Latest / Mechanical Engineering Made Simple / (#121) Designing thermal fluid systems for power

Transcript

(#121) Designing thermal fluid systems for power · Mechanical Engineering Made Simple

0:00
Welcome back design junkies to Mechanical Engineering Made
0:02
Simple. If the podcast brings you value
0:05
help, support the show with a like a follow or leave a comment
0:07
below. Glad to be here for this one.
0:09
It's it's going to be a heavy one.
0:11
Oh, absolutely. You know, usually when we talk
0:14
about a medical diagnosis, there's this expectation of
0:17
precision, right? You break your arm, the X-ray
0:19
shows that jagged white line and a doctor just points and says
0:23
there it is. Yeah, it's binary.
0:25
Like it's either broken or it's not.
0:26
Exactly. You look at the film and the
0:29
problem is just perfectly visible.
0:31
We inherently crave that kind of categorization, I think, just
0:35
because it's comforting. Right.
0:37
But then you step into the world of actual physical thermal fluid
0:40
system design and suddenly that X-ray machine is just, well,
0:43
it's completely useless. We are looking at an engineering
0:46
landscape that is honestly, it's murky.
0:49
It is incredibly murky. Because on paper everything
0:52
looks pristine. Your spreadsheet gives you a
0:54
precise efficiency number down to like the 4th decimal place,
0:58
But in reality, fluid flow and heat transfer are chaotic,
1:03
they're highly sensitive to boundary conditions, and they
1:05
almost never behave the way a simplified 1D textbook equation
1:09
suggests. Yeah, it's the absolute
1:11
definition of diagnostic Muddy Waters.
1:14
Yeah, the gap between theoretical thermodynamic
1:16
elegance and actual, you know, operational machinery is where
1:20
most critical engineering failures happen.
1:22
Exactly. And that is the mission of
1:24
today's Deep Dive. We are taking a dense, heavy
1:27
look at the source material spanning thermodynamics, fluid
1:30
mechanics and turbo machinery. And we are extracting the
1:33
practical real world constraints you actually need to know.
1:36
Which is so needed because people get lost in the theory.
1:39
Right, because today is all about mechanical engineering
1:41
made simple. We are breaking down the biggest
1:43
pain points and real world issues engineers keep running
1:46
into with thermal fluid system design.
1:48
We really need to call out the common misconceptions that just
1:51
keep screwing designs up. Oh, big time.
1:54
We're talking about massive of systemic traps, like designing
1:57
heat engines for pure theoretical efficiency instead
2:01
of maximum usable power, or ignoring the incredibly messy 3
2:06
dimensional reality of airflow inside a spinning compressor.
2:11
It's terrifying how often that happens.
2:13
So true. Then we are going to walk
2:15
through a step by step approach so you don't fall into these
2:17
same traps starting from base thermodynamic properties and
2:21
moving all the way up to optimizing complex turbo
2:24
machinery and practice. It's a journey for sure.
2:27
OK, let's unpack this. Have you ever wondered why the
2:30
beautiful, perfectly efficient thermodynamic cycles you learn
2:33
about in school completely failed to produce usable power
2:36
in the real? World man.
2:37
It is the ultimate rude awakening for a young engineer.
2:41
You spend years in school calculating the Carnot cycle,
2:44
right? Which is the theoretical gold
2:46
standard for thermodynamic efficiency.
2:47
The absolute peak. Exactly.
2:49
And the Carnot efficiency is famously simple.
2:51
It is 1 minus the ratio of the cold reservoir temperature to
2:54
the hot reservoir temperature. Mathematically, it is beautiful.
2:57
It really is. It represents a perfectly
3:00
reversible engine where absolutely no energy is wasted
3:03
to friction, to turbulence, or to, you know, uncontrolled heat
3:07
transfer. But, and this is huge,
3:09
reversible is the trap word there.
3:11
Yes, for a process to be truly perfectly reversible in
3:15
thermodynamics, there can't be any driving forces that create
3:18
entropy. Which means to transfer heat
3:22
from a hot reservoir into your working fluid without any
3:25
thermodynamic loss, the temperature difference between
3:27
the reservoir and the fluid must be exactly 0.
3:31
Precisely. And if the temperature
3:32
difference is 0, the heat transfer rate is 0, right?
3:35
That absolutely zero thermal gradient and 0 friction.
3:39
You need infinite time to transfer the heat from the hot
3:43
reservoir to the working fluid, and then, you know, infinite
3:46
time to reject it to the cold reservoir.
3:48
Which means your actual energy flow rate is 0.
3:50
Exactly 0. And since power is just energy
3:53
transferred per unit of time, your actual power output is
3:56
literally 0. Yeah, it's a harsh reality.
3:59
It's like designing a hyper aerodynamic car that gets 10,000
4:02
miles to the gallon, but it only achieves that efficiency if you
4:06
drive it at .001 mph. That's a great analogy.
4:10
Right, Like you might have record-breaking efficiency, but
4:13
if you need to actually deliver a payload across the country,
4:17
your real world productivity is entirely useless.
4:20
That is the exact reality. Think about a steady flow
4:23
engine, something like a massive steam turbine in a power plant,
4:28
or a closed cycle gas turbine. To get actual measurable shaft
4:32
work out of that engine, you must overcome the physical
4:36
thermal resistances between the heat reservoirs and your working
4:40
fluid. Because the metal itself is in
4:41
the way. Right.
4:42
The metal walls of the boiler and the condenser present a
4:45
barrier. Heat only flows across that
4:47
barrier when there is a meaningful temperature
4:49
difference. But the moment you introduce
4:51
that temperature difference to force the heat through, you
4:54
introduce irreversibility. You immediately lose that
4:58
perfect carno efficiency. Wait, so if we actually push for
5:01
maximum efficiency we literally get 0 power out of it?
5:04
Literally 0. So you are forced to make a
5:06
compromise. You deliberately sacrifice some
5:09
of that perfect theoretical efficiency because you have to
5:12
move the heat fast enough to generate actual megawatts of
5:15
power. Yes, and the math provides a
5:17
specific optimal point for that compromise if your goal is to
5:22
get the maximum power output. Not the maximum efficiency, but
5:26
the absolute most work done per second.
5:29
The optimal temperature ratio across the engine shifts
5:32
dramatically. How so?
5:34
It dictates that the optimal temperature ratio happens when
5:37
the internal ratio equals the square root of the cold
5:40
reservoir temperature divided by the hot reservoir temperature.
5:43
Or you know, TC over TH to the power of 1/2.
