You know it's a video where you can and

if you find the MIT and
you miss the delete from Stanford.

Well actually it is a trustee of
a massive technology leadership.

You know I think we're very fortunate to
have Jack here to talk about innovation

this is a welcome thank you.

It's what my talk this morning is called
innovation in systems perspective

the A model based design.

The car.

You may have driven here this
morning in the airplanes

that you travel around for vacation your
smartphone you might have in your pocket.

These are all systems.

OK systems that we use every
day there are also complex

systems that you may not realize it but
virtue all the systems were

designed by teams using model based
design that you know the car you have

is millions of lines of code in
the engine control units and

those are that code was automatically
generated by the Simulink

software system from the math works.

And so in the end this is all about
model based as I wrote you all

the companies that make these products
these consumer products have adopted model

based design sometimes we call it.

N.B.D. review.

Over the last ten years
in order to innovate and

speed up their product development.

It's really it's really been a revolution
in terms of how products are developed in

the industry and
it's actually not that well known in

education these techniques are not
quite quite as well understood and

used in education as
they are are in industry.

So anyway so
what I mean by model based design.

Well.

Model based design is on
I'll talk about today's.

It's both a methodology and
it's a tool set for

system level product design and it uses a
model throughout all phases of the product

development is what characterizes
this intensive use of models.

So my talk today.

Is most about model based design but
I'm going to start with with two

smaller topics and
start with a little bit on the economy and

a little bit on you know my
microphone is not on top of it.

There we go.

And so a little bit on the economy
a little bit on innovation megatrends

I'll talk about how those
tie to model based design.

So as you know we're currently recovering
from a recession right now in the U.S.

economy and the worldwide economy and

there's a lot of gloom in fact
about the state of the economy.

People talk about the new normal.

Here's a quote from McKinsey and
Company and

there's there's some thought of it
being different than other recessions.

You know a restructuring
of the economic order.

Quote from Steve Balmer C.E.O. of
Microsoft he talks about the new normal.

A Brookings talks about the new normal.

And he talks about need to talk about new

normality being a period
of lower household debt.

Higher personal savings less consumption.

OK So the big question then is
how long is this going to last.

OK so Michael Mandel is
an economist with Business Week.

And and this is his viewpoint
viewpoint of want to Coniston I don't

want to say it's not so you can't all
economists I'm sure they all disagree but

here's the point of one economist and
how long the New Normal might last.

He's pretty gloomy too.

He says consumer demand will stay saw for

years because it's going to take
us a long time to dig out of the.

The deficit.

OK.

Whoever he also goes on to say
without innovation it's a slow slog.

OK so and so point he's making is
innovation is the wild card but

we have all these factors lining
up against us in the economy.

Innovation is the way out new products

will will boost support to
end supply create jobs and

innovation is the missing ingredient
towards getting the economy growing again.

OK So
let's approach this from a different way.

Let's look at this innovation
with different way.

This is the gross domestic product per
person over the last eighty years for

the United States now.

I like to think of the Gross
Domestic Product as kind of imagine

the United States was a single company
comprised of of a roll up a company

was rolled up of all the companies in
the United States every single you know.

Education Institution corporate thing you
put all those together every restaurant

and you add up all the revenues.

That's really what the G.D.P.
is OK and this is the G.D.P.

per person and it's increased six
fold over the last eighty years.

So so the of that business amount
of revenues in constant dollars

per person increase it for.

That's an enormous number.

OK So why did that happen.

I like to suggest that
a lot of it happened

starting in the eighteenth hundreds.

When fundamental discoveries were made
by scientists in the areas of chemistry

physics and biology to to to determine
the underpinnings of science and

then in the late eighteenth hundreds and

in one thousand nine hundred engineers
turned that science into innovations

into concepts into inventions
practical things radio T.V. computers

farming equipment automobiles airplanes
and this amazing one hundred years ago.

You know we'd be riding around on
horses most of us would be farmers.

OK it's true.

Extraordinary how far the world has come.

In a hundred years and

in terms of the terms of that and you know
I think we owe it all to scientists and

engineers scientists for figure out
engineers for turning these things into

these innovations that
have led to this growth in

output per person of lead to us be able to
sit in air conditioning rooms and you know

have leisure time to be able to travel
where we want the world and so forth.

OK so so so so this is you know
arguably this is what's what's going

on with in terms of innovation driving
the growth in the G.D.P. and the economy.

OK so what types of innovations.

Well here's an here's an interesting list
this is a list from the National Academy

of Engineering where they've taken a
position on what the twentieth century is

greatest engineering achievements are and
look at this list.

You know number one is electrification.

You know what the automobile computers.

Look the Internet's thirteen.

You know we all think the Internet is the.

You know the bee's knees or
whatever the latest thing and

it's only thirteenth on this
list relative to the significant

innovations of the of
the twentieth century.

OK.

The.

But so

that what one of them is missing from the
list that would be my personal vote for

the most important one is the transistor
and I know that there are when we went

through the year two thousand there were a
number of people that said the transistor

was the most fundamental invention of the
innovation of the of the twentieth century

century that you know the transistor
has been has enabled computers

electronics telephones spacecraft Internet
all these things that's a huge enabler.

Did you know that every year
there were sixty sixty million

transistors manufactured for
every person on the planet.

OK So these these trends that you're
just going to hugely impacted on

on all these other mega trends.

OK.

So that's that's.

Twentieth Century.

OK.

What I'd like to suggest now is
to make the trends that we see

to to make a trans happening now happen
over the last ten fifteen years here

that are affecting us now currently
that I want to talk about.

And so the first major new mega-trend
I want to talk about is what

I'm going to call software and everything.

