Welcome back to Smarter Every Day.

This is my buddy Chad.

- Hey.

- We are absolutely giddy because we've been working

on something for how long?

- 12 years.

- Well, I'll be like that's us hanging out

but we're working on this project for how long?

- I don't know, a good four months.

- Four months.

Chad's really good at software and we had an idea

and so there's a patent pending product

that we're working on.

What it does is it eliminates kickback from handsaws.

Chad, do you wanna explain what's going on here.

- Yeah, so you'll be cutting a piece of big

a big sheet of plywood and as you get down towards

the middle of it the pieces that you've already cut

start to bend towards each other and they'll pinch

the blade and when the blade gets pinched it becomes

kind of a wheel and these teeth, since they're coming

back towards you, catch in the wood and then the whole

saw comes back at you.

- It's a big deal.

If you've ever had it happen, maybe a couple drops

of pee came out.

- Terrifying.

- You know exactly what I'm talking about.

We're gonna show how it works first with the high speed

camera and then we'll show you the fix we have

for kickback.

- [Chad] Turning it on, three, two, one.

(motor rumbling)

It wants to go.

- [Destin] It wants to go?

I'm ready.

- Three, two, one.

- [Destin] Whoa, it's still on, kill it.

What are we doing?

- [Chad] And it came out 'cause it got to the knot.

- You notice we have this chain and we've got

this danger radius there.

Will you show and make sure that the chains

not gonna hit the, yeah.

(motor revving)

(gentle music)

- Oh, yeah.

- You need something for these safety glasses.

Because of the rotation direction of the blade,

the saw is lifted up out of the workpiece and then

that rotation, again, causes it to be propelled

towards the user.

Now obviously, if this were to really happen

you would want to release the trigger on the saw

and in this test we have it zip tied so we can actually

see what's happening but I found, personally,

that my body tenses up when I get startled

and I'm not so sure that I would have enough control

in the moment to make the decision to release the trigger.

Let's look at another run and I'll stand just outside

the danger zone and I'll pull focus on the camera lens

and we'll see if we can measure the human response times

required to avoid injury.

Whoa.

- That's scary.

- [Destin] Okay, in this run, look at the timer in the top

right of the screen that's counting from the moment

the saw starts rising up out of the wood.

You can see that the blade is fully released

from the wood within 80 milliseconds

and it's already headed up towards your gonadular region.

Assuming that you don't tense up, you would need

to respond by releasing the trigger well within

a tenth of a second.

Once the blade is freed from the wood, there's no longer

any resistance and therefore the blade increases speed

because the motor is adding angular

momentum to the system.

Now that makes it even more dangerous and you can see

here when the blade falls back down and comes back

in contact with the wood, just imagine if that was flesh

and keep in mind here that this is a slow-mo video.

All of this that you're seeing here is happening

in just a tenth of a second after the blade

has lept up out of the wood.

Obviously the ideal time to respond to a kickback

event is before the saw even leaves the wood.

The problem with all this is often times humans

simply can't react that quickly.

That is no bueno.

It's a bad day, man.

Okay, so that's kickback.

It's very scary.

Now what we're gonna do is show you what we've made.

What would you call it?

It's a multi-axis detection system that can figure out

kickback based on machine learning.

- Keep going.

- [Destin] And basically dynamically break the motor.

- Yeah, so I can give you a demo

on the first one that I built.

So I'm gonna turn it on and let go with my finger

and you'll hear it turn off quickly.

See that?

- [Destin] Yeah.

- If there was no brake, it would just keep

it would keep spinning.

So I put a sensor in there to sense a kickback

so that when I do that

it shuts off automatically.

- [Destin] So your finger was still on the trigger

but you just accelerated it backwards.

- Right, this was the first one that I built

and since I just used an accelerometer

there are some false alarms

like if you bounce it around or just normal use

it can trigger and the new one that I did

I used a nine-axis accelerometer,

gyroscope and magnetometer.

So the really fancy thing that we're doing here

is we're doing a machine learning algorithm.

- [Destin] So you don't care.

- I'm not really programming in exactly

what the thresholds are that I need it to sense.