5:46
OK, let me make sure I'm visualizing the physics of this
5:47
correctly. Instead of just using the raw
5:49
external temperatures of the environment, say the burning
5:52
hole in the cold river water, we have to look at the temperature
5:55
of the working fluid inside the engine's heat exchangers as it
5:57
absorbs and rejects that heat. Correct.
6:00
Let's look at a concrete example from the sources.
6:02
To ground this imagine a heat engine connected to a scorching
6:07
hot reservoir at 1600 Kelvin and rejecting heat to a cold
6:12
reservoir at 400 Kelvin. OK 16104. 100 right?
6:16
If you are chasing maximum carno efficiency you would
6:19
theoretically design the engines internal working fluid to match
6:23
those external temperatures exactly. 1600 on the hot side,
6:26
400 on the cold side. But again, with no temperature
6:29
delta across the heat exchanger walls, no heat actually flows.
6:33
You get a perfect spreadsheet and a dead engine.
6:35
Exactly. So we apply the physical
6:37
optimization for maximum power. We take the ratio of the cold to
6:41
hot reservoirs 400 / 1600 which is 1/4 or 0.25 right sqrt 0.25
6:48
is 0 point five. This geometric mean dictates
6:51
exactly how we set our internal operating temperatures to
6:55
maximize the energy flow rate through the engine.
6:58
The working fluid needs to operate at drastically different
7:01
temperatures than the surrounding reservoirs to like
7:03
physically force that heat transfer through the metal.
7:06
How do engineers balance this in the real world without just
7:09
arbitrarily guessing where to set the operating temperatures?
7:12
You obviously don't want a machine that burns through fuel
7:16
terribly inefficiently, but you also need it to keep the lights
7:20
on in a city. What's fascinating here is that
7:23
internally reversible engines operating at maximum power
7:27
output naturally settle at a practical efficiency that
7:31
perfectly balances the carno ideal with the physical reality
7:35
of thermal resistances. Yeah, when you calculate the
7:39
engines efficiency at that maximum PowerPoint, it is
7:42
significantly lower than the carno ideal, but it represents
7:45
the absolute peak of what you can achieve while actually
7:48
extracting the maximum amount of useful work per second.
7:51
It changes the entire design paradigm.
7:53
Completely. You stop chasing a theoretical
7:55
ghost and start designing for the maximum rate of energy
7:58
transfer. That completely flips the
8:00
script. You aren't just trying to
8:01
aggressively conserve every single Joule of energy, you are
8:05
trying to move those joules as effectively as possible to
8:07
generate horsepower, but to calculate any of these power
8:11
outputs. In the real world, to figure out
8:13
how much work you can actually extract across a turbine, you
8:17
need to know the state of your working fluid.
8:19
You need to know its internal energy, its enthalpy and its
8:22
entropy. And that brings us to one of the
8:24
biggest foundational hurdles in thermal fluid design, the
8:28
measurement problem. Oh, this is a big one.
8:30
It is. You can easily install a
8:32
thermocouple into a pipe to measure temperature.
8:35
You can attach a gauge to measure pressure.
8:37
You can measure the physical volume of a containment vessel,
8:40
pressure specific volume and temperature, PV and T.
8:44
These are tangible macro scale properties.
8:47
Right, you can physically see the dial move.
8:49
Exactly. But you cannot go to an
8:51
industrial supplier and purchase an entropy meter.
8:54
Right, I can't just stick a probe into a high pressure steam
8:56
line and have a digital readout tell me the internal energy or
8:59
the entropy. They are unmeasurable in a
9:01
direct physical sense. So how do engineers bypass this?
9:06
How do we not get completely stalled trying to measure the
9:09
unmeasurable chaos of the fluid? We rely on the mathematics of
9:13
continuous surfaces, specifically the Maxwell
9:15
relations OK. These are fundamental
9:18
mathematical properties that allow engineers to relate
9:21
changes in unmeasurable properties like entropy directly
9:25
to changes in the properties we can easily measure, like
9:27
pressure, volume, and temperature.
9:30
I want to break this down because mathematics of
9:32
continuous surfaces sounds incredibly abstract.
9:35
Let's ground this. Sure, if I have a single
9:38
component system like pure water or pure nitrogen, the
9:41
relationship between all its thermodynamic properties forms a
9:44
three-dimensional surface, right?
9:47
Imagine a 3D graph. The X axis is entropy, the axis
9:50
is specific volume, and the vertical Z axis is the specific
9:53
internal energy. Yes, and because the state of
9:56
that pure fluid is smooth and continuous, assuming we aren't
10:00
crossing a phase boundary yet, that 3D surface has to be
10:03
mathematically continuous, and a continuous mathematical surface
10:06
has a very strict rule. The cross derivatives must be
10:10
equal. Meaning it's like hiking a
10:12
perfectly smooth mountain. It doesn't matter if I walk
10:14
north and east or east and N, the ultimate slope and elevation
10:18
I reach at a specific GPS coordinate on that mountain is
10:21
identical regardless of the path I took to get to that point.
10:24
Precisely. And from that mathematical
10:27
absolute of the mountain slope we derive the Maxwell relations.
10:30
OK, give me an example of 1. For example, one of the
10:33
foundational relations states that the partial derivative of
10:36
entropy with respect to specific volume at constant temperature
10:41
is exactly equal to the partial derivative of pressure with
10:45
respect to temperature at constant volume.
10:47
Let me translate that into physical hardware, because that
10:49
is a massive conceptual leap. The left side of that equation
10:53
is about entropy, the unmeasurable chaos, the
10:56
dispersion of energy as the volume changes I can't measure.
11:00
That right, you can't buy a gauge for it.
11:01
But the right side, the change in pressure when you change the
11:05
temperature inside a rigid constant volume tank, I can
11:09
absolutely measure that. I just need a steel box, a
11:12
pressure gauge, a thermometer, and a Bunsen burner.
11:14
Exactly. The Maxwell relations act as the
11:17
Rosetta Stone of thermodynamics. They allow you to calculate
11:20
precisely how much the unmeasurable entropy changes by
11:24
simply recording how the physical pressure responds to a
11:27
temperature shift on a gauge in a lab.
11:29
That's brilliant. It turns abstract physics into
11:32
measurable, actionable engineering data.
11:35
So we completely bypass the need for an entropy meter.
11:38
We just track the pressure gauge on the outside of the rigid tank
11:41
and the math bridges the gap. That is an incredibly elegant
11:45
workaround. It really is.