OK And we can illustrate this
with this with this chart.

If you look at the bottom of the chart
that color line near the bottom.

That's the number of computer
microprocessors that are being

manufactured every year for computers
like this or desktop computers for

use in standard you know running Windows
or Linux or something like that the mac.

And that's growing dramatically but

knows that it's dwarfed by the embedded
processors on that top curve.

OK.

And.

Which That which that in two thousand and
eight last year for

which you have data there were
there were ten billion shipped.

OK And this this process of
these imbedded processors

are really put the process of putting
software into everything around us.

OK So and
it's a dramatic change right now of

getting software in consumer
products all around us.

But here's an example look at the in
the one nine hundred eighty S.

Here's a typical automobile no
airbags you know knowing like brakes

you had a cassette tape for your radio.

OK.

Aside from the radio was there
even a transistor in the car.

You know I I don't think there was and
verified that but so

in the radio is probably the dominant
source of transistors so and

that was all it was just thirty years ago.

Let's fast forward to today.

Today's automobile.

Well it has software
controlling power train systems

software controlling the chassis.

Transmission brake system
traction control things like that

software controlling safety systems
airbags crash avoidance and so.

And software controlling
a convenient systems

things like climate controls
windows things like.

OK And phenomenal amount of
software in fact I've heard numbers

between fifteen and twenty percent
of the cost of a car is software.

If you buy a car.

OK if you've been out of real estate but
a car is probably the most money you'll

ever spend on software in your life you
might think you know buying games and

things like that is software it's not
compare not even close to what you spend

when you buy a car you don't think of cars
and software product but increasingly it's

the differentiating factor determines
why you choose one car over another.

If you like you like some of
the features they have in electronics

they like brake system traction control
something like that will get you to buy.

The M P.

You know the M P three connector or

whatever you know there's
a lot of stuff in there.

OK so.

So that's an example of
a place that's tremendously.

But more generally software is appearing
in virtually every consumer product out

there and and this brings me to number
two it's not just software in everything.

It's more math and algorithms and things.

OK more math and algorithms and
consumer products and this is good for

you guys universities who
were full algorithms and

full of mathematics and
those conceptual ideas years ago it

was very difficult to get that
mathematics actually used in things.

Now with software everything.

It's much easier you can
come up with an idea.

You know software sort of liquid
innovation you come up with an idea

of an algorithm.

You could make it work in software.

OK you know take something like
you know the i Phone the i

Phones using all rooms
all the Multi Touch.

There's mathematics behind all the Multi
Touch actions you do on the i Phone

communication systems and
things like that a digital camera.

Obviously the autofocus systems.

But there is you know lots of features
in there there's features in there for

recognizing smiles and knowing when
to take the picture to get everybody

in the group smiling it's you know image
processing and you know the pattern

recognition and things like that so
increasingly algorithms in the systems.

And these are often put on chips.

When you put the albums on a.

Chip within these systems.

However it can be more complicated
than this IT systems level.

Sometimes are very often these subsystems
are put in larger systems in up

with systems of systems that have multiple
subsystems with their own separate designs

and and these are all in some
cases being linked together you.

Well OK but

overall the systems are smarter they can
have active systems rather than the past

the systems you had in the past systems
that don't have a human in the loop.

Collaborative systems that work
together well to function systems so

huge trends towards increasing algorithmic
complexity really over the last ten or

fifteen years in terms of this being
deployed in the software area.

Well OK so how are we going to make how
we're going to design these systems.

How are we going to make
the software in these systems.

This is the place where there's
been a really big change but

I want to talk about the traditional
process that was used.

Maybe ten years ago in industry before
model based design came along and

here's sort of a process workflow for

making this it starts with requirements
and specs which were done in paper form

paper specification for
what YOU for the algorithms.

Then there was a design phase which
often involved physical prototype

to try things out on.

Then you wrote code where a bunch of code
and then after was kind of all done you

tested it to see if it would work
that was the traditional process.

However because of the increasing
complexity there began to be a whole host

of problems the industry was noticing
it was becoming fantastically

expensive to develop this stuff and
they were failures.

Probably the most spectacular and
the most famous software bug ever.

Is it that at least and
that I've heard of is

the maiden flight of the area on a five
launch vehicle people know about this.

Just curious myside it's a couple hands.

OK so this was the first.

Launch of this launch vehicle
was Flight five zero one.

And about twenty seconds after liftoff.

There was a malfunction
software malfunction and

they had to had to destroy the.

The launch vehicle cutaways was
going to come down hit them.

I guess or something like that.

OK And so

what happened in twenty seconds afterwards
is there is an integer overflow.

OK so the energy overflowed and
they did a you know a fault now so

the whole event but
the energy flow flows in the guides and

control system that was controlling
the vectoring of the nozzles OK And

what happened was they had
taken the guidance and

control system they used on the area on
four and they put it on the area on five.

OK And it was.

And and so but what happened twenty
seconds after lift off is the Horan's all

of a lot because it was an error in
flight was a larger rocket the horizontal

velocity of the whole platform was
larger than had been expected or

had been modeler and so
that horizontal velocity overflowed

the sixteen bit injure and once you were
full of God's control system like that.

You know just complete loss of control.

OK so this was a this was a systems
fail or it was a requirements failure.

You know the requirements weren't
right there wasn't really proper

modeling done on this.

There was a host of different
things that happened there.

And it wasn't really a coding error.

OK sometimes what a times you find
in history when you have a which

what's called a bug.

Most of time is not pure coding errors
usually there's a system of requirements

failure at the root of what caused that.

OK so so significant software failures.

