WEBVTT

00:00:00.000 --> 00:00:03.419
I want to take you back a few years to a major

00:00:03.419 --> 00:00:06.179
incident at a chemical processing plant in the

00:00:06.179 --> 00:00:08.060
Midwest. Oh yeah, the one that's basically a

00:00:08.060 --> 00:00:10.619
textbook case study now. Exactly. I mean, I won't

00:00:10.619 --> 00:00:13.160
name the company, but it's a famous case in modern

00:00:13.160 --> 00:00:16.399
engineering circles. So the facility had just

00:00:16.399 --> 00:00:18.730
invested millions in this. state -of -the -art

00:00:18.730 --> 00:00:20.929
control center. Right, the whole digital overhaul.

00:00:21.070 --> 00:00:23.949
Yeah, the centerpiece was this massive, like,

00:00:24.030 --> 00:00:27.730
wall -to -wall screen displaying a digital replica

00:00:27.730 --> 00:00:29.989
of the entire plant. Like a high -end video game.

00:00:30.030 --> 00:00:32.450
Just gorgeous graphics. You could see every pipe,

00:00:32.590 --> 00:00:35.130
every valve, every massive reaction vessel. The

00:00:35.130 --> 00:00:37.210
graphics were flawless. Which is where the trouble

00:00:37.210 --> 00:00:40.170
always starts. Right. So according to this digital

00:00:40.170 --> 00:00:42.869
dashboard, the pressure and reactor vessel number

00:00:42.869 --> 00:00:45.530
four was holding perfectly steady. It was at

00:00:45.530 --> 00:00:49.119
a safe 150 PSI. Nice and green on the screen,

00:00:49.200 --> 00:00:52.420
I bet. Exactly. The simulation algorithms were

00:00:52.420 --> 00:00:54.679
projecting this completely stable chemical yield

00:00:54.679 --> 00:00:57.119
for the next 12 hours. I mean, it looked like

00:00:57.119 --> 00:00:59.719
the absolute pinnacle of industrial control.

00:01:00.060 --> 00:01:02.500
But the reality down on the factory floor was

00:01:02.500 --> 00:01:04.510
telling a very different story, wasn't it? Oh,

00:01:04.670 --> 00:01:06.849
completely different. Down on the floor, the

00:01:06.849 --> 00:01:10.269
physical reactor was literally screaming. The

00:01:10.269 --> 00:01:13.109
metal was fatiguing. The mechanical relief valves

00:01:13.109 --> 00:01:17.409
were buckling. And within minutes, the vessel

00:01:17.409 --> 00:01:20.530
ruptured. This is wild to think about. We are

00:01:20.530 --> 00:01:23.969
talking about a catastrophic failure here. Hazardous

00:01:23.969 --> 00:01:27.030
material everywhere. An emergency shutdown of

00:01:27.030 --> 00:01:29.750
the entire grid. And the operators in the control

00:01:29.750 --> 00:01:32.269
room were just staring at their screens in absolute

00:01:32.269 --> 00:01:34.790
shock. Because right up until the alarms blared,

00:01:34.870 --> 00:01:37.790
the screen said everything was fine. Yes. Their

00:01:37.790 --> 00:01:40.390
digital dashboard insisted everything was operating

00:01:40.390 --> 00:01:43.930
in perfect harmony. Yeah, it's a terrifying scenario.

00:01:44.569 --> 00:01:46.629
And honestly, variants of it happen way more

00:01:46.629 --> 00:01:48.689
often than the industry likes to admit. Really?

00:01:48.790 --> 00:01:51.250
That often? Oh, absolutely. The control room

00:01:51.250 --> 00:01:53.489
operators had fallen into this very modern trap.

00:01:53.909 --> 00:01:55.930
They trusted the simulation over the physical

00:01:55.930 --> 00:01:57.709
reality. They thought they had a digital twin.

00:01:58.090 --> 00:02:00.299
Exactly. They believe they were looking at a

00:02:00.299 --> 00:02:02.420
digital twin, but they were actually just looking

00:02:02.420 --> 00:02:05.840
at an incredibly expensive, highly detailed animation.

00:02:06.180 --> 00:02:08.180
Okay, let's unpack this. Because we hear the

00:02:08.180 --> 00:02:11.460
term digital twin thrown around constantly today.

00:02:11.719 --> 00:02:14.120
It's the ultimate buzzword right now. It really

00:02:14.120 --> 00:02:16.080
is. We're told they're going to revolutionize

00:02:16.080 --> 00:02:19.340
everything from, you know, multi -million dollar

00:02:19.340 --> 00:02:21.780
automotive manufacturing down to personalized

00:02:21.780 --> 00:02:24.439
medicine. Right, like having a digital replica

00:02:24.439 --> 00:02:26.479
of your own cardiovascular system predicting

00:02:26.479 --> 00:02:28.449
a heart attack before it happens. which sounds

00:02:28.449 --> 00:02:31.330
like science fiction. But today's deep dive is

00:02:31.330 --> 00:02:35.129
about the invisible gap between a pretty 3D graphic

00:02:35.129 --> 00:02:38.949
and an actual functioning digital twin. And it

00:02:38.949 --> 00:02:41.310
is a massive gap. So we're stripping away the

00:02:41.310 --> 00:02:43.870
marketing hype today. The mission of this deep

00:02:43.870 --> 00:02:48.469
dive is to look at the rigorous, deeply unforgiving

00:02:48.469 --> 00:02:51.629
backbone required to make these systems actually

00:02:51.629 --> 00:02:53.569
work. Which means talking about the unsexy stuff.

00:02:53.689 --> 00:02:55.789
Right, we're talking about metrology, international

00:02:55.789 --> 00:02:58.560
standards, and the intent absolute necessity

00:02:58.560 --> 00:03:01.620
of calibration, especially when, you know, certification

00:03:01.620 --> 00:03:04.460
bodies and quality system audits are breathing

00:03:04.460 --> 00:03:06.300
down your neck. Because if we don't understand

00:03:06.300 --> 00:03:09.259
that backbone, the entire concept of the digital

00:03:09.259 --> 00:03:11.740
twin just collapses. It's just a cartoon otherwise.

00:03:12.300 --> 00:03:15.539
Exactly. A digital twin without an unbroken,

00:03:16.300 --> 00:03:18.400
documented, meticulously calibrated chain to

00:03:18.400 --> 00:03:21.680
the physical world is useless. I mean, it's actually

00:03:21.680 --> 00:03:23.639
worse than useless. It's a liability. Because

00:03:23.639 --> 00:03:26.360
it gives you false confidence. Yes. The true

00:03:26.360 --> 00:03:29.039
benefits of a digital twin, things like predictive

00:03:29.039 --> 00:03:32.340
maintenance, process optimization, early detection

00:03:32.340 --> 00:03:35.500
of failures, they only materialize when that

00:03:35.500 --> 00:03:38.819
twin is anchored to reality by rigorous metrology.

00:03:39.039 --> 00:03:41.419
And the clients paying for this know that, right?

00:03:41.419 --> 00:03:43.479
Oh, absolutely. If you're a client paying for

00:03:43.479 --> 00:03:45.680
this technology or, say, an auditor assessing

00:03:45.680 --> 00:03:48.400
its safety, you don't care how sleek the user

00:03:48.400 --> 00:03:50.340
interface is. You just want the proof. Right.

00:03:50.340 --> 00:03:52.439
You care about the proof that the data is real,

00:03:52.500 --> 00:03:54.879
that it's actually reflecting reality. So let's

00:03:54.879 --> 00:03:57.280
start with the reality check here. What exactly

00:03:57.280 --> 00:03:59.719
separates that dangerous cartoon we talked about

00:03:59.719 --> 00:04:02.099
at the chemical plant from a legitimate digital

00:04:02.099 --> 00:04:04.460
twin? Well, let's look at the baseline definition

00:04:04.460 --> 00:04:07.080
first. OK. The definition we always hear is that

00:04:07.080 --> 00:04:10.879
a digital twin is a continuously updated model

00:04:10.879 --> 00:04:13.139
that closely mirrors its physical counterpart

00:04:13.139 --> 00:04:15.719
with real -time data. Right. And the key word

00:04:15.719 --> 00:04:19.100
in that standard definition is mirrors. But people

00:04:19.100 --> 00:04:21.600
often misunderstand how that mirrors actually

00:04:21.600 --> 00:04:24.959
constructed. Let's look at the lifecycle. When

00:04:24.959 --> 00:04:27.779
a team of engineers or data scientists sets out

00:04:27.779 --> 00:04:30.620
to build a digital twin, they don't start with

00:04:30.620 --> 00:04:33.459
a perfect replica of the specific machine sitting

00:04:33.459 --> 00:04:35.560
in your factory. They don't. I thought that was

00:04:35.560 --> 00:04:37.620
the whole point. No, they begin with what is

00:04:37.620 --> 00:04:40.540
called a nominal or generic model. You can think

00:04:40.540 --> 00:04:44.180
of it as a CAD file or a baseline physics simulation.

00:04:44.399 --> 00:04:47.500
Oh, OK. So it's like a template. Exactly. It

00:04:47.500 --> 00:04:50.160
represents how the machine should behave in a

00:04:50.160 --> 00:04:53.459
vacuum based purely on its original design blueprints.

00:04:53.620 --> 00:04:56.180
So to use an analogy, if I go to a tailor to

00:04:56.180 --> 00:04:58.870
get a suit. The generic model is just the off

00:04:58.870 --> 00:05:00.870
-the -rack jacket sitting on the mannequin. That's

00:05:00.870 --> 00:05:02.850
a good way to look at it. It has two sleeves,

00:05:03.029 --> 00:05:05.730
a collar. It generally looks like a jacket. But

00:05:05.730 --> 00:05:07.769
it's not made for me. It's just the concept of

00:05:07.769 --> 00:05:10.829
a jacket. That is a fair starting point. But

00:05:10.829 --> 00:05:13.509
we need to elevate the stakes of that analogy

00:05:13.509 --> 00:05:16.790
considerably for industrial systems. OK, raise

00:05:16.790 --> 00:05:18.670
the stakes for me. We are just talking about

00:05:18.670 --> 00:05:20.610
a slight bunching of fabric at the shoulder here.

00:05:21.110 --> 00:05:23.550
Imagine that suit is a pressurized space suit,

00:05:23.810 --> 00:05:25.750
and you are about to step out of an airlock.

00:05:25.870 --> 00:05:30.189
Oh, wow. Okay. If the generic model assumes your

00:05:30.189 --> 00:05:33.250
arm is 30 inches long, but your arm is actually

00:05:33.250 --> 00:05:36.689
28 .5 inches long, you cannot reach the life

00:05:36.689 --> 00:05:39.970
support valve on your chest. You die. That escalated

00:05:39.970 --> 00:05:43.290
quickly. But I get it. In an industrial context,

00:05:43.629 --> 00:05:46.050
a generic model assumes that every single pump,

00:05:46.430 --> 00:05:49.350
motor, and pipe in a facility operates at exactly

00:05:49.350 --> 00:05:52.810
100 % of its factory -rated specification. Which,

00:05:52.889 --> 00:05:54.949
if you've ever been in a factory, is never true

00:05:54.949 --> 00:05:57.269
in the real world. Never. The moment a physical

00:05:57.269 --> 00:05:59.410
machine is turned on, it begins to deviate from

00:05:59.410 --> 00:06:01.529
its generic blueprint. Because of physics, basically.

