So the fucking final Tonight we're all
right Dr Alexander girl is not the right.

Hold your thought.

But just for yourself.

I should say that the beginning of
the fourth right after the program was so

complex and so
little your support formation.

You saw as it was said by
somebody from the company

who saw a little information that
matched the bill so much of the first

and then this last portion of
the told us that they are going

to market this software and then
basically develop a new concept called

Good evening everyone.

Your team are in my my name is Katie and
I'm late even read it but

if you go actually printed and done we
would like to give a special thank you.

Try faculty advisor.

Dr Shapiro and charts I can type out it so

are my are real her measuring heart rate
it is a software as a service provider for

right now companies in
the transportation industry

first project we focus on companies
within the transportation sector.

So while our lives are direct land we are
actually going to benefit our my customers

as such are my local real life percentage
really is certain percentage savings

of what we give to their customers for
this project we use the real

life software that are in my house
in real time it allows more users

to visit weeks to visit we manage
tests and sections out on the real.

Do you pull wool users are called meteors
here's an overview of a project to

the problem here is that currently there
is no specific way to Stengel test and

inspections result is using an
optimization model leading to a value of

a consummately one hundred nine labor
hours city per maintainer over the course

of a year now the test of
reform are done on these assets

we are working on the assets we have
signals how this switches and crossings.

So a particular type of test
that might occur is and

maintained it could go
to this crossing and for

example mention of the lights are flashing
and that the can move up and down all of

these tests are done as mandated by the
federal railroad association or a part.

If the railroad companies or have assets
that are found to be out of compliance and

the F.R.A.
levied a very heavy penalty on them.

There are many different types of tests
that come in many different frequencies.

We have thirty eight hundred ninety
data one hundred eighty data has

three sixty date has an addition to
your four year and ten year test.

Now the problem here is that real dives
does not provide a specific schedule for

maintainers follow.

So I did see a screenshot to the right.

We have a list of as ladies along
with the building next to you

now not the presentation all as
ninety's or either blurry or

jumble because of confidentiality
as you can see all the answers

are color coded based on
will be when they'll be due.

However the maintainer
can decide on his or

her own when ever they
want to perform a test.

Now let's give a three day causing turns
and example not in the most optimal case

the being can or will not perform
any more tests and I absolutely do.

This means the maintainer will perform
the task exactly on the day that it is due

to this case the first test.

Form and generate a hit and
then two N's one thirty later and

then during the fourth.

So along a six month period.

Naturally to meet him before this test.

However the star that we have
found I mean can perform tests

three point four days in advance.

Plus in the same example and say that
they perform every twenty seven days that

again the first test is done on January
the it is actually doing thirty days

just separate Of course the legal form
it three days earlier on your universe.

A lot of that same six month period
the meteor will perform one additional

This is not a desirable situation for
railroad companies

because it will result in higher
labor costs to pay for the additional

As I said before there are more than just
three day tasks as you can see here.

Now this is where the problem
becomes much more complicated

the certain tasks are what we call parent
task that means that the performing

the lower frequency test is inclusive
of the higher frequency test

was able as example and
the first sheet in the first sixty days

to give this one to thirty
eight has naturally this third.

Thirty eight test is due
on the ninetieth day.

However because often times thirty day and
ninety day test are paired

you can instead form an inclusive
ninety day times here and

include the work of thirty data
similarly again we can say that ninety

nine times are paired with one
eight as and when you need it.

Times are paired with three sixty data.

However and I said before many times.

Times are performed to early so
I decided it has performed too early.

That is due date will actually be before
that includes the one hundred eighty

day tasks so as you can see there
is an extra test performed and

along down the line additional tests
can be run in the same fashion.

This is not a desirable
situation again for a.

Company.

There are always out and

higher labor costs that we don't have
that are right we don't talk about.

OK so
in the first part of our data collection.

We receive historical data
in the form of My S.Q.L.

database from one of their largest clients
which we will refer to as a class one very

basically a class one railroad means that
it's one of the largest certainly real

real companies in the nation.