Another one is the expense of writing
software expense of writing the C.

code that goes into products like this.

This is an artifact from code and done so
it shows their development process and

you can see it has passed
over on the left side and

you can see that software
design is a code.

Box.

You have to once you've designed it you
have to verify the code line by line.

There's two coding parts in their
schedule and certainly in Denver.

When they presented this
they were trying to make

their development process more efficient.

So they could products out faster and
the coding was a real problem.

It was too time consuming too expensive.

Introduce defects and
if you know Toyota they hate the facts.

OK They're out to drive those out of out

of the system with their air
quality approach to quality.

They also didn't like to
many verifications state so

they had that if they pointed this diagram
and said can we make this go away.

You know can we can we fix this coding
process which is very inefficient.

Another problem could portability.

If you do a design and you targeted
something like a C six thousand or

a an ARM processor and so
they want to go work on another platform.

OK if you've written
a code by hand you out.

You're not a port we have to start over
again read all that code over over again

and with the proliferation of all
these target platforms out there

that was that lack of
portability was a real problem.

OK so if we go back to
the traditional develop process.

You know there.

Those were some of the problems
encountered but at the requirements level

the probably the biggest one was
paper specifications are ambiguous.

You know they're they're not precise.

OK And
there are multiple versions of truth.

The design phase.

You know it was physical prototypes which
were expensive implementation we just

talked about the problems with
Daniel coding process and

then in what happened is in traditional
testing you get to the end of the game

you do all the testing and
you find errors late in the process.

You often find problems
with the design and

you lock down a design
discover a new way to do it.

Late late in the game.

And so these were real things that
industry was telling us about we were

talking industry out there you know
because they were using Matlab but

they were playing to these things and
saying can you do something about this.

These are some of our real problems now.

It's in a lot of the.

Through these large teams in aerospace and

automotive and other places like that
in fact these were separate teams.

It was there was R. and D.
did the requirements.

You know you had a set of design that deal
with the design element you had software

engineers that did the implementation and
you had technicians that did the testing.

And so they were actually
different teams of people.

These were huge barriers
between them of communication.

They tended to throw paper over
the wall to the next phase and so.

So there were communication issues there.

But also gaps between disciplines.

You know within the design teams you
know you had mechanical engineers that

is software engineers digital or

tries are a fan want that were
also gaps in the design flow.

Here's a Power Point from
a large aerospace O.E.M..

That they presented at a conference
we had in five or ten years ago and

in there they were trying to again
describe their problem to us and

they talked about systems engineering and
software engineering and

see that see the fingers pointing
each other there in the middle.

OK So what happens is a problem
arises results don't agree.

And so everybody of course says well this
is a problem on the other side you know

there's a problem in the code or
there's a problem.

The simulation models OK.

You know the requirements and and
that sort of a slow process of trying to

resolve these things in or through
that was the one that was the problems

the industry was having OK So
what are some solutions to this.

Well in terms of solutions I
want to I'm going to focus

on kind of two aspects to it when
aspect the solution the model based

design I want to talk about is
response to these megatrends So

they're being more software and
everything and the other one is response

to the problems being encountered
in those development processes.

OK so so by a similar aspect

it was happening in other areas too and
I want to sort of used to talk about that.

By analogy.

Suppose you want to build a building.

OK these days.

A large building.

Well it's very clear you wouldn't rush
out and just construct it first model it.

OK.

Everybody knows that.

Similarly suppose you wanted
to be interested in building

a hydraulic braking system for a bicycle.

You would just go to the machine shop and
hit the wave on that.

You model it.

OK do some solid modeling they're OK where
if you want to design a total aircraft.

Well.

That involves writing code so the wrong
thing to do is just jump in there and

start writing code OK's what we're trying
to say in fact the obvious solution is to

do some modeling modeling of a system so
different type of modeling.

You know it's not it's not the static
model you get in solid modeling or

in buildings.

This is Dynamic Modeling and

you need to have differential equations
difference equations things like that.

So in order to model a complex
system like that you need a healthy

library of system models in fact we refer
this is multi to main system models.

If you believe it represent all these
different domains and I like to go through

a set of the different types demands that
we think are important in terms of doing

and in system models in order
to build systems in industry.

The first to mean is
continuous time models.

This is these are the block
diagram the control engineer might

use the system flow diagram.

These represent dynamic systems analog
behavior Vironment models the underneath

them is are ordinary systems of ordinary
differential equations nonlinear

obviously is the special case that
is linear forms of those as well.

So these are continuous time models or
are something that needs to be in

the in your toolkit of models Another
want to discrete time models.

This is the language of signal
processing image processing

video processing digital control
all use discrete time models.

A third area is discrete event models.

This is used in for servers in queues and
having delays in and

and can be used for communication
busses and things like that in systems.

Another type are so-called state charts or
state machine models.

These are for modeling mode logic or
things that have discrete states

this particular picture here is
a state chart for a power window and

automobile when you have different states
like UP AND DOWN AND emergency down in

obstacle detecting things like that
it's not a it's not a continuous

time control type thing but it's a set
of actions to take with certain states.

It turns out if you go into an automobile
in fact this is almost the dominant type

of control that happens in in a system
like a car as all these you know

as discrete events Your when she
wiper positions your cruise control

positions are all just the things have
to happen in those different modes.

Physical models is another
important area and this is an.

This is these types of models where you
know not really interested in differential

equations you would like to work.

You might like to work with resistors or
capacitors.

Or perhaps at the level of a motor where
you have an aggregate of those and or it

could be hydraulic models or a mechanical
model where you have linkages in joints.

That connect things up in
this case the lines are or

are bi directional they
represent a pressure or

a voltage there not a single
direction like in a system's model.