00:06:01.850 --> 00:06:04.750
Right. Metal extends with heat, bearings experience

00:06:04.750 --> 00:06:07.050
microscopic wear, environmental humidity affects

00:06:07.050 --> 00:06:09.430
electrical resistance. I mean, the physical world

00:06:09.430 --> 00:06:11.509
is messy. And this is where calibration steps

00:06:11.509 --> 00:06:13.589
in. This is exactly where calibration steps in.

00:06:13.850 --> 00:06:16.250
Calibration is the rigorous mathematical transformation

00:06:16.250 --> 00:06:19.449
of that generic theoretical model into a specimen

00:06:19.449 --> 00:06:22.500
-specific representation. specimen specific,

00:06:22.839 --> 00:06:26.720
so tailoring it. to the exact machine. Yes. It

00:06:26.720 --> 00:06:29.920
is the mechanism of taking real world observed

00:06:29.920 --> 00:06:32.819
data and forcing the numerical parameters of

00:06:32.819 --> 00:06:35.819
the model to align with reality. Right. So calibration

00:06:35.819 --> 00:06:37.920
isn't just saying, oh, the physical machine is

00:06:37.920 --> 00:06:40.420
running a little hot today. Let's tweak the dashboard.

00:06:40.540 --> 00:06:42.779
No, it's way deeper than that. It's a fundamental

00:06:42.779 --> 00:06:45.459
alteration of the underlying physics engine of

00:06:45.459 --> 00:06:48.519
the twin, right? So that it adopts the specific

00:06:48.519 --> 00:06:51.319
quirks, flaws and current state of the actual

00:06:51.319 --> 00:06:54.399
machine. Precisely. And this is why the demands

00:06:54.399 --> 00:06:56.560
from clients and regulatory agencies have become

00:06:56.560 --> 00:06:58.920
so intense lately. Because the stakes are so

00:06:58.920 --> 00:07:01.339
high. Exactly. In industries like aerospace,

00:07:01.779 --> 00:07:04.040
automotive or pharmaceuticals, you can't just

00:07:04.040 --> 00:07:06.060
claim your model is calibrated. You can't just

00:07:06.060 --> 00:07:09.290
say, trust me, bro. Definitely not. If you are

00:07:09.290 --> 00:07:11.889
running a digital twin of a bioreactor producing

00:07:11.889 --> 00:07:14.829
life -saving vaccines, the FDA does not care

00:07:14.829 --> 00:07:16.730
about your internal confidence. They want the

00:07:16.730 --> 00:07:18.350
paperwork. They want the paperwork, the data,

00:07:18.529 --> 00:07:21.189
everything. Certification bodies require absolute

00:07:21.189 --> 00:07:23.889
documented proof that your calibration process

00:07:23.889 --> 00:07:26.509
is rooted in a recognized standard. And they

00:07:26.509 --> 00:07:29.269
enforce this through audits, right? Yes. They

00:07:29.269 --> 00:07:31.810
mandate this through stripped quality system

00:07:31.810 --> 00:07:34.529
audits. Because think about it, if your calibration

00:07:34.529 --> 00:07:37.310
is off by a fraction of a degree, the simulation

00:07:37.310 --> 00:07:39.509
might tell you a batch of medicine is viable.

00:07:39.610 --> 00:07:42.329
When it's actually denatured and totally toxic.

00:07:42.730 --> 00:07:44.970
Exactly. So you need the receipts. You have to

00:07:44.970 --> 00:07:47.769
prove exactly how you anchor this digital fantasy

00:07:47.769 --> 00:07:51.050
to physical reality. And to do that, you have

00:07:51.050 --> 00:07:54.269
to navigate a very specific set of rules, which

00:07:54.269 --> 00:07:56.069
brings us to what the engineering literature

00:07:56.069 --> 00:07:59.029
treats as the holy trinity of digital modeling.

00:07:59.209 --> 00:08:02.569
Ah, yes. The big three. Calibration, verification

00:08:02.569 --> 00:08:05.209
and validation. Because looking at this, it seems

00:08:05.209 --> 00:08:09.220
incredibly easy to mix these up. It's actually

00:08:09.220 --> 00:08:11.459
the most common pitfall for software developers

00:08:11.459 --> 00:08:14.379
entering the industrial space. And the standards

00:08:14.379 --> 00:08:16.579
strictly forbid confusing them. So they aren't

00:08:16.579 --> 00:08:19.139
just synonyms for checking if it works? Not at

00:08:19.139 --> 00:08:21.560
all. They are three distinct legally binding

00:08:21.560 --> 00:08:24.000
phases of establishing a model's credibility.

00:08:24.199 --> 00:08:26.279
OK, break them down for us. We just covered calibration.

00:08:26.680 --> 00:08:29.379
Right. Calibration is estimating numerical parameters

00:08:29.379 --> 00:08:32.820
to reconcile the model with reality. But before

00:08:32.820 --> 00:08:35.370
you even calibrate, you have to verify. Verify,

00:08:35.370 --> 00:08:37.830
OK. Verification is the process of confirming

00:08:37.830 --> 00:08:40.789
that the computational model accurately represents

00:08:40.789 --> 00:08:43.710
the underlying mathematical model and its solution.

00:08:44.389 --> 00:08:46.370
Wait, let me make sure I'm following. So verification

00:08:46.370 --> 00:08:49.230
is essentially like a massive debugging session.

00:08:49.429 --> 00:08:52.070
In a way, yes. It's asking, did we write the

00:08:52.070 --> 00:08:54.269
code correctly? Did we build the model exactly

00:08:54.269 --> 00:08:57.889
to our own internal blueprint? Yes. But on a

00:08:57.889 --> 00:09:01.570
vastly complex scale, verification asks, are

00:09:01.570 --> 00:09:04.009
the equations being solved correctly. Can you

00:09:04.009 --> 00:09:06.309
give an example? Sure. Let's say your digital

00:09:06.309 --> 00:09:09.090
twin involves fluid dynamics, like simulating

00:09:09.090 --> 00:09:11.549
how a liquid moves through a pipe. Verification

00:09:11.549 --> 00:09:14.350
is checking the mesh resolution. It's ensuring

00:09:14.350 --> 00:09:16.509
there are no floating point rounding errors in

00:09:16.509 --> 00:09:19.110
the physics engine. It's purely a mathematical

00:09:19.110 --> 00:09:21.190
and software engineering check. So it doesn't

00:09:21.190 --> 00:09:23.450
even look at the real world yet. Exactly. It

00:09:23.450 --> 00:09:25.470
does not ask if the model represents reality.

00:09:26.029 --> 00:09:28.730
It only asks if the model represents the design.

00:09:28.840 --> 00:09:31.580
Okay, I think I see where you're going. So if

00:09:31.580 --> 00:09:34.759
verification asks, did we build the model right?

00:09:35.639 --> 00:09:38.179
Then validation must be asking, did we build

00:09:38.179 --> 00:09:40.419
the right model? That is the exact distinction.

00:09:40.519 --> 00:09:42.460
That's a great way to put it. So validation is

00:09:42.460 --> 00:09:44.659
bringing the real world back into it. Right.

00:09:44.840 --> 00:09:47.240
Validation is the process of determining the

00:09:47.240 --> 00:09:50.820
degree to which a model is an accurate representation

00:09:50.820 --> 00:09:54.429
of the real world. specifically from the perspective

00:09:54.429 --> 00:09:57.070
of the intended uses of the model. Let me push

00:09:57.070 --> 00:09:59.409
back on this for a second because the line still

00:09:59.409 --> 00:10:01.590
feels a little blurry to me. Go ahead. If I've

00:10:01.590 --> 00:10:03.970
perfectly calibrated my model using real -world

00:10:03.970 --> 00:10:06.730
data, and I've verified that my code has absolutely

00:10:06.730 --> 00:10:09.870
zero bugs, haven't I already validated it? That's

00:10:09.870 --> 00:10:11.950
a very common assumption. Right. Doesn't the

00:10:11.950 --> 00:10:13.809
fact that it perfectly matches the real -world

00:10:13.809 --> 00:10:16.450
data prove that it works? Not at all. And that

00:10:16.450 --> 00:10:18.750
assumption has caused disastrous failures in

00:10:18.750 --> 00:10:21.480
the past. Really? Why? What's fascinating here

00:10:21.480 --> 00:10:24.179
is the requirement for independent datasets.

00:10:25.360 --> 00:10:28.600
To validate a model, you must test its predictive

00:10:28.600 --> 00:10:31.559
capabilities against real -world data that was

00:10:31.559 --> 00:10:35.029
never used during the calibration phase. Because

00:10:35.029 --> 00:10:37.909
if I use the same data to calibrate the model

00:10:37.909 --> 00:10:40.509
and then to test it, it's like giving a student

00:10:40.509 --> 00:10:43.590
the answer key to memorize before the final exam.

00:10:43.830 --> 00:10:46.269
Exactly. This is a massive issue in machine learning

00:10:46.269 --> 00:10:48.269
called overfitting. Overfitting, right. I've

00:10:48.269 --> 00:10:50.330
heard of that. Yeah. So if you calibrate your

00:10:50.330 --> 00:10:53.090
digital twin using the pressure readings from

00:10:53.090 --> 00:10:56.210
a specific pump running at 50 % capacity on a

00:10:56.210 --> 00:10:59.470
Tuesday bear. Okay. And then you try to validate

00:10:59.470 --> 00:11:02.620
the model. Using that exact same Tuesday data,

00:11:03.120 --> 00:11:05.179
the model will look brilliant. It will spit out

00:11:05.179 --> 00:11:07.759
a perfect prediction. But it's just cheating.

00:11:08.100 --> 00:11:10.259
Right. It hasn't actually learned the physical

00:11:10.259 --> 00:11:12.820
laws governing the pump. It just memorized that

00:11:12.820 --> 00:11:15.139
one specific state. So if things change even

00:11:15.139 --> 00:11:17.980
a little bit, it breaks. The moment you run that

00:11:17.980 --> 00:11:20.840
pump at 90 % capacity on a Friday, the model

00:11:20.840 --> 00:11:23.600
completely breaks down. So validation requires

00:11:23.600 --> 00:11:26.620
you to throw a completely new, unforeseen scenario

00:11:26.620 --> 00:11:29.120
at the digital twin and see if it can swim. Right.

00:11:29.450 --> 00:11:31.649
Let's look at a concrete example using one of

00:11:31.649 --> 00:11:33.789
the most stringent frameworks out there, which

00:11:33.789 --> 00:11:38.409
is the ASME -VNV40 standard. ASME -VNV40, what's

00:11:38.409 --> 00:11:40.950
that? This is the American Society of Mechanical

00:11:40.950 --> 00:11:43.629
Engineers standard for assessing the credibility

00:11:43.629 --> 00:11:46.350
of computational modeling in medical devices.

