporting both within a functional hierarchy and work within product
focused team, employees will be more likely to work together with employees from other
functions and make decisions with a broad perspective. This vision has not been fully
realized. Much of the decision making is myopic, focused on ones functional objectives.
Much of the evidence of this is anecdotal but did have impact on this project when a
recommendation to introduce an inventory buffer of raw billet material was made. This
analysis, while well received and understood, was not implemented. It seems that in
Pratt's current organization structure and incentives, Manufacturing owns work in
58
process inventory. An inventory buffer would be considered work in process and
therefore, manufacturing would own the inventory. The benefit of carrying that
inventory would be reduced finished goods inventory resulting from a quicker response
time to new orders. This benefit would be had by the spares organization.
Conflicting incentives such as these that are not aligned with company objectives
were talked over water coolers and discussed at lunch but no changes were made in them
during the course of this project. Based on these observations, Pratt could benefit by
developing a team that will focus on global supply chain initiatives that is empowered to
make changes in the system. This might take the form of a Logistics Services
organization that works for the engine programs and must sell its services. By creating
such an organization that is required to sell its services, they will be motivated to do the
global analysis, demonstrating to their constituents the value in making the changes they
propose.
One of area of focus for such a team is Pratt's strategic reserve of raw materials.
Currently, Pratt maintains some inventory of raw materials. However, this is not
considered to be a buffer inventory but is held in the event that suppliers are unable to
obtain material on the market to satisfy Pratt & Whitney orders. Suppliers make requests
to access the Pratt reserve when market prices are too high or queues lengthen to the
point that will be unable to manufacture the components on time. By taking a more
holistic, total supply chain perspective, there is potential for Pratt to capture more value
for both UTC and its suppliers.
59
6.2: Project Specific Recommendations
The aircraft utilization based forecasting methodology we developed should prove
useful for Pratt & Whitney. However, there are several caveats to this statement. First, it
is a market forecasting tool and relies on certain efficiencies of the marketplace to reduce
expected error. Attempting to apply this model for customer specific forecasts without
additional refinement is likely to yield unsatisfactory results.
Consider the launch customer for the Global Materials Solutions business, United
Airlines. Over the course of three years, in efforts to emerge from bankruptcy in
February 2006, United began charging for meals on flights, renegotiated its labor
contracts reducing labor rates by 30%, cancelled its employee pensions, cut wages again,
and took on $3 Billion in new debt. Over the course of those three years, the only thing
on the mind of United's managers was survival. They put any part they could buy into
their engines regardless of how long it would last before they were required by the FAA
to replace it. Many of these parts were purchased from the secondary marketplace rather
than from the OEM, increasing lifetime maintenance costs but minimizing short term
expenses. The accumulated cycles on the parts in United's engines most likely are no
longer reflected by the accumulated cycles on their aircraft. Our method will not work
when applied to this subset of the industry because of this behavior.
To apply this method to a specific customer, Pratt must first obtain detailed
information about the current status of their engines. By gaining this data the model
could be modified to provide a customer specific forecast. I recommend that Pratt
consider purchasing this information from customers or including the provisions for
collecting it in contracts for service and materials.
60
While at Pratt, when the need for data from United or other customers was
discussed, the common response was that the airline operators engineers were probably
too stretched by the climate of layoffs and downsizing to collect and provide such data
regularly. Given this situation, Pratt might consider placing its own people on site with
the airline operators or consider paying the operators through reduced prices or promised
service levels to collect this data. Experiments would have to be undertaken in which this
data is available, a customer specific forecast generated, and the accuracy of this forecast
compared to the accuracy of a forecast without this additional data. The increased
accuracy will translate to a reduced level of safety stock required to support the desired
service level thus allowing Pratt to value the information.
In addition to the forecast for Life Limited Parts, Pratt has need for a forecasting
methodology for the non Life Limited or gas path parts. The Markov Process described
by Pratt managers was interesting and will provide an excellent starting point for a future
Leaders For Manufacturing internship. In this area, I recommend placing the intern with
the group that will have the best access to data rather than within the Spares or GMS
organizations. Muench (2003) described a data collection and cleansing process in his
thesis that could provide much of the data necessary to complete this analysis. However,
it seems that Pratt abandoned the effort shortly after Muench left the project. Renewing
efforts in this area could provide valuable data not only to the GMS program but also to
Pratt's Spares Organization and engine centers.