Another kind of models there
are important text based models sometimes

algorithms are best described in text form
they're not best described graphically and

this is an example of about
fourteen lines of matlab code

that implements an extended
common filter and

a common filter is something that's
just best written in matrix form.

That's the way you want explained that's
the way it was theories presented that's

the best way to describe it and

it only takes fourteen lines amount of
code to implement something like that.

So it's important able to have text
based models in terms of Europe and

the taken especially when
you're building out rhythms.

So we have the math works have
been working really hard over the.

Last twenty years in fact to build
a single modeling environment for

multi-day main systems it's been
a great sort of you know you guys

do you are your research here in various
areas in some ways this is our research

project that we've had going on for
twenty years of how to.

How do you mix all these systems together.

You can find a lot of software
out of those individual

pieces of these modeling domains.

OK and one at one of the project we've
been working on is how do you have

all work together.

How can you simulate some have something
that can work all these things can work

collectively and you get the interplay
you get the build off of you can.

You can apply a common filter to a circuit
without having to you know having

the linkages and things like that.

And so we've been really
working hard to Phyllis and

it's only recently that a lot of
these different software modules have

become available to support
these different to Mainz.

And and it allows you to do end
in simulation of something like

a tilt rotor aircraft that has you know
landing gear and hydraulic systems and

computer systems and you want to model
the environment things like that.

And so.

So what you use these
modeling to Mainz for our for

all that is system elements you know the
mechanical systems there at the bottom.

There's the software in the embedded
control unit the analog

hardware digital hardware and
also the environment.

You know you're not just modeling what
you're building your modeling something to

test it in the environment might be
well Hudson examples that in a second.

So so so the modeling.

So one part of model based
design is having an MA

environment they can represent
many different systems.

Well to give you an example here of this
of this these parts working together.

Well this is a picture of a wind farm.

Which in itself is a system but

I'm going to pick a single wind
turbine here to talk about.

And here is

Jim can you tell me the pointer here
point out some of these things.

So here.

This is an old.

Diagram of the wind turbine if you look
on the top part you can see the bull

a bottle of the blade in the drags and
and the blades below that

if you look down below that you find in
the cell that's the gearbox that has the.

You know the bullet shape thing
that you see up on the tower but

no that's the tower.

You can actually see we you know
model of the connection to the grid.

You know in terms of electrical elements.

If you go over the left side of
the diagram you find the wind.

So we will model of the wind
the input to the terms we have to

you know have it have an input response.

And to the right of that
is the main controller

which is going to have
several different states.

And then there's some.

Continuous time dynamic controllers in
the yard controller pitch controller

there in the middle so.

So this is really a block diagram
that's model the whole thing.

And let's let's dig into some
of the elements here and

let's go look inside than a cell.

OK in here you can see a gear train
model which is done as a physical model

you see a generator which will
be an electrical model and

there they are being connected up
to the various mechanical parts.

We can exploit now at the environment the
environment consists of two parts one is

the winter been the wind
on the left side and

the other one is the grid
on the lower right side.

So we'll open these up and
so if we open up the wind.

You can see we specify the wind
speed as a set of points.

So we're going to have the wit
an increase from two miles an hour or so

up to around fifteen and then drop and
the wind direction will vary a little bit.

As input to the model in the lower right
you can see we've got transmission line

connected to some of the motor
there on the grid so

that's how we're doing a model of that.

But the main controller The thing with
the S one S two S three in the middle

is the mode control logic
we can look inside that.

And here you see a state flow chart and

the major states of this controller
are it starts in the park mode in the.

Or left.

It's going to go into
start up in the middle.

And then generating on the right.

And there's a break mode down below.

And so this is really for controllers and
all that have to do with sequencing

between needs different states we can
now look inside to your controller and

here we get you see is a simple idea
controller implemented to control and

regulate the blade angle as well as
the direction of the of the in the cell.

We could run a simulation of this in
Simulink of all these electrical physical

components state chart components
working together and and

so it's rich running right now when the
wind speed goes about three miles an hour.

The blade pitch command goes high and
the parking brake releases.

Are you can see the blades turn in the in
the simple crude animation on the right.

The rotor speed starts to increase.

You can see in the top right scope and
then as the wind direction shifts.

Than a cell will rotate to
track the wind direction.

You know you can see this in
the red line the top of scope and

you can see than the cell moving
in the animation the cells

going to lag actual wind direction because
the controller limits the maximum torque.

As the wind speed varies the blade pitch
is regulated to maintain the rotor speed

optimum fourteen revolutions permit it.

You see in the operate scope as
the interaction shifts again the cell in

the animation rotates to the right.

And then when the wind
speed drops below three.

C. the wind dropping the blade pitch
will be the blades will be feathered

to minus ninety degrees and
which you can see in the animation and

it will come to a stop here.

So so it's a fairly complex system
model going through multiple modes of

start up you know it's not just of
this is not a linear control problem.

You know this is a more
complex system model here.

That's modeling things the way
mechanical electrical type people would

with might build a model.

OK so I talked about software and
everything and more math and

algorithm systems as being real innovation
mega-trends and have suggested that

system modeling is an important response
to these megatrends that industry is using

in order to be productive and
to rise to these challenges.

This is a second part to it though
it's a system Ali is not enough.

There's also a deeper process in bed
then that's called model based design.

And let's go let's go back to our
traditional thing with the walls there and

show how model based design changes
this traditional development process.

But the text based requirements
is replaced by a model.

And the industry calls this
an executable specification.

So rather than a paper spec that you
can you know if you get a mistake.

You don't even know you've got a model so
you run the thing to see if it works.