00:11:46.629 --> 00:11:49.759
Okay, so high stakes. Very high stakes. Let's

00:11:49.759 --> 00:11:52.600
say you are building a digital twin of a human

00:11:52.600 --> 00:11:56.059
heart to test a new pacemaker design. Oh, wow.

00:11:56.379 --> 00:11:59.100
It's an incredibly complex electromechanical

00:11:59.100 --> 00:12:02.399
model dealing with tissue resistance, fluid dynamics

00:12:02.399 --> 00:12:05.200
of blood, and electrical impulses. That sounds

00:12:05.200 --> 00:12:07.600
like an absolute nightmare to try and simulate

00:12:07.600 --> 00:12:11.899
perfectly. It is, which is why the ASME VNV40

00:12:11.899 --> 00:12:15.240
framework forces engineers to abandon the arrogant

00:12:15.240 --> 00:12:18.330
claim of having a perfect digital twin. Because

00:12:18.330 --> 00:12:21.029
perfection is impossible there. Right. Instead,

00:12:21.210 --> 00:12:23.350
it demands a transparent audit of limitations.

00:12:23.549 --> 00:12:25.710
What does that look like? Well, during verification,

00:12:26.029 --> 00:12:28.190
you might check that the code simulating electrical

00:12:28.190 --> 00:12:30.289
conductivity doesn't crash when it hits a certain

00:12:30.289 --> 00:12:32.690
threshold. Just checking the math. Right. Then

00:12:32.690 --> 00:12:34.750
during calibration, you might feed the model

00:12:34.750 --> 00:12:38.269
data from a specific patient's MRI to shape the

00:12:38.269 --> 00:12:40.850
digital ventricles. Tailoring the suit. Exactly.

00:12:41.210 --> 00:12:43.970
But for validation. You have to take this digital

00:12:43.970 --> 00:12:46.490
heart, simulate the introduction of the pacemaker,

00:12:46.990 --> 00:12:49.750
predict the exact change in blood flow, and then

00:12:49.750 --> 00:12:52.429
compare that prediction against an independent

00:12:52.429 --> 00:12:55.350
physical bench test. Or like animal data, right?

00:12:55.409 --> 00:12:58.490
Yes, in vivo animal data that the algorithm has

00:12:58.490 --> 00:13:00.950
absolutely never seen before. And if the simulation

00:13:00.950 --> 00:13:04.129
says the blood flow increases by 10%, but the

00:13:04.129 --> 00:13:06.470
physical bench test shows it only increased by

00:13:06.470 --> 00:13:10.110
2 %? Your model fails validation. Full stop.

00:13:10.250 --> 00:13:12.899
Wow. It doesn't matter how beautiful the code

00:13:12.899 --> 00:13:16.480
is. Exactly. It fails validation for that specific

00:13:16.480 --> 00:13:19.399
context of use. Context of use, meaning it might

00:13:19.399 --> 00:13:21.720
be good for one thing, but not another. Yes.

00:13:22.220 --> 00:13:24.419
The model might be perfectly valid for predicting

00:13:24.419 --> 00:13:27.159
heart rate, but entirely invalid for predicting

00:13:27.159 --> 00:13:29.600
blood volume. And auditors care about that distinction.

00:13:29.940 --> 00:13:32.980
Oh, it is vital for quality system audits. Auditors

00:13:32.980 --> 00:13:34.879
don't want to hear that your digital twin is

00:13:34.879 --> 00:13:37.019
a magic oracle that knows everything. They want

00:13:37.019 --> 00:13:39.179
the boundaries. They want to see your validation

00:13:39.179 --> 00:13:42.059
reports detailing exactly where the twin is reliable

00:13:42.059 --> 00:13:44.500
and precisely where the predictive power drops

00:13:44.500 --> 00:13:47.110
off. OK, this raises an incredibly important

00:13:47.110 --> 00:13:49.529
question for me. What's that? I get that we have

00:13:49.529 --> 00:13:52.269
to validate the digital twins predictions against

00:13:52.269 --> 00:13:54.929
independent physical data, like your bench test.

00:13:55.409 --> 00:13:59.309
But if I'm the auditor for the FDA or an aerospace

00:13:59.309 --> 00:14:01.970
client reviewing a manufacturer's claims, I'm

00:14:01.970 --> 00:14:03.529
not just going to trust the bench test either.

00:14:04.029 --> 00:14:06.600
Ah. You're catching on. Right. If you tell me

00:14:06.600 --> 00:14:09.480
the physical bench test measured exactly 2 %

00:14:09.480 --> 00:14:12.039
blood flow, how do I know the flow meter you

00:14:12.039 --> 00:14:14.200
used on the bench test wasn't broken? Bingo.

00:14:14.460 --> 00:14:16.679
Who was checking the tools that we used to check

00:14:16.679 --> 00:14:19.700
the digital twin? And there it is. You have just

00:14:19.700 --> 00:14:22.899
identified the foundational bedrock of all digital

00:14:22.899 --> 00:14:25.379
metrology. Which is? Metrological traceability.

00:14:25.960 --> 00:14:28.200
This is the golden thread that tethers a digital

00:14:28.200 --> 00:14:31.399
twin floating in the cloud to the actual physical

00:14:31.399 --> 00:14:33.820
laws of the universe. The golden thread. I love

00:14:33.820 --> 00:14:35.580
that. When we talk about certification bodies

00:14:35.580 --> 00:14:38.240
and quality system audits, we are primarily talking

00:14:38.240 --> 00:14:42.220
about adherence to standards like ISO IECL -1T025

00:14:42.220 --> 00:14:45.399
and ISO 9001. Oh, man. Anyone listening who works

00:14:45.399 --> 00:14:47.620
in manufacturing, lab testing, or supply chain

00:14:47.620 --> 00:14:49.980
logistics just groaned out loud at the mention

00:14:49.980 --> 00:14:53.179
of ISO 1T025. I know. I know. It's notorious.

00:14:53.460 --> 00:14:56.440
It is the absolute gold standard and a massive,

00:14:56.600 --> 00:14:59.279
massive bureaucratic undertaking for laboratory

00:14:59.279 --> 00:15:01.789
competence. It is stringent by design, though.

00:15:02.029 --> 00:15:06.169
Under ISO 17805, you cannot simply buy a sensor

00:15:06.169 --> 00:15:08.370
off the internet, plug it into your factory,

00:15:08.970 --> 00:15:11.590
and feed its data into a digital twin. You can't

00:15:11.590 --> 00:15:14.490
just trust the Amazon reviews. Definitely not.

00:15:14.690 --> 00:15:17.049
You have to establish metrological traceability.

00:15:17.279 --> 00:15:19.399
What does that mean, practically? This means

00:15:19.399 --> 00:15:22.519
there must be an unbroken, documented chain of

00:15:22.519 --> 00:15:25.840
calibrations linking your specific sensor's measurements

00:15:25.840 --> 00:15:29.039
all the way back to specified reference standards.

00:15:29.360 --> 00:15:31.379
Reference standards, like what? Which are usually

00:15:31.379 --> 00:15:33.700
primary SI units, the International System of

00:15:33.700 --> 00:15:36.700
Units. Unbroken chain. Okay, let's trace that

00:15:36.700 --> 00:15:39.700
chain. Say I have a digital twin of an aircraft

00:15:39.700 --> 00:15:42.620
engine and it's ingesting data from a physical

00:15:42.620 --> 00:15:45.200
temperature sensor on the turbine. Okay, a high

00:15:45.200 --> 00:15:47.539
heat environment. Right. And it says it is exactly

00:15:47.539 --> 00:15:50.159
1 ,000 degrees Kelvin. Trace that back for me.

00:15:50.179 --> 00:15:53.139
How do I prove that? To satisfy an auditor, you

00:15:53.139 --> 00:15:55.940
have to produce documentation showing that the

00:15:55.940 --> 00:15:58.820
specific turbine sensor was calibrated in the

00:15:58.820 --> 00:16:01.580
factory against a specialized reference thermometer.

00:16:01.679 --> 00:16:04.879
Okay, step one. But it doesn't stop there. You

00:16:04.879 --> 00:16:07.059
must prove that the factory's reference thermometer

00:16:07.059 --> 00:16:10.139
was calibrated by an accredited secondary laboratory

00:16:10.139 --> 00:16:12.940
using an even more precise thermal standard.

00:16:13.419 --> 00:16:16.419
Oh, wow. It keeps going. Then you must prove

00:16:16.419 --> 00:16:18.440
that the secondary laboratory's equipment was

00:16:18.440 --> 00:16:20.620
calibrated by a national metrology institute.

00:16:20.740 --> 00:16:24.059
Like who? Like NIST in the United States or NPL

00:16:24.059 --> 00:16:26.570
in the UK. And where do the National Metrology

00:16:26.570 --> 00:16:29.049
Institutes get their truth? They can't just calibrate

00:16:29.049 --> 00:16:31.090
against another bigger thermometer forever. It

00:16:31.090 --> 00:16:33.289
has to stop somewhere, right? This is where we

00:16:33.289 --> 00:16:36.009
touch the absolute fabric of reality. Seriously?

00:16:36.269 --> 00:16:39.009
Seriously. Historically, traceability ended at

00:16:39.009 --> 00:16:41.789
physical artifacts. If you wanted to trace a

00:16:41.789 --> 00:16:44.409
kilogram, your chain eventually led back to a

00:16:44.409 --> 00:16:46.450
vault at the International Bureau of Weights

00:16:46.450 --> 00:16:49.889
and Measures, the BIPM, outside of Paris. Oh,

00:16:49.889 --> 00:16:52.610
I've heard of this. In that vault sat Le Grand

00:16:52.610 --> 00:16:55.389
Quai. a physical cylinder of platinum iridium

00:16:55.389 --> 00:16:58.789
forged in 1889. The grand care, the big care.

00:16:59.190 --> 00:17:02.970
The entire world's concept of mass was digitally

00:17:02.970 --> 00:17:06.950
and physically traced back to that one specific

00:17:06.950 --> 00:17:09.849
lump of metal. reading about this, the problem

00:17:09.849 --> 00:17:12.490
was that Lagrange was actually losing mass over

00:17:12.490 --> 00:17:14.730
the decades, right? Yes, exactly. Like, microscopic

00:17:14.730 --> 00:17:16.789
amounts of metal were wearing off when they cleaned

00:17:16.789 --> 00:17:18.650
it, which meant the definition of a kilogram

00:17:18.650 --> 00:17:21.190
was literally changing. Right. And a changing

00:17:21.190 --> 00:17:24.309
baseline is a disaster for high -precision digital

00:17:24.309 --> 00:17:27.109
twins. Because all the math falls apart. If the

00:17:27.109 --> 00:17:30.789
primary standard drifts, the entire global chain

00:17:30.789 --> 00:17:34.089
of calibration shifts with it. That is why, in

00:17:34.089 --> 00:17:37.930
2019, the scientific community radically redefined

00:17:37.930 --> 00:17:40.049
the SI units. Oh, they got rid of the metal cylinder.