11:46
But we also have to deal with specific heat capacities, the
11:50
exact amount of thermal energy it takes to raise the
11:52
temperature of the fluid. And there is a very strict
11:55
unforgiving rule about the specific heat constant pressure,
11:59
which we call CP versus the specific heat at constant volume
12:03
CV. Yes, The math supported by these
12:06
same partial derivatives dictates that the specific heat
12:09
at constant pressure CEP must always be mathematically greater
12:13
than or equal to the specific heat at constant volume CV.
12:17
And physically that makes sense. If I heat a gas inside a rigid
12:20
steel tank, which is constant volume, all the heat goes purely
12:23
into raising the temperature of the gas, correct?
12:26
But if I heat that same gas in a cylinder with a movable piston,
12:29
so constant pressure, the gas expands as it heats up.
12:32
It has to push the piston against the weight of the
12:34
atmosphere. It has to do physical work.
12:36
Exactly. It's pushing back against the
12:37
environment. So I need to add more total heat
12:40
to achieve the same temperature because some of that heat is
12:42
being spent moving the piston, therefore CP is greater.
12:46
That is the macroscopic physical view, and it is entirely
12:49
correct. But the underlying mathematical
12:51
reason goes even deeper, and it involves another partial
12:56
derivative. It involved the rate of change
12:59
of pressure with respect to specific volume at a constant
13:03
temperature. This specific derivative must
13:05
always be negative. Wait, why is that so crucial?
13:08
What happens if that derivative isn't negative?
13:11
Think about what a positive value would mean physically if
13:14
the rate of change of pressure with respect to volume were
13:17
positive. It would mean that as you
13:19
squeeze a substance, as you decrease its physical volume,
13:22
its internal pressure would also decrease.
13:24
That sounds bizarre. Or, conversely, if you pull on
13:27
it and increase its volume, its internal pressure would
13:30
spontaneously increase. That sounds like an absolute
13:33
nightmare. If I squeeze is a balloon, it
13:35
naturally pushes back harder against my hands.
13:37
The pressure goes up as the volume goes down.
13:39
That's a negative relationship. Exactly.
13:42
If it were positive, applying even a tiny amount of pressure
13:45
to the outside of the balloon would cause the pressure inside
13:47
to drop, meaning the atmosphere would crush it even faster,
13:51
lowering the internal pressure further.
13:53
Causing it to shrink more until it completely collapsed out of
13:56
existence in a spontaneous microsecond.
13:59
Yeah, if a substance had a positive derivative there, it
14:02
would be fundamentally structurally unstable.
14:06
The mathematical reality that CP must be greater than or equal to
14:10
CV is tied directly to the fact that matter as we know it
14:14
inherently resists being compressed.
14:16
It literally saves us from the physical universe just
14:18
spontaneously collapsing and on itself.
14:21
It's wild how a simple mathematical derivative dictates
14:24
the structural stability of reality.
14:26
It's the hidden scaffolding of the universe.
14:28
But as we mentioned earlier, these smooth continuous surfaces
14:31
only apply when the fluid is well behaved.
14:33
We don't always deal with nice continuous states.
14:36
Sometimes the fluid boils, sometimes liquid water violently
14:40
flashes into high pressure scheme.
14:42
That is a massive discontinuity. The mountain suddenly has a
14:45
sheer Cliff face. Phase changes are where
14:48
engineering gets incredibly dangerous.
14:50
When water boils into steam, it expands violently.
14:54
If we didn't have a way to mathematically model that exact
14:57
transition, our safety pressure relief valves, our industrial
15:01
boilers, and our nuclear reactor containment vessels would be
15:04
designed based on complete guesswork.
15:06
And a guessing game with high pressure steam usually ends with
15:09
a crater where your facility used to be.
15:12
So how do we handle the mathematics of that sheer Cliff
15:15
face? Through the Clausius cluppy or
15:17
an equation, when a fluid changes phase, say from liquid
15:21
water to vapor, it happens at a constant pressure and a constant
15:24
temperature. We call those the saturation
15:26
values. During that exact phase change,
15:29
the fluid absorbs a massive amount of thermal energy, which
15:32
we call the latent heat of vaporization.
15:34
It absorbs all this energy without changing its temperature
15:37
at all. Right.
15:38
A pot of boiling water stays at exactly 100°C.
15:41
You can crank the burner as high as you want, but the liquid
15:44
won't get hotter, it will just convert into Steam Fest.
15:47
Using the exact same continuous surface math from the Maxwell
15:51
relations, the Claudius Clapeyron equation proves that
15:54
this massive invisible latent heat of vaporization is directly
15:59
tied to the physical slope of the saturation line on a simple
16:03
pressure temperature diagram. We just the slope.
16:05
Just the slope. By taking a fluid in the lab and
16:08
simply measuring how it's boiling point changes as you
16:11
increase the pressure, which is literally just plotting a curve
16:14
on a graph, you can calculate exactly how much energy is
16:17
required to tear the liquid molecules apart into a gas.
16:21
That's crazy. You don't need to measure the
16:23
intermolecular bonds, you just need to know the slope of the
16:25
boiling curve. It's another brilliant shortcut.
16:28
We've mathematically defined the fluids behavior.
16:30
We know how to extract maximum power, we know how to calculate
16:33
its unmeasurable thermodynamic properties using simple lab
16:36
gauges, and we know exactly how much energy it absorbs when it
16:39
boils. Which is a huge toolbox.
16:42
But a fluid just sitting stationary in a tank isn't doing
16:45
much useful work. For us to be a functioning
16:47
machine, we have to move it. We have to force it through
16:50
pipes, drive it down cooling channels, and blast it over
16:53
microchips. And that brings us to a massive
16:57
chronic pain point for thermal fluid engineers.
16:59
Internal flow, the physical reality of forcing A viscous
17:04
fluid through a highly confined space.
17:07
I cannot tell you how many design spreadsheets and
17:10
preliminary CFD models I've seen where the engineer just assumes
17:14
the fluid is perfectly behaved from the very millimeter it
17:17
enters the pipe. Too many.
17:18
Far too many. They assume the flow is
17:20
instantly fully developed right at the inlet, but treating the
17:23
entrance of a pipe like the middle of a pipe is a dangerous
17:26
misconception. The flow entrance region behaves
17:28
completely differently, and ignoring it leads to massive
17:32
catastrophic miscalculations in both pressure drop and heat
17:35
transfer. It is an incredibly trap because
17:38
fully developed flow is mathematically easy to solve.