You know Malia to be very precise it to be
very precise in terms of how you describe

the requirements and also becomes
a spec that the whole team can use.

It's not just done by R. and D..

It's a whole system includes
the environment and

it's going to be used throughout
development in fact test starts right away

you start validation you start testing
to make sure it works right away.

Don't not wait until the end.

Here's an example of a three G.
wireless system.

The thing to point out the there's a base
station and that is the top box the middle

box is the handset user equipment and
then on the right you see a channel model.

So there's the environmental model
connecting it to that that's part of

model based design.

If we go to the design stage design.

It is replaced by a process of
taking the initial model and

adding behavioral descriptions to it.

Adding design details things like fixed
point timing component interfaces things

that represent the hardware and
you have a chance at this stage

to do design optimization to try a bunch
of different things different approaches.

It's very flexible stop being done in
hardware it's being done in software and

you can find problems before
you go build with them.

Here's an example of a decoder
which is a signal processing

up algorithm and if you look on here
you can see there's down the middle.

There's three or so different versions
of this right here we do coder and

on the left side there simply being
difference to compare it with the output.

If you look at the top one here.

This one is the golden reference.

It's written in a Purpura
mathematical form.

The way from a pure algorithm standpoint

this next one at turns
it into fixed point.

So it's modeling fixed point effects.

And the third version is adding
hardware pipelining these

are the successive elaborations you do
to turn something into a real design and

within this single window you can compare
the results of all of them to see how they

do relative to these real words design
constraints that been added to them so

it's a concept of what happens in design.

When you go to implementation.

Rather than writing code by hand.

This is one of the more dramatic
parts of model based design.

In that there in that hand coding is
gone and it's really transformed.

You know the how you know like your
car with it with a million lines of

code that into my anymore it's auto
generated now from models and it and

the wonderful thing is nobody can write
forty thousand lines of code without bugs

just can't be done and
this bug free code comes out.

I mean the problem Yagur of the and
it's not going to fix that but

the code will be correctly represent what
you have in the algorithm and that's a.

Huge cost savings for industry and

Abel's it eliminates the gap between
software and hardware engineers in

terms of implementing algorithms were the
domain experts are the main experts and

software engineers in terms
of implementing algorithms.

Here's an example from Honeywell where
they're using out of a code generation to

for a million lines of flight
navigation flight control code

in this case on it on a business jet and
you know just enormous to

many many cases in history have design
teams they use the gold the new method and

they made actual comparisons about
which one was more cost effective.

In the last step is
the testing verification and

instead of doing the testing at the end.

Well OK The design is model aberration.

But instead of testing at the end
the testing is done all the way along

the process and the models are the
requirements the gold references and

the testing is done throughout that
process rather than being left to the end

and you can actually reuse a lot of test
suites throughout the design process.

So so that's an overview of
the workflow changes that happened and

the different elements that and
we're meeting with our When we meet

with our major customers they typically
point to one phase of this workflow and

they say can you add this here.

You know they give us.

They're all trying to improve
this workflow in ways that

they can tie requirements to to
the code and various other things like

that that help them more more
efficiently use these processes.

Actually one more thing
that might say on here.

Well.
Well the first was

back in the early days when we were
first rolling out designing was big and

was being when I went to visit
an automotive manufacturer in Germany and

they they took me out to drive in
a car that they were running some

they had an active suspension system
that they were had electronics some

motors they put on the front wheels
that could lift the differences.

It's up in a way to so if you turn right.

You know cars do this in the motors would
push up and straighten the car out so

they would take me out on the on
the autobahn to for a test run on this and

that is kind of scary because the guy was
typing at a computer mounted in the front

seat he was driving on the Autobahn about
hormones are taking on a keyboard so much.

I found a little unnerving.

The when he was turning
hard to show how it worked.

You know sort of going back and forth.

OK but
it was the code running in real time.

Actually I think part of what he
was doing was is he knew that I and

other people matter is responsible for
the code running in real time and

he wanted me to feel the safety
aspects of driving out

if we're writing code that
matters that created there but.

Anyway.

What he what the story he
told me is when he took his

first model based design product
using that process to the test track.

They've been working in the design
labs for you know a few months prior

to this and they're ready to go
test it out on the test track so

they they go out there you know with
their software and they go out and

get in the car and
they were on a test track and it worked.

The first time.

And there was there was a couple
technicians helping out and

technicians looked at them and and
these were old grizzled technicians and

they said this is the first time in my
career an engineer has brought something

to the test track that
worked the first time.

OK because they're used to engineers
would show up with a bunch of code

written it had really been tested very
well and you know and but now this thing.

The reason it worked
because it had been tested.

They didn't run against models.

OK so they worked out all these scenarios.

OK You know you get out of the car and

you're left with a few science you might
have forgotten at the last moment.

You know and some of the interfaces and

things like that typically don't work out
and so you know it's a dramatic change

to how how this stuff is developed
in the automotive industry.

I Caterpillar uses the model based as I.

They they compared regular methods
with model based design they found two

extra forks reduction in person hours and.

Is huge for a company that's a huge
reduction in development efforts.

And a reduction of calendar time.

OK In terms of time the market.

Caterpillar and

other companies often talk about going
from manual calling into model products.

And they talk about you know we you know
modeled all all you know hybrid control or

something like that is that the knacker
they use is a model system.

Here's a news article motive news
about a year ago talk from talking

about General Motors that I just want to
kind of you know just give you an example

of where some of the stuff is used this
is a quote by the director who's in

charge of the software for power train and
his middle com is I don't think you can do

a hybrid control system
without model based design and

development the name of the game
in hybrid is integration.