I'm taking data from dozens to hundreds of people

using a saw in normal, everyday ways

and feeding that information into a neural network

and I'm comparing,

stop grinning.

- We're cutting plywood here.

- So machine learning works by you take a big data set

of people using it in normal ways and then another

big data set of kickback events and the computer

automatically takes all those, all those data sets

and figures out what the anomalies are.

What's unique to kickback that's different from normal use.

- [Destin] So you hogged out the handle it looks like.

- [Chad] I did.

- [Destin] And so there is a legit computer in there.

It has accelerometers and gyros in it.

- [Chad] Yep.

- Three, two, one, go.

So it stopped itself?

- Yeah.

- It stopped itself.

That's what it does?

- That's what it does.

- That's awesome.

- [Chad] To better understand how the system works,

we've overlaid the data on top of the slow motion.

We're measuring nine different sensors here

but for simplicity's sake let's just look

at the overall magnitude of the acceleration

which is in yellow and the magnetic field

which is in red.

You can see that when the saw itself,

not the blade, but the saw itself starts to accelerate

the saw's braking system is applied and the magnetic

field starts to work against the rotation of the blade.

The cool thing about using magnetometers

is that the algorithm might be able to detect

the magnetic field from the motor and then make

decisions based on that information as well.

- [Destin] And so you just combined those different

things together and make it make a decision.

- Right, I just let it capture all that data.

I don't tell it what's important.

After the fact, if it's been a kickback, I annotate

the data with this was a kickback event

and then I have a whole bunch of other files

that are regular use cases and it figures out

what's important in the kickback event

to trigger output.

- [Destin] That's awesome dude.

- Yeah.

- [Destin] Three, two, one.

- Ooohh.

- It did stop, though.

- [Chad] So we performed all of those tests

by starting the saw while it was already

bound up in the wood but what would happen

if we were to spin the saw up first

before we induced the kickback event?

This graph is stinkin' awesome and it tells the whole story.

Let me walk you through it.

Okay, if you look back here on the left

you can see this red spike.

That's when the motor was first spun up.

That's the torque associated with accelerating

that blade rotationally, right, so as we move along

here and we bump the front of the saw, that induces

a kickback event and you can see that in the data

by this large yellow acceleration and at that point

that tells the algorithm oh, oh, there's kickback

and so it applies the brake and you can see

this red spike in the magnetic field.

That is the saw's brake being applied and trying

to stop that rotation of the saw blade

so you can actually see that happening and it tapers

off as the velocity of the saw blade decelerates.

This is awesome.

I mean like high five.

We just used science to stop a kickback event.

Well, I mean, it's even cooler because it's

high five and not high four which is what could happen

if the kickback, I'll shut up now.

- Okay, here's the final test.

I know how to run a circular saw the correct way

and I also know how to do it the wrong way

so I'm gonna try to induce kickback, with the algorithm

running, from my hand and see what happens.

Okay here we go.

It stopped.

This should be on saws.

If you can detect the profile of a kickback event

in Newtonian physics then you can implement

the brake that's already on the saw.

- Yeah, so the brake's already in there.

All you need is a little sensor to figure out

when to hit the brake and that's all we're doing.

- Yeah so if you think this needs to be in tools

please tweet this video to your favorite tool

manufacturer and then tool manufacturer,

come talk to us.

We'll put a link down in the video description.

- Wait till you see the chainsaw experiment.

(beep)

- I'm about to tell you something

that's a pretty big deal in my life.

It's a book called Seven Eves by Neal Stephenson.

I listen to it on Audible which sponsored this episode.

You can get Seven Eves by Neal Stephenson

by going to audible dot com slash smarter

or texting the word smarter to 500 500.

The moon blew up without warning, and for no apparent reason

which is like the best opening line to any book

for my brain ever.

The whole premise of the book is the moon is exploded

and humanity has to get off the earth.

The technology developed in order to do that,

we could totally do this.

There's a swarm of spacecraft in orbit

and if you change your orbit on one side of the earth

just a little bit that affects it on the other side

as well, right?

But if you have a swarm you have to account for all of that.

They developed something called the perambulator

and I read this book like eight months ago.

I still think about the perambulator because