17:42
But think about the physical reality.
17:44
When fluid first enters a pipe from a large reservoir, the
17:47
velocity profile is basically uniform.
17:51
It's moving at the exact same speed all across the cross
17:53
section. But immediately the very first
17:56
layer of fluid touching the metal wall stops moving entirely
17:59
due to friction. We call this the no slip
18:02
condition because fluids have viscosity internal friction.
18:06
This stationary layer slows down the next layer, which slows down
18:09
the next layer. It cascades.
18:11
Exactly. This creates A boundary layer of
18:14
slow moving fluid that grows thicker and thicker as the flow
18:16
moves further down the pipe. The easiest way to visualize
18:19
this is a traffic jam. If you have a highway where
18:22
everyone is doing 60 mph, but suddenly a car breaks down on
18:26
the right shoulder, the cars in the right lane have to slow
18:28
down. Then the cars in the middle lane
18:30
have to hit their brakes to merge, and slowly that pocket of
18:34
slow moving traffic expands from the wall out into the center
18:37
lengths of the highway. That's a perfect analogy.
18:39
Because the fluid near the walls is drastically slowing down, the
18:43
fluid in the exact center of the pipe actually has to speed up to
18:46
compensate. It has to accelerate to maintain
18:49
the same overall mass flow rate through the pipe.
18:51
That makes sense. This entire region where the
18:54
boundary layer is growing in the core is accelerating, is the
18:57
hydrodynamic entrance region. Only after these boundary layers
19:01
grow so thick that they meet perfectly in the center of the
19:03
pipe does the flow become fully developed, settling into that
19:07
classic, unchanging parabolic velocity profile known as Hagen
19:12
Poiseuille flow. So the critical question is, how
19:14
long does that take? How far down the pipe does this
19:17
traffic jam extend? Because if I'm designing A
19:19
compact heat exchanger with very short pipes, my fluid might
19:23
never actually reach that fully developed parabolic state.
19:27
The scaling for this flow entrance length is well
19:29
documented in the sources. For laminar flow.
19:32
The physical distance it takes to become fully developed is
19:34
roughly 10 to the power of -2 times the Reynolds number times
19:38
the diameter. And since a high Reynolds number
19:41
means the fluid has high inertial forces, it's either
19:44
moving very fast or it has very low viscosity, like a thin gas.
19:49
A highly turbulent, fast moving flow is going to stretch that
19:52
entrance region incredibly far down the pipe.
19:55
Precisely, and in that entrance region, the physical friction
19:58
against the wall is significantly higher than the
20:01
fully developed region. Why is that?
20:03
Because the boundary layer is very thin at the entrance, which
20:06
means the velocity gradient, the difference in speed between the
20:09
stationary wall and the fast moving core, is incredibly
20:12
steep. If you calculate the pressure
20:15
drop required for your pump, assuming a nice easy, fully
20:18
developed flow, but your actual physical system is comprised of
20:22
short Pikes that are entirely in the entrance region, your
20:24
friction calculations will be way off.
20:26
So your pump will be vastly undersized.
20:28
Entirely undersized. You'll build the physical
20:31
prototype, turn the pump on, and a trickle will come out the
20:34
other end because the friction choked it to death.
20:38
But it's not just about the pressure drop, is it?
20:39
It's also about thermal management.
20:41
Just like there is a velocity boundary layer, there is a
20:44
thermal entrance length as well. Yes, just as the friction from
20:48
the wall slowly creeps into the center of the pipe, if the wall
20:51
is at a different temperature than the fluid, a thermal
20:54
boundary layer grows. Heat slowly diffuses from the
20:57
wall into the core of the fluid, and the length of this thermal
21:00
entrance region depends heavily on a specific fluid property,
21:05
the Prancel number. The Prancel number.
21:07
It's the ratio of a molecular momentum diffusivity to thermal
21:10
diffusivity. In simple terms, it describes
21:12
how fast physical momentum spreads through the fluid
21:15
compared to how fast heat spreads through it.
21:17
Exactly. For a fluid like heavy engine
21:20
oil, which has a massive prancel number, momentum spreads much
21:24
faster than heat. The fluid is highly viscous but
21:28
poorly conductive, so the velocity profile might become
21:31
fully developed relatively quickly, but the temperature
21:34
profile takes a huge physical distance to develop.
21:37
The thermal entrance length is incredibly long, and here's the
21:41
trap. The heat transfer coefficient in
21:44
this thermal entrance region is extremely high right at the
21:47
inlet where the boundary layer is paper thin, but it decays
21:51
exponentially as the flow moves downstream and the boundary
21:54
layer thickens. So if I design A cooling channel
21:56
assuming a uniform, fully developed heat transfer
21:59
coefficient across the whole length, my design is going to be
22:01
wildly inaccurate. The front of the channel will
22:03
over cool and the back of the channel will literally melt.
22:06
That is exactly what happens in poorly designed electronics
22:09
cooling. Here's where it gets really
22:11
interesting, because we aren't just talking about long generic
22:13
water pipes anymore. We are talking about micro
22:17
channels cooling the latest, most powerful, incredibly
22:20
credibly dense computer chips in massive hyperscale server farms.
22:25
You don't just arbitrarily guess a channel with anymore.
22:28
We actually have a specific step by step approach to optimize
22:31
these channel sizes for cooling electronics.
22:34
You absolutely cannot afford to guess when you have millions of
22:37
dollars of silicon crammed into a tiny rack space generating the
22:40
heat of a small oven. The engineering goal here is to
22:43
maximize the heat transfer density, figuring out exactly
22:47
how much heat you can remove from a highly constrained
22:49
specific physical volume. And the absolute key to this
22:53
optimization is a dimensionless parameter called the Bijan
22:56
number. Let's define the physics of
22:57
that. The Bijan number mathematically
22:59
captures the specified pressure drop across the channel.
23:02
So delta key times l ^2 divided by the fluids dynamic viscosity
23:07
MU times its thermal diffusivity alpha.
23:10
The true beauty of the Bijan number is that perfectly
23:13
encapsulates the exact physical constraints of a real world
23:17
electronics cooling problem. You have a strictly fixed volume
23:21
for your heat sink, a fixed length of the fins, and a highly
23:25
specific limited pressure drop that your small onboard cooling
23:28
fan can provide. You can't just put a bigger fan
23:31
on it, you are completely constrained.
23:33
So given those rigid constraints, how does this step
23:36
by step optimization work in practice?