You can't make an engine and
transmission separately anymore.

And then integrate it into the last
minute that's to be conceptualized

as a family is a system system
is an exception as well.

You're in products not
going to work very well.

But other companies like Toyota to

one of their fundamental beliefs as
a company is really design innovations or

early design process is really is a
process before they walk into design they

want to try a lot of things can many
companies commit to a designer early and

then work it out and make changes to it
you know near the end to adapt to it but

I would like to put try to look
different out of box things and

the modeling really helps that really
allows them to do these multiple design

iterations on paper where they're not as
expensive as once you go to hardware.

Here's a study by or if you do little that

shows the trends in amount of time
spent in different activities and

you can see that the these two curves
which are implementation in test

are trending Dow they're industries doing
lest implementation and test and so.

The economy will support is Will companies
to companies know in order to be

successful in growing their own grow
their own revenues in their economy

innovation is the way.

And so the model bases and
gives companies competitive edge is here

in terms of having really
design interation it's.

We have our customer
base is very interesting.

We can break it into kind of
well loosely into three camps.

There's industries that are very
collaborative aerospace not a lot of very

collaborative they don't they're not
trying to compete on using our tools and

they have industry meetings where they
share lots of information about how to use

our tools and we work with OK.

There's the finance industry to use our
tools we have no idea what they use it for

they won't tell us a thing.

OK completely secretive.

OK.

And then there's a number of consumer
products companies who are individually.

Unbelievably secret some of adopted
are because they feel like they've got

a real edge in using our products and

they're sure their competitors haven't
caught on a model based design and

they repeatedly make us where
we want to mention their name.

OK you're the financier they don't
mind if you mention their name.

They don't tell you what they do but but

some of these consumer companies they just
don't want it known that you know they may

be on a development process that you know
is five years ahead of their competitors.

So that's sort of what we see out there
that the secret ones are actually hard for

us because we can't get many requirements
for what they need the ones that

collaborate more like automotive it's
been a very good industry for us because

they're very aggressive as you seem of
all the software in the automobile and

they're full of ideas and
collaborate very well in that.

So another key driver model
based design has been to

eliminate the manual coding that's
a huge thing to not have to

write their code you know some companies
model based just for that step.

Improving quality is why some companies
adopt model based as zine quality

is very important and the really early
the envy helps so that the collaboration

across disciplines because there's
a common modeling environment.

You can build a shared models.

Clabber development stages.

You mentioned before how it used to
be the walls between the groups.

So this is allows everybody to have
a goal model they can work for

you go to the model rather than going
to a piece of paper some place you go

to the model to find out what's the truth.

So this is a summary of model
based assign an industry.

If you switch now to education.

There's a curriculum survey that was
written came out I think last year

in the control system society was
a survey of industry University about

what should be being taught
in controls engineers and

their list first slide is pretty
good agreement between industry and

professors in terms of the top four items
on hands on experience in history focused

design computer software and look
mathematical modeling of dynamic systems.

These are the top four.

It went down below forty.

Well below forty in the other ones.

And there's another question about what
types of models should be being taught in

school here there's quite
a bit of disagreement.

Unlike the previous slide.

There was agreement on linear models but

after that it started to diverge
where industry wanted controller and

it models sim simulation models non-linear
models finite state machine models and

real time models for hardware in the loop
and so there's a gap there between

what in a few things at once and
what education thinks you should have.

So one of the reason brings to attention

is a lot of these things are you know I'll
tell you why this is showing up here.

It's because the model
based design industry.

You know this is what industry introducing
model is designed to use things and

they're there they're hoping
that there's more training.

It's interesting.

Japan is actually implementing of the big
curriculum to try to train train what they

call M.B.D. engineers trying to develop
curriculums to train this methodology so

people can hit the ground
running when they get industry.

Up models based designs being used
in education a lot of time a lot

of cases been in project design.

Design projects and project courses
like the eco car in the lower left

to simple things like Lego
types of experiments.

So the project course is certainly
one place that it's being used.

Well Kate.

So the key idea tried to present you
present to you today is that model

based design has has really transformed
how complex systems are developed

an industry and it's happening reasonably
suddenly over a ten year period and

not that well understood by you know
people in education in a lot of ways.

And the second one is my conjecture for
the future.

I believe the model based design is going
to transform education we Mathworks has

experience with how matlab transformed
teaching of computation in different areas

of engineering over the last
twenty five years or

so you know from the advent
of the personal computer

producing software that can be used for
teaching we've seen happen in matlab.

And it's my belief that model
based design is going to have

a similar effect in education
over the coming years.

And and here is you know some of the areas
in which we think it will will have a role

in teaching will will will have
an opportunity work with models.

Directly in teaching with those.

Project based learning I mentioned you
know a lot to educate lot of design

a lot of the interest in
education innovation.

You know there's courses in it there's
entrepreneurship courses things like that.

So that's an element of it you know
trying different alternative designs

another thing education education
is full of fancy mathematics.

One of the question is how do
you get that math into systems.

Oftentimes you learn in school you go out
in history in the math isn't used and

now with the power software out there.

You really can get those more
advanced algorithms out there.

And I think that's great for education
certainly limiting low level coding.

I mean nobody here wants to do low
level coding is not something that you

know there's been an obstacle
to the use of step to.

The design in the past.

And so we're getting it out
that gets rid of a major hurdle

in terms of trying a lot of these things.

But the club aeration I know you
know universities always try to

you know it's Goldman you
collaboration between different areas.

So it's an opportunity to do that within
the modeling environment and then

preparing students for industry because
this is the type of stuff industry wants.