00:17:40.509 --> 00:17:42.769
We stopped tracing back to physical objects in

00:17:42.769 --> 00:17:45.930
vaults entirely. Today, the National Metrology

00:17:45.930 --> 00:17:48.029
Institutes calibrate their primary standards

00:17:48.029 --> 00:17:50.369
against the fundamental constants of the universe.

00:17:50.569 --> 00:17:52.349
Wait, really? How does that even work in practice?

00:17:52.589 --> 00:17:54.819
Take the kilogram. Instead of tracing back to

00:17:54.819 --> 00:17:57.480
Lagrange, laboratories now use something called

00:17:57.480 --> 00:17:59.759
a kibble balance. That kibble balance. It's an

00:17:59.759 --> 00:18:02.559
incredibly complex instrument that measures mass

00:18:02.559 --> 00:18:06.200
by balancing the gravitational force on an object

00:18:06.200 --> 00:18:09.640
against an electromagnetic force. Okay. This

00:18:09.640 --> 00:18:12.559
allows us to define the kilogram using Planck's

00:18:12.559 --> 00:18:14.880
constant, which is a fundamental principle of

00:18:14.880 --> 00:18:17.359
quantum mechanics. That is absolutely staggering.

00:18:17.500 --> 00:18:20.059
So when your digital twin records the mass of

00:18:20.059 --> 00:18:22.700
a chemical payload, the metrological traceability

00:18:22.700 --> 00:18:25.180
chain stretches from your factory floor through

00:18:25.180 --> 00:18:28.319
the accredited labs and ultimately anchors itself

00:18:28.319 --> 00:18:30.940
to an immutable constant of quantum physics.

00:18:31.279 --> 00:18:34.000
The digital twin on my laptop is anchored to

00:18:34.000 --> 00:18:36.680
quantum mechanics by a continuous chain of paperwork.

00:18:37.200 --> 00:18:39.680
That is wild. It's pretty amazing when you step

00:18:39.680 --> 00:18:41.619
back and look at it. But wait, what happens if

00:18:41.619 --> 00:18:43.940
this aircraft engine is built using parts from

00:18:43.940 --> 00:18:46.259
all over the world? Which they usually are. Right.

00:18:46.279 --> 00:18:48.500
So if the temperature sensor was calibrated in

00:18:48.500 --> 00:18:51.119
Germany, the pressure sensor in Japan, and the

00:18:51.119 --> 00:18:53.740
mass flow meter in the United States, how does

00:18:53.740 --> 00:18:56.720
the digital twin know these countries are all

00:18:56.720 --> 00:18:58.859
interpreting these quantum constants the exact

00:18:58.859 --> 00:19:01.960
same way? That is managed through the CIPM mutual

00:19:01.960 --> 00:19:05.359
recognition arrangement or the CIPM MRA. CIPM

00:19:05.359 --> 00:19:08.019
MRA. Got it. Along with frameworks from Isla

00:19:08.019 --> 00:19:10.980
Hess, the International Laboratory Accreditation

00:19:10.980 --> 00:19:14.400
Cooperation. So global treaties, basically. Exactly.

00:19:14.619 --> 00:19:17.720
These are massive global geopolitical agreements

00:19:17.720 --> 00:19:20.819
where national metrology institutes essentially

00:19:20.819 --> 00:19:23.339
peer review and audit each other constantly.

00:19:23.440 --> 00:19:26.220
Keeping everyone honest. Right. They run international

00:19:26.220 --> 00:19:29.119
comparison tests to ensure that a Kelvin measured

00:19:29.119 --> 00:19:32.299
in Tokyo is mathematically identical to a Kelvin

00:19:32.299 --> 00:19:34.559
measured in Berlin. That's incredible. Because

00:19:34.559 --> 00:19:37.319
of these agreements, a digital twin can ingest

00:19:37.319 --> 00:19:40.319
a vast array of global data and actually trust

00:19:40.319 --> 00:19:43.059
its cohesion. So all of this traceability, these

00:19:43.059 --> 00:19:44.980
international agreements, the quantum physics,

00:19:45.660 --> 00:19:48.440
this is what makes up the evidence packages that

00:19:48.440 --> 00:19:50.579
clients and certification bodies demand during

00:19:50.579 --> 00:19:53.400
an audit. Yes. And it is critical not just for

00:19:53.400 --> 00:19:55.920
your own hardware, but for third -party AI modules.

00:19:56.079 --> 00:19:59.559
Oh, like off -the -shelf software. Exactly. Say

00:19:59.559 --> 00:20:02.160
you buy a predictive maintenance AI off the shelf

00:20:02.160 --> 00:20:04.960
to plug into your digital twin. That AI was trained

00:20:04.960 --> 00:20:07.220
on massive data sets provided by the vendor.

00:20:07.529 --> 00:20:10.230
But if that vendor cannot provide the traceability

00:20:10.230 --> 00:20:12.329
chain for the sensors used to gather their training

00:20:12.329 --> 00:20:16.529
data, an ISO auditor will force you to discard

00:20:16.529 --> 00:20:20.170
the AI. Wow, really? Even if it works well? Yes.

00:20:20.410 --> 00:20:23.170
Opaque training data introduces a black box of

00:20:23.170 --> 00:20:25.630
risk. If you don't know the calibration lineage

00:20:25.630 --> 00:20:27.750
of the data that trained your algorithm, you

00:20:27.750 --> 00:20:30.230
cannot trust the algorithm's decisions. Okay,

00:20:30.250 --> 00:20:32.869
so we've established this magnificent unbroken

00:20:32.869 --> 00:20:35.809
chain of trust, linking our digital twin to the

00:20:35.809 --> 00:20:38.029
constants of the universe. It's a beautiful theory.

00:20:38.230 --> 00:20:40.269
It is, but let's bring it back down to Earth.

00:20:40.440 --> 00:20:42.619
Because as much as we want to anchor things to

00:20:42.619 --> 00:20:46.680
quantum perfection, a vibrating 200 -degree manufacturing

00:20:46.680 --> 00:20:49.440
floor is inherently chaotic. Very messy. The

00:20:49.440 --> 00:20:52.339
sensor on that turbine isn't perfect. The wiring

00:20:52.339 --> 00:20:55.220
has resistance. The environment introduces noise.

00:20:55.579 --> 00:20:57.720
If I'm an auditor, how do I deal with the fact

00:20:57.720 --> 00:20:59.819
that perfection is physically impossible? You

00:20:59.819 --> 00:21:02.019
deal with it by embracing the mathematics of

00:21:02.019 --> 00:21:04.240
doubt. The mathematics of doubt? I like the sound

00:21:04.240 --> 00:21:07.299
of that. In metrology, we don't pretend perfection

00:21:07.299 --> 00:21:10.630
exists. we quantify the imperfection. This brings

00:21:10.630 --> 00:21:13.769
us to a foundational text in the field, the guide

00:21:13.769 --> 00:21:15.950
to the expression of uncertainty and measurement,

00:21:16.529 --> 00:21:19.430
universally known as the G -U -M. The D -U -M.

00:21:19.480 --> 00:21:24.339
along with policies like ILSI P14, the GUM provides

00:21:24.339 --> 00:21:26.640
the mathematical framework for dealing with doubt.

00:21:26.940 --> 00:21:30.240
So if my digital twins dashboard says a pipe

00:21:30.240 --> 00:21:34.259
is at exactly 100 PSI, the GUM is the standard

00:21:34.259 --> 00:21:36.839
that tells me how much of a lie that number is.

00:21:37.180 --> 00:21:38.799
Well, let's correct the terminology slightly

00:21:38.799 --> 00:21:40.940
because this is a vital distinction in quality

00:21:40.940 --> 00:21:43.440
audits. You have to draw a hard line between

00:21:43.440 --> 00:21:46.690
error. uncertainty oh they aren't the same thing

00:21:46.690 --> 00:21:48.829
no they're entirely different concepts error

00:21:48.829 --> 00:21:51.029
is the difference between a measured value and

00:21:51.029 --> 00:21:54.609
the true value it is a specific knowable mistake

00:21:54.609 --> 00:21:56.950
knowable mistake can you give an example for

00:21:56.950 --> 00:21:59.630
example if you know your pressure sensor consistently

00:21:59.630 --> 00:22:02.130
reads 2 psi higher than reality because of a

00:22:02.130 --> 00:22:04.930
slight manufacturing defect that is a known error

00:22:05.069 --> 00:22:08.769
you can and must mathematically correct for known

00:22:08.769 --> 00:22:11.569
errors during calibration. Okay, so error is

00:22:11.569 --> 00:22:14.009
like knowing my bathroom scale always adds exactly

00:22:14.009 --> 00:22:15.670
three pounds. I just subtract three pounds in

00:22:15.670 --> 00:22:18.369
my head and I've corrected the error. Exactly.

00:22:18.990 --> 00:22:21.829
But uncertainty is entirely different. Uncertainty

00:22:21.829 --> 00:22:25.170
is a parameter that characterizes the dispersion

00:22:25.170 --> 00:22:28.069
of the values that could reasonably be attributed

00:22:28.069 --> 00:22:30.609
to the measurement. That's a very technical way

00:22:30.609 --> 00:22:33.450
to say it. To simplify, it's a quantification.

00:22:33.660 --> 00:22:36.400
of how much a reported value might differ from

00:22:36.400 --> 00:22:39.140
the true value, stated at a specific confidence

00:22:39.140 --> 00:22:42.619
level, usually 95%. So sticking with the bathroom

00:22:42.619 --> 00:22:45.839
scale, uncertainty is knowing that even after

00:22:45.839 --> 00:22:48.039
I subtract the three pounds of air, depending

00:22:48.039 --> 00:22:49.980
on where I stand on the glass or the temperature

00:22:49.980 --> 00:22:52.640
of the room, or how the digital springs are feeling

00:22:52.640 --> 00:22:55.539
that day, my wheat could still fluctuate by plus

00:22:55.539 --> 00:22:58.519
or minus one pound. Precisely. You can never

00:22:58.519 --> 00:23:00.460
eliminate uncertainty. You can only calculate

00:23:00.460 --> 00:23:02.880
its boundaries. Right. And to pass an audit,

00:23:03.039 --> 00:23:05.880
For a critical digital twin, you must produce

00:23:05.880 --> 00:23:08.759
what is called an uncertainty budget for every

00:23:08.759 --> 00:23:11.660
single data stream entering the system. An uncertainty

00:23:11.660 --> 00:23:13.799
of budget. Here's where it gets really interesting.

00:23:13.859 --> 00:23:16.759
It's crucial for the audit process. We are literally

00:23:16.759 --> 00:23:19.720
creating an itemized spreadsheet, but instead

00:23:19.720 --> 00:23:22.140
of budgeting dollars and cents, we are budgeting

00:23:22.140 --> 00:23:25.400
our own ignorance. We are tracking every single

00:23:25.400 --> 00:23:28.119
place where doubt creeps into the system. That's

00:23:28.119 --> 00:23:30.200
exactly what it is. Let's build one. Let's go

00:23:30.200 --> 00:23:32.440
back to that temperature sensor on the jet engine

00:23:32.440 --> 00:23:35.000
turbine. What goes into an uncertainty budget

00:23:35.000 --> 00:23:37.819
for that one data point? We start with the obvious,

00:23:38.680 --> 00:23:40.960
the calibration uncertainty of the sensor itself,

00:23:41.019 --> 00:23:42.720
which comes from its calibration certificate.