23:39
You analyze the two extreme physical asymptotes of the
23:42
design. Let's say you're designing a
23:45
stack of parallel plates, your classic aluminum heat sink fans.
23:48
If you make the spacing between the plates extremely large, you
23:52
can only physically fit a few plates in your given volume.
23:54
Makes sense now. The fan easily pushes the air
23:57
through very low friction, but your total metal surface area
24:00
available for heat transfer is incredibly small.
24:03
The thermal resistance is too high, the air doesn't pick up
24:05
enough heat and the electronic components completely overheat.
24:08
OK, so the large spacing asymptote is a failure due to
24:11
lack of surface area. What if I go the other way?
24:14
What if I pack them in really really tight?
24:16
Thousands of paper thin fins to maximize surface area?
24:20
If the spacing is too small, the internal friction becomes
24:23
immense. The pressure drop required to
24:25
force air through those microscopic gaps is drastically
24:30
higher than what your small cooling fan can actually
24:32
deliver. The fluid essentially chokes.
24:35
It can't flow, it stagnates between the fins.
24:37
Oh wow. And once again the whole volume
24:39
overheats, this time due to lack of airflow.
24:41
So you have a large spacing failure and a small spacing
24:44
failure. The optimal physical design has
24:46
to be the exact intersection of those two extremes.
24:49
That is the exact mathematical approach.
24:51
The optimal spacing for parallel plates is found exactly exactly
24:54
where these two asymptotes cross.
24:56
The math give the exact optimal geometry.
24:59
It states that the ratio of the optimal spacing D opt to the
25:03
length of the channel L is approximately 2.7 times the
25:07
Bijan number to the power of -1 fourth.
25:09
That's incredibly specific. It is specifically as your
25:13
allowable pressure drop increases, meaning you have a
25:16
stronger fan, the optimal spacing gets tighter.
25:19
Let's visualize this inside that massive server farm, rack after
25:23
rack of buzzing circuit boards, the computing power of that room
25:27
is strictly limited by how fast you can remove the heat.
25:30
By using this exact Bijan number methodology, you maximize the
25:34
heat transfer density. You tune the spacing of the fins
25:37
perfectly so that the thermal boundary layers growing off of
25:40
adjacent fins just barely meet exactly at the exit plane of the
25:44
channel. That is the crucial underlying
25:46
physical insight. Think about what happens if you
25:49
get the spacing wrong. If the boundary layers meet too
25:52
early near the inlet, you have fully developed flow for the
25:55
rest of the channel, which means the heat transfer coefficient
25:58
drops and becomes highly inefficient.
26:00
Conversely, if the fins are too far apart and the boundary
26:03
layers never meet, you have a massive channel of wasted space
26:06
in the center between the plates, where cold air is just
26:09
bypassing the hot metal entirely without picking up any heat.
26:14
The absolute mathematically optimal design occurs when the
26:17
thermal boundary layers merge just as the fluid leaves the
26:20
heat sink. That is so deeply satisfying
26:23
from a design perspective. It means that every single cubic
26:27
millimeter of air pushed by that fan is working as hard as
26:30
possible to absorb heat right up until the exact moment it
26:34
exhausts out the back. If you nail that Bijan number
26:37
optimization, you can pack drastically more computing power
26:40
into the exact same physical server chassis without melting
26:43
the chips. We connect this to the bigger
26:45
picture. This geometric optimization
26:47
principle isn't just a clever trick for parallel plates, it is
26:50
a universal physical truth. The exact same methodology
26:54
intersecting the asymptotes of extreme physical geometries
26:57
based on the Beyond or Rayleigh numbers is used to optimize
27:00
natural convection cooling, forced turbulent duct flow,
27:03
complex arrays of staggered plates, and even bundles of
27:05
cooling cylinders. It scales to everything.
27:07
It is a fundamental law of allocating flow architecture in
27:10
a confined volume. But what happens when the fluid
27:13
isn't confined inside a channel? We've perfectly optimized our
27:17
internal flows, but what happens when the fluid flows over the
27:20
outside of our components, say an exposed sensor on an aircraft
27:24
or a standalone component in a wind tunnel that takes us out of
27:28
the constrained pipe and into the open ocean, so to speak?
27:31
External force convection. It is an entirely different
27:35
fluid, dynamic beast. When an object is in a free
27:38
stream, the boundary layer is no longer constrained by a physical
27:41
wall on the other side. It can grow, it can evolve, and
27:45
crucially, if the pressure conditions are wrong, it can
27:48
separate entirely from the object.
27:50
The mechanics of external flow cover flow over flat plates,
27:53
cylinders, and spheres, and the common thread tying all of this
27:56
together is the boundary layer battlefield.
27:59
It dictates absolutely everything.
28:01
The friction drag pulling on the object, the heat transfer
28:04
cooling the object, and the pressure drag behind the object.
28:07
And solving the mathematical equations for these external
28:10
boundary layers is notoriously punishingly difficult.
28:14
The Navier Stokes equations, which govern the conservation of
28:17
mass, momentum, and energy in a fluid, are highly nonlinear
28:20
partial differential equations. To actually solve them without a
28:24
supercomputer, classical engineers had to use a brilliant
28:28
mathematical trick called the similarity Transformation
28:31
technique. This is where we take a 2D
28:32
problem and collapse it. We combine the X&Y spatial
28:36
dependence, the distance along the plate, and the distance away
28:39
from the plate into a single independent variable, usually
28:42
called ETA. Why does this physical hack
28:45
allow the math to work? Because it turns an unsolvable
28:48
partial differential equation into a solvable ordinary
28:51
differential equation. Think about it physically.
28:54
If you take a snapshot of the velocity profile inside the
28:57
boundary layer one inch down the flat plate, and then take
28:59
another snapshot 10 inches down the plate, the boundary layer is
29:02
much thicker, but the fundamental shape of the curve
29:05
looks geometrically identical if you just scale up the zoom.
29:08
Because nature scales perfectly here, engineers can collapse 2
29:12
messy spatial dimensions into one single mathematical
29:15
variable. It takes an impossible 2D
29:17
calculus nightmare and turns it into a solvable line graph.
29:21
By using that single variable ETA, we can find an exact
29:24
analytical solution for the velocity and temperature fields
29:27
over a flat plate. But a perfectly flat plate
29:30
aligned perfectly with the wind is the easiest possible
29:33
scenario. Real engineering is messy.
29:35
The airflow hits things at angles.
29:37
This leads to Faulkner scan flow past a wedge.
29:40
Why do we care about a wedge shape?