So I guess in summary
it might my talk here

just kind of comment going back and
talking covering some things I talked

about the economy is in a recession
the way out is through innovation.

Innovation is the key driver for
the economy and

our standard of living both as companies.

And as countries in fact.

I like to tell people
the world is counting on us.

Scientists and engineers to drive
this innovation and continue.

Our call our standard of
living in the economy.

It's coming from science and engineering.

I don't believe anybody else who
tells you anything different.

So my opinion software software and
everything and more math and algorithm

systems are two important trends that are
happening over the last fifteen years and

they're mega trends they're big and we're
seeing embedded systems in everything.

And companies with high tech
products have to respond to this.

It's not possible to not put software in
what you're doing these days and that and

finally the system modeling in model based
design are a response to these trends

they're countermeasure to help with
companies building products and

responding to these
rather dramatic trends.

Thank you for listening and

fine.

We have

stopped teaching

what coding stop teaching coding.

My pain on that I have pain so
that the biased.

I think that I think computer
computer training is some

languages are essential for
people to get industry.

I think there's I think there's
two that ought to be learned.

You gotta learn a computer science you
have to learn matlab of course I have to

say that that's a problem that's a problem
solving language for solving problems but

you ought to know either C. or Java too.

And you get industry you're really
you know we live in a computer era.

You really and I think Java and
that lab are the foundations for

which you can do almost anything out there
in industry but I think you're handicapped

if you don't know either of
those to be honest the right

place where

they're right.

So the question of verification of
validation is actually one of the leading

sets of things we're working on
with our customers right now

we are used in in what we call
safety critical applications.

You know which is flight
control of the air spaces.

Is some of the more demanding areas and

we have a sort of we have a set of
artifacts that can help go through

search through certification steps that
are required to validate code is correct.

There are quite generators are extremely
high quality because they're used so

heavily and we have an enormous
investment in testing but

we don't have we can't prove it's correct.

OK So but we're we are working on
validation tools to help ensure that.

So I don't.

Jim you know we used in new clear critical
applications I can't speak to that I don't

believe we are critical of
a lot of your question is

are these are the same with our love that.

We sort of use ways or.

That you take advantage of that you

are sure of that.

Yeah I mean I suspect there are code
generator exercise more than

the codes that you're talking about
just aren't our entire customer base and

you know so some ways just using something
heavily gives us some confidence

in it although that we're working on
other ways to try to triangulate on that

what you need using yours.

He's hired

Russian Russian That's a tough one.

A little above that.

Well I give you a related one what
the maybe a little bit of that

one of things we've encountered this week
is you take this tool set you put it

engineers desk and
they can use it any way they want to and

what happens is organizations
actually don't want that

they want to restrict they want to
limit it in certain ways they want to.

They want to come up with guidelines and
best practices.

OK So in some ways the next level.

Conceptually is a set of best practices
about how you model things you

know how you put the pieces together and
how they work and that's probably the area

that's of most active interest to a lot of
our large customers right now is so-called

guideline development to you
know ensure best practices there

above that I don't think I have a
particular quick answer on that that I can

think of beyond that with
a good history with the going

that we started with all of

the discussion of all of your love.

But there we sort of locked our
brother jack of all trades by

what it was you mean because
you're liable for your systems.

So you don't want to lose that.

So the question is how do you do you

let people there while they did so
there's an interesting upward pressure.

There's a writer Richard all

or that is the more

first to the multi-core
questions you absolutely right.

The multi-course or
make programming more difficult.

I think that probably depends
what the what problems

the universe you're trying to solve for
a lot of stuff you know if you're using

Matlab to work on an algorithm multi-core
most time isn't that important.

You know you're working on time to insight
or time to discovery and actually speeding

something up a little bit you know your
most important thing is how long it takes

to write that rhythm down and simulate OK
So a good portion doesn't really matter.

It's the response a part of
it is response to the vendors

like us to make sure we use
the multi-core you know so if you.

You know if you work with them.

You know on a large matrix computation and
you have a multi-core it's going

to go you know ideally you know four
times faster if you for course there so

we can provide a lot of that built in.

And we're also trying to provide
construction matlab that allow you

to take advantage of that on the Simulink
side we one of the key elements is

be able to partition models into different
parts that would run on different course.

So what we can provide tools
that you if you're trying to

do a design problem you could
separate the different parts there.

So I think.

Really.

Unless you're really interested in
parallel processing research the best

thing to do is is to kind of wait for the
tool sets out there to support that and

give you the freedom to play with
those the way you play with the other

elements of the systems you

made where you're

very OK So

the quest questions about how to integrate
with other types of models and packages.

There's a whole ecosystem of kind of kitty
out there you know we we we have a variety

of connections and links we do to other
modeling domains and environments and

whatnot that we try to you know based on
customer man develop ourselves customers

or develop their own other vendors or some
of these other modeling environments built

their own connections so largely it's a
tremendous amount of connections you know

one of our one of our fundamental design
tenets is a company is extensibility So

we're we have a pretty extensive A.P.I.
for adding things in there.

You call Java directly from inside
matlab you know just to any any any Java

function you need.

I mean I was looking for
you know I needed to find my P.

address in my computer an automated
fashion so I just typed in Google.

You know Java you know host ID IP
address or something like that and

I got a line of Java code to call and
I typed it into Matlab and

got my IP address you know so it's that's
all automated biassed by providing

the seamless Java A.P.I. and you can tap
into all the java connections for example.

So it's part of our job is to you know
provide that kind of to video and so

forth but there isn't one
answer depends on the area.

We hired three

zero zero zero zero

zero zero zero zero zero zero

zero zero zero zero zero zero so.

So actually we see the opposite.