Now we received this information from your
spending two thousand and two thousand and

ten however we wanted to model present
their activity as closely as possible so

we only looked at data from
the two thousand and ten.

Next we identified identified ninety two
different tests required by the federal

railroad association for each of
these tests we assign them a unique

identification number from one to ninety
two which we'll refer to as Georgia Tech

and these and then finally we remove
that layer information so we can model

the average maintainer and in that way we
get the most generic solution possible.

The next phase of our project involves
calculating just how long it takes for

a maintainer to perform each test.

Now when we reached out to our mine for

this information we found out that they
actually don't track this at all and

rather they only know what tests were
performed on what they buy what maintain.

Now in order to come up with another
to my schedule we had to calculate

this information somehow.

So we did was we assigned a variable
to each one of the tests and

using the average maintainer
workday of eight hours we set

up a system of equations which would
solve to give us a corresponding weight.

If you're very attention here on the right
this is a very simplified version of what

we did so we have a test data and test B.
and we assign variables X.

and Y. in on more particular day two
of the tests they were performed and

one of the tests be were performed and

one of those on another day two were to
spirit perform and the corresponding ways

would be two hours in four hours
per test day in Test B. respect.

Now in our specific case we had some
test actually didn't occur at all

in the year two thousand and ten as
well as some tests that one applicant.

To the specific program.

So in total we ended up with seventy
three different variables and

we actually had over three
hundred thousand total equations.

So what we did was we fit
the least squares method

to solve this system of equations and
give us their corresponding weights.

Now I want to stress that these weights
are very crude estimate it's because

they do not take into account factors
like maintain or variability and

general overhead such as Time
to maintain or spend one

of the most difficult parts of calculating
your way through the system of equations

was a very wide range of test occurrences
as you can see from this graph we actually

had some tests that occurred in the tens
of thousands of ranges an example being

ID seventy seven which occurred over fifty
seven thousand times now for these types

of tests we were able to calculate weights
for which we were highly confident but

at same time we'll set some tests that
occurred like less than one hundred times

the example being ideal which actually
only occurred once in this particular year

so for these ways we weren't quite so
confident and so what we did was we

reached out to our might ask them to
survey may take a sample of maintainers as

well as other field experts and
get some ballpark assessments so

we could compare it to our results and
come up with a final I said a week.

So the recovery weights.

I'm from our mathematical calculations
alone we were able to calculate

thirty five out of the seventy three
total weight so just under half and

while that may not seem that
impressive if you look here.

On the right from those thirty five
weights alone they represent almost ninety

nine percent of the total test
performed two thousand and ten so

specifically out of three hundred ten
thousand total test corresponding to three

hundred five thousand.

Now again I want to stress this
was something never before done by

our mind and actually keeping this in
mind we hadn't answered these weights

are already been implemented into
other facets of our minds business so

that doesn't stop the lame speak
about her after the session you.

The representation of our most of our
model is a mix and right now the salt and

sea life is a compensation software
the model schedules tasks for

and that means painter over
the course of a year and

has a ninety five thousand constraints and
up to seventy thousand and

these numbers vary due to the fact that
the end points for each maintainer

are now long as there are checksum
function and parts constraints and now

here is a close up look of our object
function simplified are intact and

to minimize labor costs or
hours are comprised of right here in time

hours as well as overtime hours for
the purpose of our project and

anything exceeding forty hours
a week was considered overtime.

Now there are some different inputs for

fire front end user the first
two time horizon and weight loss

the entire horizon for our project was
spent consuming Zoran year and wage cost.

While we couldn't use monitoring values
we chose to use a weight of one for

regular time hours and
one point five per hour and

we did this in order to about why
our solution in terms of hours and

also required are the number and
type of the expected times for

that individual maintainer for
every test to know that test.

Thank you so in this example we have I
need seventy four which is a very take

crossings us also is the initial
tagline that SAS national that minus

the number six until the test is due for
one starting from one of them will run.