29:42
Because a wedge fundamentally introduces a pressure gradient,
29:45
As fluid approaches a wedge, the free stream velocity changes
29:49
depending on the angle of that wedge.
29:51
If the fluid is being squeezed and forced to speed up, which we
29:54
call a favorable pressure gradient, the boundary layer is
29:57
physically pushed down against the surface.
29:59
It stays thin, tightly attached and heat transfer is excellent.
30:03
OK, that sounds ideal. But if the fluid is moving past
30:06
the wedge and beginning to slow down and adverse pressure
30:08
gradient, the pressure is actually pushing back against
30:11
the flow. And when the pressure pushes
30:13
back against a slow moving boundary layer that is already
30:17
losing energy to friction, the fluid literally stops, turns
30:21
around and separates from the surface entirely.
30:23
Exactly. And flow separation is the
30:26
absolute enemy of thermal fluid design.
30:28
It creates a massive turbulent wake, it skyrockets your
30:31
pressure drag, and it completely destroys your predictable heat
30:35
transfer. The Faulkner scan solutions
30:37
mathematically prove how fundamentally the heat transfer
30:39
off a solid surface shifts depending on whether that
30:42
pressure gradient is helping you or fighting you.
30:44
Let's bring this abstract wedge physics directly back to a real
30:48
world electronic cooling issue. Plate fin versus pin fin heat
30:52
sinks. We just spent all this time
30:54
perfectly optimizing the spacing of parallel plate fins using the
30:57
Bijan number. We nailed the design, but then
31:00
reality throws a massive curveball.
31:03
It does. As modern electronic systems
31:05
become increasingly compact. Think about the inside of a high
31:07
end laptop or a densely packed server blade.
31:10
The path for the cooling airflow gets severely constrained.
31:13
You can absolutely no longer guarantee a nice, clean,
31:16
perfectly straight airflow from your fan to your heat sink
31:18
because there's. There's just too much stuff in
31:20
the way. Right.
31:20
The air is violently bouncing off other components, turning
31:24
sharp corners and hitting the heat sink from completely
31:27
unpredictable shifting directions.
31:28
And plate fin heat sinks, the ones we just optimized, degrade
31:32
incredibly rapidly if the air doesn't hit them perfectly
31:35
straight on. If the air comes in, even at a
31:37
slight angle, it hits the broad side of the very first fin, the
31:41
flow separates massively, it creates a huge turbulent wake,
31:45
and the rest of the carefully spaced channel is entirely
31:48
starved of cooling air. Your Bijan number optimization
31:51
is ruined because the air bypasses the channel.
31:54
The physical solution to this unpredictable, chaotic airflow
31:58
is the pin fin heat sink. Instead of long continuous
32:01
aluminum plates, you machine a forest of small vertical
32:05
cylindrical pins. Pin fins are physically robust
32:08
against the direction of the airflow.
32:09
Because they're round. Exactly because they are
32:12
cylinders, it doesn't matter what angle the air approaches
32:14
from, the cross flow dynamics remain generally the same.
32:18
The air will weave its way through the forests of pins and
32:20
continue to extract heat regardless of the approach
32:23
angle. There is a very common, highly
32:26
optimized physical setup for this, placing a small axial fan
32:30
directly above a pin fin heat sink, blowing straight down onto
32:33
the chip. This is a brilliant use of
32:35
external flow geometry. The air from the fan blows
32:38
straight down. Initially it flows
32:40
longitudinally down the central pins, which behaves almost
32:43
exactly like flow along a flat plate.
32:46
But then, because it physically hits the solid baseboard
32:49
attached to the hot microchip, it has to abruptly turn 90°.
32:53
It then transitions into crossflow, moving horizontally
32:55
across the outer perimeter pins before finally exhausting out
32:59
the open sides of the heat sink. It's like a dual action cooler.
33:02
It leverages both internal parallel flow characteristics in
33:05
the center and external crossflow characteristics on the
33:08
edges. But wait, if pin fins are so
33:10
adaptable and they completely solve the problem of
33:13
unpredictable flow direction, why don't we just use them for
33:16
absolutely everything? Why do plate fins even still
33:19
exist? As always in mechanical
33:21
engineering, you cannot escape the harsh physical trade-offs.
33:25
Pin fins have massive drawbacks. First, simple manufacturing
33:29
complexity. Extruding a massive long piece
33:32
of aluminum with straight plate fins is incredibly cheap and
33:35
fast. You just push metal through a
33:37
die. Yeah, that makes sense.
33:39
But machining, forging or casting thousands of tiny,
33:42
perfectly spaced cylindrical pins is a slow, expensive
33:45
manufacturing process. Right, the cost scales up
33:48
quickly, but what about the fluid dynamics?
33:50
There has to be at a physical penalty too.
33:52
The penalty is the pressure drop.
33:54
While pin fins offer fantastic heat transfer, especially
33:57
because they generate constant turbulent wig zones that makes
34:00
the fluid, the aerodynamic drag coefficient of a bluff cylinder
34:03
in cross flow is extremely high compared to the sleek profile of
34:07
a flat plate perfectly aligned with the flow.
34:10
So it's harder to push the air through.
34:11
Much harder. You pay a massive penalty and
34:14
pressure drop. It requires a much stronger,
34:16
louder, much more power hungry fan to physically force the air
34:20
through a dense pin fin array. You only spend that money and
34:23
that fan power when the unpredictable flow direction
34:26
makes plate fins physically unviable.
34:28
It's always a balancing act. Heat transfer versus pumping
34:31
power, manufacturing costs versus thermal performance.
34:34
We've talked extensively about cooling, stationary objects,
34:37
servers, microchips, flat plates.
34:39
Now let's ramp up the complexity and look at the ultimate
34:42
mechanical engineering challenge, moving the fluid
34:45
itself with massive rotating blades.
34:48
Turbo machinery, where absolutely all of the simple 1D
34:51
rules become significantly more complicated and honestly
34:54
dangerous. We are diving into axial flow
34:57
compressors and fans. Think of the giant spinning
35:00
titanium blades at the front of a commercial jet engine, or a
35:02
massive industrial ventilation fan in a mine shaft.
35:06
The biggest misconception here, the absolute trap that ruins
35:09
amateur designs, is treating the flow as purely 2 dimensional and
35:12
entirely ignoring radial equilibrium.
35:15
When you look at a cross section of a compressor blade drawn on a
35:17
piece of paper or in a simple CAD model, it's very easy to
35:21
just assume the air moves straight through horizontally
35:24
from the front to the back in a nice clean 2 dimensional plane.