00:23:42.779 --> 00:23:44.980
Okay, that's line item number one. But that's

00:23:44.980 --> 00:23:47.740
just the start. Okay. What about the analog to

00:23:47.740 --> 00:23:50.299
digital converter, the ADC? What does that do?

00:23:50.640 --> 00:23:53.539
The physical sensor reads analog voltage, but

00:23:53.539 --> 00:23:56.640
your digital twin requires digital ones and zeros.

00:23:57.460 --> 00:24:01.160
The ADC has a quantization error. It's a slight

00:24:01.160 --> 00:24:03.539
rounding of the physical reality to fit into

00:24:03.539 --> 00:24:05.599
digital memory. Oh, wow. So that rounding goes

00:24:05.599 --> 00:24:07.839
in the budget. Yes, that goes in the budget.

00:24:08.019 --> 00:24:09.640
What about the environment? If the turbine is

00:24:09.640 --> 00:24:12.480
at 30 ,000 feet, the ambient temperature and

00:24:12.480 --> 00:24:14.759
the vibration must affect the wiring, right?

00:24:15.140 --> 00:24:17.900
Absolutely. Environmental drift is a major line

00:24:17.900 --> 00:24:20.619
item. The resistance of the copper wire connecting

00:24:20.619 --> 00:24:23.799
the sensor to the processor changes with ambient

00:24:23.799 --> 00:24:26.579
temperature. which slightly alters the voltage

00:24:26.579 --> 00:24:29.039
reading. So every little detail matters. Then

00:24:29.039 --> 00:24:31.359
you have method variability. Is there a slight

00:24:31.359 --> 00:24:34.000
delay in how fast the sensor responds to a rapid

00:24:34.000 --> 00:24:35.740
temperature spike? I wouldn't have even thought

00:24:35.740 --> 00:24:38.279
of that. All these individual sources of doubt

00:24:38.279 --> 00:24:41.920
are quantified as standard uncertainties. Then,

00:24:42.000 --> 00:24:44.720
using the GUM methodology, you mathematically

00:24:44.720 --> 00:24:47.900
combine all these distinct doubts, often using

00:24:47.900 --> 00:24:50.420
the root sum of the squares method to calculate

00:24:50.420 --> 00:24:53.900
a final expanded uncertainty. So, after doing

00:24:53.900 --> 00:24:56.019
all this math, I might conclude that my digital

00:24:56.019 --> 00:24:59.299
twin's reading of 1000 degrees Kelvin has an

00:24:59.299 --> 00:25:02.240
expanded uncertainty of, say, plus or minus 4

00:25:02.240 --> 00:25:05.079
degrees at a 95 % confidence level. Exactly.

00:25:05.200 --> 00:25:07.099
But why does the auditor care so deeply about

00:25:07.099 --> 00:25:09.619
this? If the engine melts at 1500 degrees, who

00:25:09.619 --> 00:25:11.859
cares about a 4 degree margin of doubt? Because

00:25:11.859 --> 00:25:14.400
in complex systems, margins of doubt compound

00:25:14.400 --> 00:25:16.140
and they define your safety boundaries. Safety

00:25:16.140 --> 00:25:18.299
boundaries. Let's look at predictive maintenance.

00:25:18.730 --> 00:25:21.690
Your digital twin is simulating the wear on a

00:25:21.690 --> 00:25:24.490
titanium turbine blade based on vibration and

00:25:24.490 --> 00:25:27.130
thermal data. The simulation predicts the blade

00:25:27.130 --> 00:25:29.849
will suffer a catastrophic fracture at exactly

00:25:29.849 --> 00:25:32.789
10 ,000 flight hours. So I tell the airline to

00:25:32.789 --> 00:25:36.210
replace the blade at 9 ,900 flight hours to be

00:25:36.210 --> 00:25:39.009
safe. But what if you never calculated your uncertainty

00:25:39.009 --> 00:25:42.240
budget? What if the combined doubt from the sensors,

00:25:42.799 --> 00:25:45.680
the ADC quantization, and the environmental drift

00:25:45.680 --> 00:25:48.420
means your digital twins prediction actually

00:25:48.420 --> 00:25:51.539
has a massive expanded uncertainty of plus or

00:25:51.539 --> 00:25:54.240
minus 500 hours? Oh, I see. That blade could

00:25:54.240 --> 00:25:56.960
snap at 9 ,500 hours, taking down the aircraft,

00:25:57.359 --> 00:25:58.740
while your maintenance crew thought they had

00:25:58.740 --> 00:26:01.789
hundreds of hours left. That's terrifying. Calculating

00:26:01.789 --> 00:26:04.230
uncertainty is how you establish the safe operating

00:26:04.230 --> 00:26:06.309
envelope of a digital twin. Because you have

00:26:06.309 --> 00:26:08.390
to assume the worst -case scenario within that

00:26:08.390 --> 00:26:10.829
doubt. Exactly. If your uncertainty is plus or

00:26:10.829 --> 00:26:13.730
minus 500 hours, you mandate maintenance at 9

00:26:13.730 --> 00:26:16.269
,000 hours. You budget your doubt to guarantee

00:26:16.269 --> 00:26:18.730
human safety. And I imagine certification bodies

00:26:18.730 --> 00:26:21.250
are pretty strict about this. They enforce strict

00:26:21.250 --> 00:26:23.390
rules on how this is communicated, particularly

00:26:23.390 --> 00:26:26.329
the Uricum rules on rounding. Uricum? What's

00:26:26.329 --> 00:26:30.049
that? Uricum is a network of organizations in

00:26:30.049 --> 00:26:33.109
Europe focusing on analytical measurement quality.

00:26:34.049 --> 00:26:36.170
But their guidelines on expressing uncertainty

00:26:36.170 --> 00:26:38.549
are widely adopted globally. And what do they

00:26:38.549 --> 00:26:40.990
say about rounding? One of their core tenets

00:26:40.990 --> 00:26:43.849
is that you cannot report results with fake precision

00:26:43.849 --> 00:26:47.480
just to impress a client. Fake precision. Yeah.

00:26:47.700 --> 00:26:49.579
If your uncertainty budget proves your margin

00:26:49.579 --> 00:26:52.900
of doubt is plus or minus 1 .5 degrees, you are

00:26:52.900 --> 00:26:55.180
strictly prohibited from programming your digital

00:26:55.180 --> 00:26:58.079
twins dashboard to display the temperature to

00:26:58.079 --> 00:27:01.700
three decimal places, like 100 .125 degrees.

00:27:01.799 --> 00:27:04.579
Ah, because adding those decimal places implies

00:27:04.579 --> 00:27:06.500
a level of certainty that you don't mathematically

00:27:06.500 --> 00:27:09.119
possess. Right. It's lying with UI design. It's

00:27:09.119 --> 00:27:12.180
like a false promise of accuracy. Exactly. The

00:27:12.180 --> 00:27:14.200
rounding rules mandate that the resolution of

00:27:14.200 --> 00:27:17.079
your reported data must match the reality of

00:27:17.079 --> 00:27:19.539
your expanded uncertainty. If the digital twin

00:27:19.539 --> 00:27:21.839
claims a precision it hasn't earned, it will

00:27:21.839 --> 00:27:25.359
fail the audit. OK, so in a perfect world, we

00:27:25.359 --> 00:27:28.500
have our verified code, our independent validation

00:27:28.500 --> 00:27:31.500
data, our unbroken chain of metrological traceability

00:27:31.500 --> 00:27:35.079
back to the kibble balance in Paris, and a meticulously

00:27:35.079 --> 00:27:38.140
calculated uncertainty budget preventing us from

00:27:38.140 --> 00:27:40.240
making dangerous assumptions. That's the dream.

00:27:40.599 --> 00:27:43.559
It sounds like an airtight, perfect mathematical

00:27:43.559 --> 00:27:46.900
utopia. But what happens when we take this pristine

00:27:46.900 --> 00:27:50.240
theory and drop it onto a greasy, chaotic factory

00:27:50.240 --> 00:27:52.799
floor? Real -world friction happens. massive

00:27:52.799 --> 00:27:54.920
amounts of it. I knew there was a catch. The

00:27:54.920 --> 00:27:57.319
gap between the theory of metrology and the actual

00:27:57.319 --> 00:28:00.119
deployment of industrial digital twins is currently

00:28:00.119 --> 00:28:02.259
filled with brutal bottlenecks. Let's dig into

00:28:02.259 --> 00:28:04.319
these bottlenecks because this is where the grand

00:28:04.319 --> 00:28:06.940
promises of digital twins really start to face

00:28:06.940 --> 00:28:08.970
reality. Yeah, it's not all smooth sailing. The

00:28:08.970 --> 00:28:11.430
first major obstacle highlighted by engineers

00:28:11.430 --> 00:28:13.789
in the field is what they call the paper problem.

00:28:14.170 --> 00:28:15.650
And frankly, when I read this, it sounded like

00:28:15.650 --> 00:28:17.670
a joke. I wish it were a joke. We are talking

00:28:17.670 --> 00:28:21.269
about cloud -based AI, quantum physics, and massive

00:28:21.269 --> 00:28:24.589
sensor networks. What does paper have to do with

00:28:24.589 --> 00:28:27.230
anything? It is the great, embarrassing irony

00:28:27.230 --> 00:28:29.630
of the industrial Internet of Things, the IIOT.

00:28:29.789 --> 00:28:32.609
How so? We have these astonishing digital twins

00:28:32.609 --> 00:28:35.230
capable of ingesting millions of data points

00:28:35.230 --> 00:28:38.009
a second, running complex neural networks. Right,

00:28:38.289 --> 00:28:40.730
incredibly advanced stuff. But the accredited

00:28:40.730 --> 00:28:43.750
calibration certificates, the actual legal documents

00:28:43.750 --> 00:28:46.349
proving the metrological traceability and outlining

00:28:46.349 --> 00:28:48.609
the uncertainty budgets we just spent 20 minutes

00:28:48.609 --> 00:28:52.089
discussing, are almost entirely stuck as physical

00:28:52.089 --> 00:28:55.299
paper documents, or at best, static. PDFs. I'm

00:28:55.299 --> 00:28:57.640
sorry, what? I know. If I have a digital twin

00:28:57.640 --> 00:29:00.559
monitoring a 100 ,000 square foot gigafactory

00:29:00.559 --> 00:29:02.920
and a technician installs a new pressure sensor,

00:29:03.380 --> 00:29:06.019
how does the digital twin know the sensor's uncertainty

00:29:06.019 --> 00:29:08.700
budget? Very often a human being has to physically

00:29:08.700 --> 00:29:11.200
open the PDF certificate provided by the calibration

00:29:11.200 --> 00:29:14.559
lab, read the parameters, and manually type the

00:29:14.559 --> 00:29:17.039
uncertainty values and calibration curves into

00:29:17.039 --> 00:29:20.029
the digital twins database. That is insane. If

00:29:20.029 --> 00:29:22.150
we connect this to the bigger picture of AI,

00:29:22.250 --> 00:29:24.970
this seems completely absurd. We have large language

00:29:24.970 --> 00:29:27.450
models that can write software code. Why can't

00:29:27.450 --> 00:29:30.069
we just digitize a calibration certificate? Why

00:29:30.069 --> 00:29:32.230
hasn't the industry solved this with a simple

00:29:32.230 --> 00:29:35.089
machine -readable file? It's not a technology

00:29:35.089 --> 00:29:38.349
problem. It's a legal and proprietary formatting

00:29:38.349 --> 00:29:41.049
nightmare. Okay, explain that. Historically,

00:29:41.240 --> 00:29:45.299
Every major calibration lab, every sensor manufacturer,

00:29:45.480 --> 00:29:48.799
and every national metrology institute use their

00:29:48.799 --> 00:29:51.500
own proprietary software and reporting formats.