Also the time it takes to complete
the toss and finally and parents us.

So in this example I am seventy six

which is a ninety day crossing
task which income doesn't work.

I need seventy four hour
there a few times and

constraints are model time constraints and
testing trainings time trains and

plane that nobody can work from or
as well.

Maintainer may not work more than twelve
hours a day as mandated by government

regulations and Tyson is trying to bar
that every time it's must me International

that life and her merit out must mean for
now he's starting its frequency.

So for
example at thirty nine times must go.

Yes once every thirty days and
then finally must be more together.

Now the final outcome our model
has both scheduling time as

well as the optimal song on waivers
the last one can see in the Southwest

seen as a box where we have
our optimal solution and

ours was a schedule and
matrix form to interpret this matrix and

reroute represents and
every column represents a test ID So

this example lets maintainers for four and

final one and position for
January first on this particular house and

I see now where you want
in a row as in the case one

times are the same now
hours will speak for

her so what we provided to our
mind was the following an S.Q.L.

statement that queries all the relevant
data from the database such as the passage

from nation and initial guidelines.

Next we have an Excel macro that develops
the inputs as I mentioned the entire

model and display in Excel
not through the final step.

So here's a quick demonstration of his
Excel file you like to maintain or

religion is made here five
sixty five this button.

This guy has all the inputs for C.

plex and here you see the final step
is the transformation of the zeros and

ones major sees and hears

example the same from the demo on
the left you see information away.

You have.

We did in two thousand and ten and

on the right is the data from
our optimized schedule so

we estimate that there is maintain your
home were eighty plus hours before and

thirty six less tests and what to say
this railroad an estimated three thousand

dollars and we did this in a similar
fashion with a larger sample size and

we calculated on average an estimate
one hundred nine Labor our savings per

maintainer claim mentioned before I
rejected a value over optimization

model is in terms of hours and this is
inclusive of regular hours as well as

the analyze overtime hours before this
hundred ninety also counts for overtime.

So I have a night multiplied
by thirty five which

are my gave us as the average hourly wage
and we calculated an average savings of

three hundred dollars per maintainer now
you'll see that our savings in hours

ranged from negative fifty six to nearly
six hundred and these negative savings

mean that our Those go to more tasks than
actually performed in two thousand and

ten and this was due to those maintainers
consistently performing their tests late.

So our that's going to more tests and
order to keep those assets in compliance.

So we took that thirty one hundred from
before which was the our savings for

maintainer and you multiply that by
the total number of maintainance and

we calculated an estimated two point
three five total dollars total savings.

Now these maintainers are not work for
our mild their cost one road or

as he mentioned are my realizes a certain
percentage of their clients' savings and

this is generally ten percent.

So our mind blower and estimated two
hundred thirty five thousand dollars and

for many this good when I read them.

I also know that I just revenue will
be realized once applied to their other

clients who are most currently preparing
to integrate our out rhythm into their

software and there are several other ways
which are my eyes and many of our projects

in two weeks are my well intended
railway system supplier Expo.

Showcase our project to put
prospective clients as well

as their favorite customers right
now are my right software for

five years and are currently looking
to expand their market share.

Additionally I distantly our model

maintainers Now they made Here's how this
work hard that measures their performance.

Right now and
this only trapped in the eyes from

one test that there is no
point in the test early.

So the help of our model are more
of the maintainers form the test

is scheduled tested and
there are a deadline and

a test performed outside
that ring earlier analyzed.

Finally our data analysis.

Our minds are aspects of their software in
a way so as I mentioned something in her

read several hundred hours sittings Now
these findings help visualize those

maintainers who are possibly workers and
also those who are underutilized So

the summary for the semester are my
software only provided maintainers

with a list of deadlines of tasks that
they can perform whenever they want it so

we provide our my child optimization model
to test get all these tests in such a way

that will save this room one hundred nine
labor hours per maintainer as most earn

are my as five thousand dollars and
this concludes our presentation

of the email how much

of the night.