35:28
But physical reality is 3D right?
35:31
You have a fluid moving through an annular duct, a giant donut
35:34
shaped ring between the central in the outer casing, and that
35:38
fluid is swirling violently because the blades are spinning
35:42
at thousands of RPM. And swirling fluid means
35:45
centrifugal force. The air has mass, and it's
35:48
spinning. It physically wants to get flung
35:50
outward toward the outer casing, exactly like you get pushed hard
35:54
against the door of a car when the driver takes a sharp turn at
35:56
high speed. Exactly, to physically prevent
35:59
all the air from piling up against the outer wall and
36:01
starving the inner hub, there must be a balance of forces.
36:05
This is the radial equilibrium equation.
36:07
OK. What does that balance look
36:08
like? There absolutely must be a
36:10
radial pressure gradient, specifically a higher static
36:13
pressure at the outer casing and a lower static pressure at the
36:17
inner hub to balance the centrifugal forces of the
36:19
swirling fluid pushing outward. If you ignore this pressure
36:23
gradient in your design, your air won't go where your
36:26
spreadsheet thinks it will. The flow will completely
36:28
separate from the blades and your compressor will
36:31
catastrophically fail. This brings us to a classic
36:34
design approach called Free Vortex Flow.
36:37
So if I'm designing a compressor, my math looks
36:39
incredibly clean if I specifically designed the twist
36:42
of my blade to create a free Vortex swirl distribution.
36:46
Because if I do that, the math magically guarantees that the
36:49
axial velocity of the speed of the air moves straight through
36:52
the engine from front to back remains perfectly constant from
36:55
the inner hub all the way to the outer dip.
36:57
It was highly attractive to early aviation designers because
37:00
it is mathematically elegant. If the axial velocity is
37:03
constant across the entire radius, you can essentially
37:06
design the compressor stage as if it were a simple 2D cascade.
37:10
It makes the spreadsheet math incredibly easy to solve.
37:12
But practically, doesn't that mean I have to force the blade
37:15
to do extreme things to make that constant velocity a
37:18
reality? Don't I have to twist the blade
37:21
at massive angles near the hub? Exactly.
37:23
And that extreme twist is where the physical reality shatters
37:26
the mathematical elegance. To achieve that constant axial
37:30
velocity, the free vortex design physically forces the fluid to
37:34
deflect at extreme, almost impossible angles near the inner
37:38
wall, the hub. The blades have to be highly
37:41
twisted, which causes the flow to separate, creating a massive
37:44
stall zone right at the root of the blade.
37:47
And it gets worse at the outside edge, right?
37:49
Because the tip of the blade is spinning physically faster than
37:51
the root. Yes.
37:52
Conversely, near the outer wall, the tip casing, the free vortex
37:56
design forces the flow into dangerously high velocities,
37:59
easily pushing it into high Mach numbers.
38:01
To make my spreadsheet clean and my math easy, I physically
38:04
created supersonic shock waves tearing off the tip of my
38:07
blades, creating massive drag and noise, and I've created a
38:11
dead stall zone at the root of my blades.
38:13
Precisely, it is a perfect historical example of
38:17
theoretical mathematical elegance causing absolute
38:20
practical failure in hardware. Modern jet engine designers have
38:24
to move completely away from strict free vortex flow.
38:28
They intentionally designed for non constant axial velocity just
38:31
to physically manage those Mach numbers at the tip and prevent
38:34
flow separation at the root, even if it makes the 3D radial
38:38
equilibrium math exponentially harder to solve.
38:41
And the math gets even harder when the compressor isn't
38:43
running at it's perfectly designed, laboratory tested
38:46
speed, because no machine operates perfectly at its single
38:49
design point. No engine operates at its exact
38:52
design .100% of the time. You have to start it U from
38:55
zero, you have to throttle it down, you have to respond to
38:57
turbulent air conditions. The key Arameter engineers use
39:01
to manage this is the stage loading factor, denoted by the
39:04
Greek letter CI. It's a measure of exactly how
39:07
much thermodynamic work each single stage of the compressor
39:09
is physically doing on the air. And this loading factor is
39:12
strongly tied to the flow coefficient Phi, which
39:15
essentially represents the physical mass flow rate of the
39:17
air moving axially through the engine.
39:19
There is a truly fascinating counter intuitive insight from
39:23
experimental compressor cascade data regarding this.
39:26
If you have a heavily loaded stage where the stage loading
39:29
factor CSI is very high approaching one, it is actually
39:32
the most flexible and robust design off design.
39:35
It physically produces very little variation of performance,
39:39
even when the flow, the mass flow of the air changes
39:42
significantly. Wait, really?
39:44
You'd intuitively think a compressor stage working at its
39:47
absolute maximum thermodynamic limit would be fragile, on the
39:50
edge of failure, and prone to breaking down if the wind
39:53
changes even slightly. It is entirely counterintuitive,
39:56
but the physical physics of the velocity triangles prove it.
39:59
It's actually the lightly loaded stages where PSI is very low,
40:02
doing very little work per stage, that are highly sensitive
40:05
and fragile. Why is that?
40:07
In a lightly loaded stage, even a tiny change in the physical
40:10
mass flow rate causes massive disproportionate changes in the
40:13
relative flow angles. The air suddenly hits the
40:16
physical metal blades at the completely wrong angle of
40:18
attack. The boundary layer violently
40:20
separates and it leads directly to stall and catastrophic
40:23
compressor surge. Let's visualize this with a
40:26
commercial jet engine. Say you design an engine purely
40:30
for cruising altitude. That's your ristine design
40:32
point. The air is thin, the speed is
40:35
constant. It's a perfectly controlled
40:37
environment if you decide to use lightly loaded stages to
40:40
maximize your aerodynamic efficiency.
40:43
At that specific point, you get great gas mileage at 30,000
40:46
feet. But what happens on the runway?
40:50
When that jet is taking off, when the air at sea level is
40:53
incredibly dense, the pilot is throttling the engine up to
40:56
maximum power, and the mass flow is completely different from the
40:59
cruise condition. Those lightly loaded stages will
41:02
be pushed incredibly far off their design point, the flow
41:05
angles will shift wildly, the boundary layers will detach from
41:08
those long blades, and the engine could stall violently,
41:11
losing all thrust. Right when the aircraft needs
41:14
maximum power to leave the ground, the design fails because
41:17
you haven't accounted for the off design flow coefficient
41:19
shifts. This raises an important
41:21
question. How do modern engineers actually
41:24
handle this in the real world? If free vortex math is flawed
41:28
and lightly loaded stages are dangerously sensitive, and the
41:32
3D radial equilibrium math is a nightmare, how do we actually
41:37
build functioning jet engines? They rely on incredibly complex
41:40
computational time marching methods.