00:29:51.799 --> 00:29:54.839
There was no universal ontology for metrology.

00:29:55.099 --> 00:29:57.119
So everyone speaks a different language. Exactly.

00:29:57.259 --> 00:29:59.940
If you tried to automate it, an OCR scanner might

00:29:59.940 --> 00:30:02.619
misread a comma as a decimal point on a PDF from

00:30:02.619 --> 00:30:05.039
a German lab, corrupting the uncertainty budget

00:30:05.039 --> 00:30:07.480
by an order of magnitude. Which changes the safety

00:30:07.480 --> 00:30:09.579
boundaries entirely. Right. And when lives are

00:30:09.579 --> 00:30:11.880
on the line, liability is a massive barrier.

00:30:12.220 --> 00:30:14.920
Who is legally responsible if a script misreads

00:30:14.920 --> 00:30:17.319
a PDF and a jet engine explodes? Nobody wants

00:30:17.319 --> 00:30:19.720
that lawsuit. So industry's stuck with the safety

00:30:19.720 --> 00:30:22.809
of human -verified paper. but this manual entry

00:30:22.809 --> 00:30:25.109
completely kills scalability. I can imagine.

00:30:27.480 --> 00:30:30.759
across a network of 10 ,000 smart sensors, if

00:30:30.759 --> 00:30:34.180
you need a team of interns manually typing in

00:30:34.180 --> 00:30:37.160
10 ,000 PDFs every time the annual calibration

00:30:37.160 --> 00:30:39.619
interval rolls around. So the paper problem is

00:30:39.619 --> 00:30:42.460
a major bureaucratic and physical bottleneck.

00:30:42.720 --> 00:30:45.339
Huge bottleneck. But there is also a conceptual

00:30:45.339 --> 00:30:47.559
bottleneck that engineers fall into, which is

00:30:47.559 --> 00:30:50.900
called the misapplication. There is a very stern

00:30:50.900 --> 00:30:53.700
warning in control systems engineering that a

00:30:53.700 --> 00:30:56.500
digital twin is not a controller. This is a critical

00:30:56.500 --> 00:30:58.680
distinction. that often gets ignored by over

00:30:58.680 --> 00:31:01.240
-enthusiastic executives who don't understand

00:31:01.240 --> 00:31:03.319
the underlying systems. The executives just see

00:31:03.319 --> 00:31:05.839
the shiny dashboard. Right. A CEO will look at

00:31:05.839 --> 00:31:08.680
the digital twin dashboard and think, well, the

00:31:08.680 --> 00:31:10.779
twin has all this real -time data, and it is

00:31:10.779 --> 00:31:12.819
predicting exactly what the machine will do next,

00:31:12.859 --> 00:31:14.759
so let's cut out the middleman. Let's just hook

00:31:14.759 --> 00:31:17.119
it up. Exactly. Let's just hook the digital twin

00:31:17.119 --> 00:31:19.339
directly into the automated actuation systems

00:31:19.339 --> 00:31:22.200
and let the AI run the factory. They try to use

00:31:22.200 --> 00:31:24.359
the complex digital twin as a substitute for

00:31:24.359 --> 00:31:27.500
traditional localized hardware like a PLC or

00:31:27.500 --> 00:31:29.559
a Smith predictor. I'm going to push back here

00:31:29.559 --> 00:31:32.779
just like that CEO would. Go for it. Intuitively,

00:31:32.940 --> 00:31:36.000
why wouldn't we do that? If I have a brilliantly

00:31:36.000 --> 00:31:39.180
validated digital twin that knows the exact uncertainty

00:31:39.180 --> 00:31:42.079
budget of my chemical reactor, why shouldn't

00:31:42.079 --> 00:31:44.160
I let it automatically shut the pressure valve

00:31:44.160 --> 00:31:47.140
when things get dangerous? Why add another layer

00:31:47.140 --> 00:31:50.480
of hardware? because of determinism versus probabilism

00:31:50.480 --> 00:31:53.220
and the harsh reality of time delays. OK, unpack

00:31:53.220 --> 00:31:55.319
those terms for me. A traditional controller,

00:31:55.839 --> 00:31:59.460
like a PLC, a programmable logic controller running

00:31:59.460 --> 00:32:03.500
a simple PID loop, is deterministic. It is a

00:32:03.500 --> 00:32:05.779
localized piece of hardware bolted to the machine.

00:32:06.039 --> 00:32:08.539
So it's right there on the floor? Yes. If a pressure

00:32:08.539 --> 00:32:12.380
switch hits 150 PSI, the PLC is hardwired to

00:32:12.380 --> 00:32:14.509
shut the valve. It does this in milliseconds

00:32:14.509 --> 00:32:17.710
with absolute guaranteed certainty. OK. And a

00:32:17.710 --> 00:32:19.910
digital twin. A digital twin, on the other hand,

00:32:20.309 --> 00:32:22.809
is a highly complex probabilistic simulation.

00:32:22.990 --> 00:32:25.799
It doesn't just read pressure. It analyzes history,

00:32:26.180 --> 00:32:28.339
cross -references thermal dynamics, and simulates

00:32:28.339 --> 00:32:30.039
five different future scenarios in the cloud.

00:32:30.220 --> 00:32:32.559
Which takes computational time. Exactly. It takes

00:32:32.559 --> 00:32:34.640
time to process and it relies on network latency.

00:32:34.779 --> 00:32:37.579
Ah, the internet connection. Right. If that chemical

00:32:37.579 --> 00:32:40.740
reactor spikes and your simple PLC handles it,

00:32:40.980 --> 00:32:44.299
the valve shuts in 10 milliseconds. If you rely

00:32:44.299 --> 00:32:46.839
on a cloud -based digital twin to run that valve,

00:32:47.400 --> 00:32:50.930
a sudden spike in network traffic momentary compute

00:32:50.930 --> 00:32:53.670
bottleneck on the server or even a brief Wi -Fi

00:32:53.670 --> 00:32:56.109
drop on the factory floor could cause a two -second

00:32:56.109 --> 00:32:58.549
delay. And two seconds is a lifetime in that

00:32:58.549 --> 00:33:00.910
scenario. In chemical processing, a two -second

00:33:00.910 --> 00:33:04.309
delay means a catastrophic explosion. Wow. And

00:33:04.309 --> 00:33:06.130
you mentioned a Smith predictor earlier that's

00:33:06.130 --> 00:33:08.250
a specific type of predictive controller used

00:33:08.250 --> 00:33:11.089
to compensate for known dead time delays in a

00:33:11.089 --> 00:33:13.900
physical system. A digital twin can inform the

00:33:13.900 --> 00:33:15.740
tuning of a Smith predictor by modeling those

00:33:15.740 --> 00:33:18.319
delays, but it should never replace the localized

00:33:18.319 --> 00:33:20.980
hardware executing the command. So to use an

00:33:20.980 --> 00:33:23.619
analogy, the digital twin is the brilliant general

00:33:23.619 --> 00:33:25.759
sitting in the war room, looking at the satellite

00:33:25.759 --> 00:33:28.039
maps, analyzing troop movements and coming up

00:33:28.039 --> 00:33:30.079
with the master strategy. I like that. But the

00:33:30.079 --> 00:33:32.200
PLC is the soldier in the trench actually pulling

00:33:32.200 --> 00:33:34.920
the trigger. You absolutely do not want the general

00:33:34.920 --> 00:33:37.019
trying to remotely pull the trigger from 5 ,000

00:33:37.019 --> 00:33:39.799
miles away. That is a perfect way to conceptualize

00:33:39.799 --> 00:33:42.720
it. The mandate here is fit for purpose. Fit

00:33:42.720 --> 00:33:45.240
for purpose. A digital twin is intended for high

00:33:45.240 --> 00:33:47.519
-level planning, for validating new processes

00:33:47.519 --> 00:33:50.500
virtually, for long -term optimization, and for

00:33:50.500 --> 00:33:53.200
scheduling predictive maintenance. It is an analytical

00:33:53.200 --> 00:33:56.099
tool, not a real -time actuator. So using it

00:33:56.099 --> 00:33:59.339
as a switch is just overkill. Deploying a massive

00:33:59.339 --> 00:34:02.420
computationally heavy digital twin to do a job

00:34:02.420 --> 00:34:05.160
that a $500 hardwired switch can do instantly

00:34:05.160 --> 00:34:08.119
is a dangerous misapplication of resources. Okay.

00:34:08.230 --> 00:34:10.349
That makes total sense. So we keep the twin as

00:34:10.349 --> 00:34:14.210
an advisor, not a trigger puller. But even in

00:34:14.210 --> 00:34:17.070
its advisory role, the twin still relies on all

00:34:17.070 --> 00:34:19.329
those physical sensors scattered around the factory.

00:34:19.949 --> 00:34:22.090
And you mentioned environmental drift earlier.

00:34:22.289 --> 00:34:24.710
If I have a gigafactory that is 100 ,000 square

00:34:24.710 --> 00:34:27.070
feet, the physical environment is not uniform.

00:34:27.269 --> 00:34:29.800
Far from it. In a massive facility, you deal

00:34:29.800 --> 00:34:32.300
with intense spatial drift. Spatial drift? What

00:34:32.300 --> 00:34:34.860
does that look like? The temperature near the

00:34:34.860 --> 00:34:37.340
40 -foot ceiling is vastly different from the

00:34:37.340 --> 00:34:40.119
temperature near the loading docks. The vibrations

00:34:40.119 --> 00:34:42.780
near the heavy stamping presses are fundamentally

00:34:42.780 --> 00:34:44.820
different from the vibrations near the assembly

00:34:44.820 --> 00:34:47.000
line. So different zones have different stress

00:34:47.000 --> 00:34:50.139
levels. Exactly. The electromagnetic interference

00:34:50.139 --> 00:34:53.639
from welding arms wreaks havoc on unshielded

00:34:53.639 --> 00:34:57.230
wiring. Over time, the sensors in these different

00:34:57.230 --> 00:35:00.570
zones will drift physically due to localized

00:35:00.570 --> 00:35:02.869
environmental stress. Their calibration degrades.

00:35:03.030 --> 00:35:05.829
Yes. Their accuracy degrades, but they degrade

00:35:05.829 --> 00:35:07.909
at completely different rates depending on exactly

00:35:07.909 --> 00:35:09.849
where they are in the building. So a sensor by

00:35:09.849 --> 00:35:12.449
the furnace loses calibration way faster than

00:35:12.449 --> 00:35:14.329
a sensor in the air -conditioned packaging wing.