Regarding the specific forecast and inventory target recommendations developed
as part of this project, Pratt would do well to put a rigorous framework around the
determination of service level and fill rate in the Spares organization. Fill rate is one of
the key measures that the managers are held responsible for in this area but there seems to
61
be little thought into what the specific target should be. As a sidebar to our efforts, we
completed a newsvendor type analysis for the GMS parts, demonstrating that the
expected return for keeping a high service level for low margin products is much less
beneficial than maintaining a similar service level for high margin products. While the
specific business model at GMS, where high service level has potential to win long term
customers is not the area at Pratt where experiments in this regard should begin, there is
potential for Pratt to reduce its inventory holdings and improve its profitability by adding
rigor to the Fill Rate target selection process. One area to begin such an effort would be
for parts on Pratt's legacy engines where PMA's have developed suitable alternative
supply. By concentrating on these parts, Pratt could reduce its inventory position without
increasing significantly the risk to customers that they will be unable to locate a needed
part.
In conclusion, the forecasting method developed for Life Limited Parts at Pratt at
Pratt & Whitney is robust and effective when applied to the entire market. In many
industries, this would be of little value. However, for the commercial gas turbine engine
business, OEM's have a defacto monopoly on Life Limited Parts through the regulatory
restrictions of the FAA. Therefore, the forecasting method will continue to provide
value. In its current form it is limited to market wide use but with additional, customer
specific data, the functionality could be extended to provide customer oriented output.
62
BIBLIOGRAPHY
Chenevert, Louis. Speech. Aviation Industries Association. Aviation Safety Alliance
Newsmakers Breakfast. Washington, DC. 31 Mar. 2004. 8 May 2007
<http://www.aia-aerospace.org/aianews/speeches/speeches_2004.cfm>.
Flint, Perry. "Picking Up the Spares." Air Transport World (2007): 62. 29 Apr. 2007
<http://www.atwonline.com/magazine/article.html?articlelD=1843>.
Graves, Stephen C. "A Single-Item Inventory Model for a Non-Stationary Demand
Process." Manufacturing and Service Operations Management 1.1 (1999): 50-61.
Hoaglin, David C. "A Poissonness Plot." The American Statistician 34 (1980): 146-149.
Montgomery, Douglas C., Lynwood A. Johnson, and John S. Gardiner. Forecasting &
Time Series Analysis. 2nd ed. New York: McGraw-Hill, 1990.
Pinkham, Myra. "Producers Boosting Capacity to Meet Soaring Demand." American
Metals Market (1997). 29 Apr. 2007
<http://findarticles.com/p/articles/mim3MKT/is-n69_vi05/ai_19303943>.
Smallen, David. "Press Releases." Bureau of Transportation Statistics. 9 Jan. 2002.
United States Department of Transportation. 29 Apr. 2007
<http://www.bts.gov/pressreleases/2002/bts001_02.html>.
"Timeline of United Airlines Bankruptcy." USA Today 01 Feb. 2006. 15 Apr. 2007
<http://www.usatoday.com/travel/flights/2006-02-01 -united-timelinex.htm>.
UTC 2005 Annual Report. United Technologies Corporation. 2006. 20 Feb. 2007
<http://www.utc.com/annualreports/2005/html/index.htm>.
"JT8D." Pratt & Whitney. 18 Feb. 2007 <http://www.pw.utc.com/vgn-exttemplating/v/index.jsp?vgnextrefresh=1&vgnextoid=cfd3b839652db01OVgnVC
MI 000000881 OOOaRCRD>.
"Original Design Approval Process." Federal Aviation Administration. 6 June 2006.
United States Department of Transportation. 2 Mar. 2007
<http://www.faa.gov/aircraft/air-cert/designapprovals/origdesapprovproc/>.
63
"Parts Manufacturer Approval (PMA)." Federal Aviation Administration. 26 May 2006.
United States Department of Transportation. 2 Mar. 2007
<http://www.faa.gov/aircraft/aircert/designapprovals/pma/>.
64
Download