Now and
I have to in some ways I have to toaster

when I first graduated went to
college when I went to college.

When I graduated college and
started working as a control engineer.

I was doing had to design control systems.

I was assigned a programmer
to do my programming.

OK And that was a caste system.

You know my company had the programmers
and I had the engineers and

it was very clearly two level and
I didn't write code.

I gave stuff for the programmers to write.

OK that's gone most programmers are not

in the there's no shortage of programming
work nobody's lost jobs here.

Get me wrong on that but

it juniors are a lot of Usenet can
use malam cells to do their own work.

They don't need somebody else to write the
code and it makes you phenomenally more of

effective to be able to think in that
thing and you know I think that's really

what's happened we've seen with the
Simulink deployments there really hasn't

been you know the programmers
who used to write all that code.

There is so much embedded systems work
they find plenty of other stuff to do.

OK it just sort of
drudgery of implementing

somebody else's algorithm goes away.

So we so we really haven't seen that
in some ways almost the reverse.

It's by you know it's like compilers
automate creating assembling

which you don't have to be in
assembly language programmer anymore.

You know it's moving
these higher levels so.

Well cake generally moderate generally
modifying the generated code is a no no.

OK Because you.

You can't put you can't regenerate it and
replace it.

We certainly have customers they do
that and they try to automate it but

the minute you go there and you hand
modify generated code you know you do

you make it so you've broken
the integrity of the design process.

Well there are places the bill post
processors that transform it and

various ways.

So there is some of that done but

generally these things don't need to be
processed they're self-contained units.

Well a lot of time series a lot
of times code generation is

not used to the entire system it
could be used for a component or

you could have software
engineers that are building.

You know multitasking systems operating
systems all that stuff you know that used.

There's plenty of software that isn't done
by this I don't want to say that spans

all software.

I don't even come close to that but
it's particularly good for

algorithmic stuff and design stuff and
some of the mains I'm talking about but

very often it's fitting into some larger
system full of lots of other code that's

there for different reasons
many times a legacy codes.

You know other types of code that has to
be there but you want these to plug in and

basically be as components that you try to
modify it just breaks your whole cycle.

Let me ask you just a little more what

Yet all users

have world or

I really like this software

that I suppose.

Software for example factories.

You know processors have a systems.

Yeah software goes right back.

So here's my question.

We're down three or
now right where we actually have

just about everybody matter what.

All right it's all right so
conceptually there is no boundary.

This could spend very wide in reality
we have to be domain experts in

the individual areas and we have to we
have to become good at them one at a time.

OK so you can't you can't boil the ocean.

So you know we there's always an area
that you're particularly good at and

you were particularly
strong in control systems

strong in signal processing
areas communications.

OK And and but but but there's lots of
other domains that are tied to that

to particular strong automotive to control
in aerospace you know consumer products.

Like I said we've got secret companies and
other companies that haven't adopted it.

You know so there's varying degrees
of adoption in different areas so but

actually this Q one more
example take factories for

except factories use peel Cs sees
a special type of code right.

There's nothing that would say we couldn't
develop a could you here for that but

we don't have one.

So we're not useful that area.

So I'm just saying.

Conceptually go pretty far.

But yeah but we can you know we we
try to focus on one at a time and

using tarring people that know
something about that stuff and

experts in the domain that
can help us shape an area.

Yes just yeah yeah.

In fact our different domains
generate internally what we call and

I our immediate representation
of the different things and

then we we have global optimizations we
run on the I.R. In fact one of the big

problems we have is those optimizations
are starting to take a long time.

Those are code generation
used to be one second.

Now.

Now we're trying to conserve
memory execution time run

a multi-core every time we add another
layer of those optimizations it

takes longer to compile your model and so
we're not the point we have to trade you

know it's complex on Microsoft's compilers
if you if you know that you have these

global optimization flags that
will analyze the actual use of

code in the execution threads
in prison observations of that.

The product where they call that
program guided optimization.

But they can take hours to run.

OK And so we're in that game where
the performance actually does matter and

some of those optimization So there
are there are challenging problems there

that we're working on trying
to solve to try to improve.

OK.

Models to develop models.

Yet there would be some of your.

I'm thinking about.

The yeah you could measure models little
mamma and a modeling languages of sorts.

So yeah you know that's that's
a few years ahead of us.

OK One last question love.

Well.

Lynn PAC is interesting is we've
now we now fund that project

we donate money and used to be National
Science Foundation is like that but but

we now donate money to the work
there which is being done

at university Tennessee and
Berkeley as to the place it's being done.

So we're a supporter of
that that software and

it's not a static you might think it's
that problem is solved well it's not

because there's multi-core it's
come along and challenge things.

There's also these G P U's that have
come along and challenge things so

there's always new stuff to do in terms
of applying the pack type of algorithms.

So there is ongoing work there with
what's happened with that lab.

I mean I never been able
everybody to be algorithm authors.

So if you're you know in terms of a lot
of these out there people people trade

Matlab is sort of the unit of
algorithm exchange in a lot of ways.

I mean I've heard I
heard talk of be you or

somebody in their talk was talking about
you know you know some of the saying

you know all the important
algorithms have been that lab.

You know sort of you know it represent
a state of doneness you know and

they were looking for
new ones to work on in some area.

OK So you know I think that there's
there isn't a single repository but

I think there's lots of
exchange of that happening.

And certainly we you know we certainly

have active investment in algorithms
all over the place in ourselves.

But you know the world is a much
bigger and more complex place than we

can do ourselves so we try to
focus on some of the core ones and

the ones you can be built on
top of OK thank you all for

me and thank you again Jack
a very interesting to work.