00:35:15.070 --> 00:35:17.030
Precisely. But wait, haven't we solved this with

00:35:17.030 --> 00:35:19.849
software? I've read a lot about self -calibrating

00:35:19.849 --> 00:35:22.429
algorithms. Things like Kalman filtering, where

00:35:22.429 --> 00:35:25.210
the sensor network uses math and historical data

00:35:25.210 --> 00:35:27.789
to constantly correct its own drift in real time.

00:35:28.329 --> 00:35:30.010
Doesn't that solve the problem without needing

00:35:30.010 --> 00:35:32.869
a technician to go out with a wrench? Computational

00:35:32.869 --> 00:35:35.510
data processing like Kalman filtering is incredibly

00:35:35.510 --> 00:35:38.110
elegant. So it works. Well, for those unfamiliar,

00:35:38.349 --> 00:35:41.829
a Kalman filter is an algorithm that uses a series

00:35:41.829 --> 00:35:44.510
of measurements observed over time. containing

00:35:44.510 --> 00:35:47.750
statistical noise and produces estimates of unknown

00:35:47.750 --> 00:35:50.070
variables that tend to be more accurate than

00:35:50.070 --> 00:35:52.730
those based on a single measurement alone. Like

00:35:52.730 --> 00:35:55.070
it averages out the noise. It works on a predict

00:35:55.070 --> 00:35:58.170
-update loop. It is fantastic for smoothing out

00:35:58.170 --> 00:36:00.849
noisy data and extending the usable lifespan

00:36:00.849 --> 00:36:03.769
of a sensor network. But I hear a butt coming.

00:36:04.130 --> 00:36:07.489
A massive butt when it comes to quality audits.

00:36:07.710 --> 00:36:10.849
Self -calibration through an algorithm does not

00:36:10.849 --> 00:36:13.389
provide legal metrological traceability. Because

00:36:13.389 --> 00:36:15.869
there's no physical anchor. You cannot trace

00:36:15.869 --> 00:36:18.650
a Kalman filter's mathematical guess back to

00:36:18.650 --> 00:36:21.389
the kibble balance in Paris. Right. Because it's

00:36:21.389 --> 00:36:23.769
essentially just a sophisticated math trick trying

00:36:23.769 --> 00:36:26.389
to guess the drift based on historical patterns

00:36:26.389 --> 00:36:29.230
rather than a physical comparison against a known

00:36:29.230 --> 00:36:32.030
immutable standard. Yes. And in complex environments

00:36:32.030 --> 00:36:34.559
with spatial drift, self -calibrating algorithms

00:36:34.559 --> 00:36:37.840
can actually be dangerous. If a sensor slowly

00:36:37.840 --> 00:36:40.619
drifts, because the furnace nearby is radiating

00:36:40.619 --> 00:36:43.079
more heat than it used to, the Kalman filter

00:36:43.079 --> 00:36:46.199
might slowly adjust to this drift, assuming the

00:36:46.199 --> 00:36:48.639
rising temperature is the new normal for the

00:36:48.639 --> 00:36:52.219
process. Oh, it just accepts the error. It will

00:36:52.219 --> 00:36:55.460
smooth the data out, masking the drift. You end

00:36:55.460 --> 00:36:57.780
up with systematic errors that the algorithm

00:36:57.780 --> 00:37:00.429
itself is fundamentally blind to. because it

00:37:00.429 --> 00:37:03.329
has lost touch with objective reality. It's just

00:37:03.329 --> 00:37:05.789
hallucinating that everything is fine. Exactly.

00:37:06.650 --> 00:37:10.170
This is why auditors will outright reject a digital

00:37:10.170 --> 00:37:12.949
twin that relies solely on self -calibration

00:37:12.949 --> 00:37:15.170
for critical safety metrics. You have to physically

00:37:15.170 --> 00:37:17.329
check the sensors against a traceable standard

00:37:17.329 --> 00:37:20.190
to keep the math honest. So to summarize this

00:37:20.190 --> 00:37:22.690
absolute gauntlet of real -world friction...

00:37:22.699 --> 00:37:25.460
Our biggest hurdles are the paper problem, trapping

00:37:25.460 --> 00:37:28.480
calibration records and PDFs, the danger of using

00:37:28.480 --> 00:37:30.820
twins as real -time controllers, and the physical

00:37:30.820 --> 00:37:33.440
impossibility of maintaining traceable calibration

00:37:33.440 --> 00:37:35.699
across tens of thousands of drifting sensors

00:37:35.699 --> 00:37:38.139
without spending an absolute fortune on manual

00:37:38.139 --> 00:37:40.559
labor. That sums it up pretty well. How on earth

00:37:40.559 --> 00:37:43.380
is the industry planning to fix this? How do

00:37:43.380 --> 00:37:45.780
we scale this technology if we are bogged down

00:37:45.780 --> 00:37:48.480
by paper and manual calibration? This is where

00:37:48.480 --> 00:37:50.840
we look to the horizon. to the emerging field

00:37:50.840 --> 00:37:53.500
of digital metrology. The industry is building

00:37:53.500 --> 00:37:55.980
a highly automated machine -readable infrastructure

00:37:55.980 --> 00:37:59.199
to solve these exact bottlenecks. OK, good. So

00:37:59.199 --> 00:38:01.800
there is hope. There is. The first step is completely

00:38:01.800 --> 00:38:04.630
eradicating the paper problem. We're moving towards

00:38:04.630 --> 00:38:07.570
universal implementation of digital calibration

00:38:07.570 --> 00:38:11.050
certificates or DCCs and something called Transducer

00:38:11.050 --> 00:38:13.829
Electronic Data Sheets, known as TEDs. TEDs.

00:38:13.929 --> 00:38:16.389
I like that. How does TEDs kill the PDF? TEDs

00:38:16.389 --> 00:38:20.230
is defined under a specific standard, IEE 1451.

00:38:21.010 --> 00:38:23.289
Essentially, it allows a smart sensor to store

00:38:23.289 --> 00:38:26.090
its own metrological identity directly on a microchip

00:38:26.090 --> 00:38:27.929
inside the sensor itself. Inside the hardware?

00:38:28.230 --> 00:38:30.969
Yes. The moment a technician plugs that sensor

00:38:30.969 --> 00:38:32.869
into the network, the digital twin pings it.

00:38:32.940 --> 00:38:35.400
The sensor transmits its TEDs data, it tells

00:38:35.400 --> 00:38:37.300
the twin exactly what kind of sensor it is, its

00:38:37.300 --> 00:38:39.659
serial number, its entire calibration history,

00:38:39.719 --> 00:38:41.900
and its mathematically precise uncertainty budget.

00:38:42.280 --> 00:38:44.519
So the SIRSR basically hands the digital twin

00:38:44.519 --> 00:38:47.400
its own digital resume and background check instantly.

00:38:47.699 --> 00:38:51.480
No humans, no PDFs, no typos. Exactly. It creates

00:38:51.480 --> 00:38:54.360
an end -to -end, completely machine -readable

00:38:54.360 --> 00:38:57.460
calibration chain. The digital twin instantly

00:38:57.460 --> 00:39:00.159
ingests the uncertainty budget and updates its

00:39:00.159 --> 00:39:02.639
own overall system margins dynamically. That

00:39:02.639 --> 00:39:05.900
is brilliant and feels long overdue. But what

00:39:05.900 --> 00:39:08.219
about the spatial drift problem? Right, the environment

00:39:08.219 --> 00:39:10.719
still degrades the sensor. Yeah, even with TEDs,

00:39:10.840 --> 00:39:13.260
a sensor's physical calibration will still degrade

00:39:13.260 --> 00:39:16.059
over time in a hot factory. Do we still have

00:39:16.059 --> 00:39:18.860
to pay technicians to manually remove 10 ,000

00:39:18.860 --> 00:39:22.280
TED sensors and send them to an ISO 17025 lab

00:39:22.280 --> 00:39:24.219
every year? Because that sounds economically

00:39:24.219 --> 00:39:26.460
ruinous. That is the multi -million dollar question.

00:39:26.579 --> 00:39:29.039
And the solution is a completely new architectural

00:39:29.039 --> 00:39:31.179
approach to calibration networks. OK, what is

00:39:31.179 --> 00:39:33.630
it? The industry is shifting away from calibrating

00:39:33.630 --> 00:39:36.289
every single sensor individually and moving toward

00:39:36.289 --> 00:39:38.849
what are called directed acyclic graph calibration

00:39:38.849 --> 00:39:42.199
structures. Directed acyclic graph. A DAGS. I

00:39:42.199 --> 00:39:44.500
know they use DAGs in computer science and data

00:39:44.500 --> 00:39:46.400
modeling, but how does that apply to physically

00:39:46.400 --> 00:39:48.659
calibrating hardware on a factory floor? Think

00:39:48.659 --> 00:39:52.039
of a DAG as a physical hierarchy of trust distributed

00:39:52.039 --> 00:39:54.019
throughout the facility. Hierarchy of trust.

00:39:54.280 --> 00:39:56.639
Instead of trying to maintain traceable calibration

00:39:56.639 --> 00:40:00.519
on 10 ,000 cheap sensors, you strategically install

00:40:00.519 --> 00:40:03.519
a small number of ground truth nodes. Ground

00:40:03.519 --> 00:40:05.719
truth nodes, okay. These are highly precise,

00:40:06.059 --> 00:40:08.860
robust, and expensive reference sensors placed

00:40:08.860 --> 00:40:11.929
strategically maybe one securely mounted in the

00:40:11.929 --> 00:40:14.349
ceiling, one near the floor, one near the furnace.

00:40:14.570 --> 00:40:18.309
Okay, so you create localized pockets of absolute

00:40:18.309 --> 00:40:21.409
truth. Right. These ground truth nodes are the

00:40:21.409 --> 00:40:23.949
only instruments in the factory that require

00:40:23.949 --> 00:40:27.989
rigorous manual physical calibration with high

00:40:27.989 --> 00:40:29.789
-end metrology equipment. Because there's only

00:40:29.789 --> 00:40:32.570
a few of them. Exactly. They're explicitly traced

00:40:32.570 --> 00:40:35.750
back to the national standards. Then, the digital

00:40:35.750 --> 00:40:38.289
twin uses the continuous data from those trusted

00:40:38.289 --> 00:40:41.570
ground truth nodes to constantly verify and virtually

00:40:41.570 --> 00:40:44.289
calibrate the thousands of cheaper, lower precision

00:40:44.289 --> 00:40:46.989
sensors around them in real time. Let me visualize

00:40:46.989 --> 00:40:49.570
this. It's like a castating waterfall of trust.

00:40:49.710 --> 00:40:51.809
That's a great image. The National Metrology

00:40:51.809 --> 00:40:54.630
Institute calibrates the lab. The lab calibrates

00:40:54.630 --> 00:40:56.849
the ground truth node in the factory ceiling,

00:40:57.590 --> 00:41:00.170
and the ground truth node calibrates the 50 cheap

00:41:00.170 --> 00:41:03.510
temperature sensors operating below it. The calibration

00:41:03.510 --> 00:41:06.489
flows in a directed graph, always moving downward,

00:41:06.710 --> 00:41:09.150
never looping back on itself. That is exactly

00:41:09.150 --> 00:41:12.469
why it is a directed acyclic graph. It balances

00:41:12.469 --> 00:41:15.309
the high cost of manual traceable calibration

00:41:15.309 --> 00:41:18.289
with the massive spatial coverage needed for

00:41:18.289 --> 00:41:20.500
a digital twin. Which solves the money problem.

00:41:20.679 --> 00:41:22.900
It allows you to maintain metrological integrity

00:41:22.900 --> 00:41:24.980
without bankrupting the maintenance department.

00:41:25.320 --> 00:41:28.260
That is incredibly elegant, but I have one final,

00:41:28.619 --> 00:41:30.679
somewhat cynical question. I expect nothing less.

00:41:31.000 --> 00:41:33.679
If all of this data is flowing digitally, the

00:41:33.679 --> 00:41:36.519
TED's data updating automatically, the ground

00:41:36.519 --> 00:41:38.920
truth nodes calibrating the DAG network remotely,

00:41:39.840 --> 00:41:41.900
How does an auditor know someone didn't just

00:41:41.900 --> 00:41:45.300
hack the system? Ah, security? Right. If a factory

00:41:45.300 --> 00:41:47.739
manager wants to pass a safety inspection, couldn't

00:41:47.739 --> 00:41:50.199
they just go into the server and alter the digital

00:41:50.199 --> 00:41:52.360
calibration certificates to make it look like

00:41:52.360 --> 00:41:54.659
the uncertainty budget is tighter than it is?

00:41:55.039 --> 00:41:57.159
Which brings us to the final piece of the digital

00:41:57.159 --> 00:42:01.119
metrology puzzle. Immutable security. You cannot

00:42:01.119 --> 00:42:04.260
have machine -to -machine trust. without cryptographic

00:42:04.260 --> 00:42:07.039
proof. Cryptography, like blockchain. Yes. This

00:42:07.039 --> 00:42:09.000
is why the cutting edge of digital twin deployment

00:42:09.000 --> 00:42:11.800
is integrating distributed ledger technology,

00:42:12.260 --> 00:42:14.639
specifically blockchain, at the edge of the network.

00:42:14.900 --> 00:42:17.380
OK. Whenever I hear blockchain in an industrial

00:42:17.380 --> 00:42:20.489
context. I immediately think of the crypto hype

00:42:20.489 --> 00:42:23.929
cycle. Of course, it gets a bad rap. Isn't running

00:42:23.929 --> 00:42:26.710
a blockchain incredibly energy intensive and

00:42:26.710 --> 00:42:30.590
slow? How do you run that on low power IoT sensors

00:42:30.590 --> 00:42:33.610
on a factory floor? That is a very valid skepticism,

00:42:33.610 --> 00:42:35.849
but you have to separate public cryptocurrency

00:42:35.849 --> 00:42:39.030
blockchains, which use energy intensive proof

00:42:39.030 --> 00:42:41.889
of work mechanisms from private permissioned

00:42:41.889 --> 00:42:44.170
industrial ledgers. OK, so they operate differently.

00:42:44.489 --> 00:42:47.059
Very differently. In digital metrology, we aren't

00:42:47.059 --> 00:42:49.719
mining coins. We're using lightweight cryptographic

00:42:49.719 --> 00:42:52.360
hashing. When a ground truth node calibrates

00:42:52.360 --> 00:42:54.960
a lower tier sensor, that specific calibration

00:42:54.960 --> 00:42:57.500
event, along with the TED's data and the calculated

00:42:57.500 --> 00:43:00.119
uncertainty, is hashed and recorded on a private

00:43:00.119 --> 00:43:02.300
decentralized ledger. Where does the ledger live?

00:43:02.510 --> 00:43:04.809
It's distributed across the factory's local servers

00:43:04.809 --> 00:43:07.989
and secure cloud environments. So if someone

00:43:07.989 --> 00:43:11.130
tries to go back and secretly edit a calibration

00:43:11.130 --> 00:43:13.929
certificate from three months ago to hide a failing

00:43:13.929 --> 00:43:16.909
sensor, the cryptographic hash breaks and the

00:43:16.909 --> 00:43:19.889
system immediately flags the tampering. Oh, exactly.

00:43:20.199 --> 00:43:22.760
To ensure the digital twin of the factory is

00:43:22.760 --> 00:43:25.119
honest, we must build the digital twin of the

00:43:25.119 --> 00:43:28.300
calibration process itself and lock it in an

00:43:28.300 --> 00:43:30.760
immutable vault. That's brilliant. And if we

00:43:30.760 --> 00:43:33.500
connect this to the bigger picture, this infrastructure

00:43:33.500 --> 00:43:36.730
isn't just a luxury. It will soon be mandatory

00:43:36.730 --> 00:43:39.710
for survival in complex industries. Why mandatory?

00:43:39.889 --> 00:43:42.510
Because the nature of auditing is changing. We

00:43:42.510 --> 00:43:45.050
are entering an era of AI assisted conformity

00:43:45.050 --> 00:43:48.769
assessments. AI auditors? Yes. Regulatory bodies

00:43:48.769 --> 00:43:50.869
won't be sending humans with clipboards to spy

00:43:50.869 --> 00:43:53.510
check your PDFs anymore. They will deploy software

00:43:53.510 --> 00:43:56.349
bots, AI auditors, that plug into your digital

00:43:56.349 --> 00:43:58.570
twins API. Oh man, they can audit everything

00:43:58.570 --> 00:44:01.820
instantly. These bots will execute smart contracts

00:44:01.820 --> 00:44:04.920
that instantly trace the DAG network, query the

00:44:04.920 --> 00:44:07.280
blockchain for the provenance of every TED certificate,

00:44:07.960 --> 00:44:10.579
and recalculate your entire factory's uncertainty

00:44:10.579 --> 00:44:13.659
budget in milliseconds. So there's no hiding

00:44:13.659 --> 00:44:16.980
anything? None. If your system relies on manual

00:44:16.980 --> 00:44:19.980
entry and paper trails, the AI auditor won't

00:44:19.980 --> 00:44:22.360
even be able to read your compliance. you will

00:44:22.360 --> 00:44:24.980
fail by default. You just get a failing grade

00:44:24.980 --> 00:44:27.579
instantly. The digital metrology infrastructure

00:44:27.579 --> 00:44:30.619
is the only way organizations can maintain the

00:44:30.619 --> 00:44:33.380
speed of modern innovation while satisfying the

00:44:33.380 --> 00:44:36.239
unrelenting demands of quality systems and certification

00:44:36.239 --> 00:44:39.000
bodies. What an absolute journey. We have covered

00:44:39.000 --> 00:44:41.659
a monumental amount of ground today. It is a

00:44:41.659 --> 00:44:43.820
lot to take in. To synthesize this for everyone

00:44:43.820 --> 00:44:46.710
listening, We started with the seductive illusion

00:44:46.710 --> 00:44:49.530
of the digital twin and uncovered the fact that

00:44:49.530 --> 00:44:52.090
its real power lies in the strict mathematical

00:44:52.090 --> 00:44:54.769
discipline of calibration, verification, and

00:44:54.769 --> 00:44:57.510
validation. Holy Trinity. We traced the golden

00:44:57.510 --> 00:45:00.269
thread of metrological traceability all the way

00:45:00.269 --> 00:45:03.070
from a factory floor to the quantum constants

00:45:03.070 --> 00:45:05.889
of the universe. We learned how to budget our

00:45:05.889 --> 00:45:08.829
doubt to guarantee safety, navigated the frustrating

00:45:08.829 --> 00:45:11.429
realities of paper certificates and spatial drift,

00:45:11.570 --> 00:45:13.809
the real -world friction, and finally looked

00:45:13.809 --> 00:45:16.150
at the DAG networks and cryptographic ledgers

00:45:16.150 --> 00:45:18.849
that will build the automated, machine -readable

00:45:18.849 --> 00:45:22.820
future. It is a vast, invisible infrastructure

00:45:22.820 --> 00:45:25.159
of trust that keeps the modern world running.

00:45:25.380 --> 00:45:29.159
It truly is. And as we close, there is one final,

00:45:29.179 --> 00:45:31.099
almost philosophical concept I want to leave

00:45:31.099 --> 00:45:33.920
you with, which emerges when you look at the

00:45:33.920 --> 00:45:36.239
trajectory of this technology. Ooh, I love a

00:45:36.239 --> 00:45:38.480
good philosophical concept. We discussed how

00:45:38.480 --> 00:45:40.980
machine learning allows a digital twin to constantly

00:45:40.980 --> 00:45:44.030
update itself. With a perfect DAG network and

00:45:44.030 --> 00:45:46.590
continuous data, the digital twin is constantly

00:45:46.590 --> 00:45:49.309
refining its understanding of physics, theoretically

00:45:49.309 --> 00:45:52.010
becoming more perfect over time. Right. It's

00:45:52.010 --> 00:45:53.949
always learning. It gets better and better. But

00:45:53.949 --> 00:45:56.090
simultaneously, the physical machine on the factory

00:45:56.090 --> 00:45:58.690
floor is doing the exact opposite. It is aging.

00:45:58.909 --> 00:46:01.789
The physical world degrades. The metal is fatiguing.

00:46:01.989 --> 00:46:04.190
The bearings are wearing down. Entropy is taking

00:46:04.190 --> 00:46:07.670
hold. We are rapidly approaching a paradigm in

00:46:07.670 --> 00:46:10.769
advanced manufacturing where the digital representation

00:46:10.769 --> 00:46:13.909
is mathematically and structurally more perfect

00:46:13.909 --> 00:46:16.230
than the physical reality it was originally built

00:46:16.230 --> 00:46:18.170
to mirror. That is a profound thought. Which

00:46:18.170 --> 00:46:20.070
brings up the ultimate question for you, the

00:46:20.070 --> 00:46:22.730
listener, to mull over as you go about your week.

00:46:23.010 --> 00:46:25.940
If you are sitting in a control room, and your

00:46:25.940 --> 00:46:27.840
digital twin and your physical machines start

00:46:27.840 --> 00:46:30.579
to disagree, and both of them have perfectly

00:46:30.579 --> 00:46:33.719
traceable, blockchain -verified, unbroken chains

00:46:33.719 --> 00:46:37.320
of calibration. Which one is the real truth and

00:46:37.320 --> 00:46:40.000
which one is just the reflection? A very dangerous

00:46:40.000 --> 00:46:42.840
question for an engineer to face. Indeed. Thank

00:46:42.840 --> 00:46:45.079
you all for joining us on this incredibly deep

00:46:45.079 --> 00:46:47.579
dive into the hidden, rigorous world of digital

00:46:47.579 --> 00:46:50.860
twins, standards, and metrology. Remember to

00:46:50.860 --> 00:46:53.179
always question the data around you. Look for

00:46:53.179 --> 00:46:55.539
the traceability. and we will catch you on the

00:46:55.539 --> 00:46:56.400
next deep dive.