41:43
They use powerful CFD software to simulate the flow.
41:47
The computation physically starts from an assumed guest
41:49
flow field, and the governing Navier Stokes equations are
41:52
mathematically marched forward in time, step by step, until
41:55
they reach a steady, converged physical solution.
41:58
It can handle local supersonic patches, shockwaves, and
42:01
incredibly complex 3D secondary flows that manual math could
42:04
never touch. But you cannot just blindly
42:06
trust the software. Absolutely not, and this is the
42:09
most critical take away for any thermal fluid engineer.
42:11
If you do not intimately understand the base physics we
42:14
just discussed, the radial equilibrium enforcing air, the
42:17
casing, the boundary layer separation on a wedge, and the
42:19
extreme danger of lightly loaded stages, you will just suffer
42:22
from garbage in garbage out software.
42:24
Errors. The software will just lie to
42:25
you beautifully. Exactly.
42:27
The CFD software will happily output a beautiful, colorful,
42:31
completely physically impossible 3D model.
42:35
The software is merely a tool. The human engineer must
42:37
fundamentally understand the physical reality it is
42:40
attempting to simulate. We've covered a massive amount
42:43
of physical ground today. We started at the absolute base
42:47
with the Karno cycle, realizing that maximum theoretical
42:50
efficiency physically guarantees 0 power output.
42:53
We bypass the impossibility of measuring entropy by using the
42:56
Maxwell relations to measure pressure.
42:58
Instead, we optimize server farm cooling channels by finding the
43:02
exact physical intersection of pressure drop and boundary layer
43:05
growth using the Bijan number. We navigated the chaotic
43:08
external boundary layers using pin fins and we just survived
43:12
the 3D swirling, shockwave inducing nightmare of turbo
43:14
machinery. Let's pull this all together by
43:17
answering the specific painful questions you, the listener, are
43:20
likely screaming at your dashboard right now.
43:22
It is always essential to connect this high level theory
43:24
directly back to the daily physical frustrations engineers
43:27
face of their CAD stations. Exactly.
43:29
I'm going to rapid fire the most common painful questions
43:33
engineers asked about these topics, and let's apply this
43:35
exact deep dive directly to their daily work.
43:38
Question number one, dealing with internal flow.
43:42
Why does my physical heat sink prototype perform drastically,
43:45
measurably worse in the lab testing than my simple design
43:48
spreadsheet calculated it would? This is the absolute classic
43:52
entrance region and bypass failure.
43:55
Your spreadsheet failed because you likely used a fully
43:57
developed Nusselt number correlation assuming perfect
44:01
unchanging heat transfer for a physical flow that is actually
44:04
still entirely trapped in the entrance region.
44:06
If your physical heat sink is short, the thermal boundary
44:09
layer never fully develops, and your average heat transfer is
44:12
entirely different from what you calculated.
44:14
And what about the physical spacing?
44:15
That's the other part. You almost certainly didn't
44:18
physically account for flow bypass.
44:20
As we proved with the Bijan number optimization, channel
44:24
spacing is the critical physical constraint.
44:26
If your fins are packed too tightly to maximize surface
44:29
area, the internal friction, the pressure drop is too high for
44:32
your fan, the air will absolutely not force its way
44:35
through your perfectly calculated, highly restrictive
44:38
fins. It will obey the laws of
44:40
physics, take the path of least resistance, and simply flow
44:43
around the outside of your heat sink entirely.
44:45
The air just dodges it. Yep, your spreadsheet assumed
44:48
the mass flow had to go through the fins.
44:50
The physical reality of fluid dynamics proved otherwise.
44:53
A tough, expensive lesson in physics, always taking the easy
44:55
way out. All right, question 2.
44:59
Diving into the terrifying world of turbo machinery.
45:03
Why do my physical compressor blades keep stalling and
45:05
searching when I simply throttle down the mass flow?
45:08
You are physically hitting the surge line on your compressor's
45:11
performance map. When you reduce the mass flow of
45:14
the throttle valve, but you maintain a constant rotational
45:17
speed of the spinning blades. You are directly and drastically
45:20
dropping your physical flow coefficient.
45:23
As we explored in the OFF Design Performance Physics, this drop
45:26
in mass flow radically alters the relative flow angles of the
45:30
air hitting the moving blade. So it hits at the wrong angle.
45:34
Exactly. The air is now physically
45:36
approaching the leading edge of the blades at far too steep of
45:39
an angle of attack. The boundary layer simply cannot
45:42
navigate that sharp turn and remain attached.
45:45
The fluid violently separates from the low pressure suction
45:47
side of the blade, creating a dead wake zone and you get a
45:50
local blade stall. And if that cascade fails.
45:52
If this happens across enough blades simultaneously, the
45:55
entire compressor loses its ability to hold back the high
45:57
pressure downstream. The high pressure air violently
46:00
reverses direction, blasting backward out the front of the
46:04
engine, which is a compressor surge and it can physically tear
46:07
the engine off its mounts. Wow.
46:09
OK, final question. Question three.
46:11
Dealing with external flow and boundary layers.
46:15
How do I physically deal with turbulent flow over rough
46:18
surfaces? If my manufactured part has a
46:20
high surface roughness, does it completely ruin my aerodynamic
46:24
heat transfer? Not necessarily.
46:26
In fact, depending on your physical goals, a rough surface
46:29
might be exactly what you want for heat transfer, but you have
46:32
to pay the toll and power. Physical surface roughness
46:35
actually acts to trip the boundary layer.
46:37
It forces the smooth laminar flow to break down into chaotic
46:40
turbulence much earlier along the plate, and it physically
46:44
thickens the turbulent sub layer near the wall.
46:46
OK, so it gets messy fast. Yes, this increases the local
46:49
skin friction coefficient, meaning the physical drag
46:51
pulling on the object skyrockets.
46:54
But through what is mathematically known as the
46:56
Reynolds analogy, a direct physical increase in skin
46:58
friction corresponds directly to an increase in the heat
47:01
transfer, specifically the dimensionless Stanton number.
47:04
Wait, so making the metal physically rougher actually
47:06
cools the component better? Yes, the physical roughness
47:10
violently churns the turbulent prantal number in the sub layer,
47:14
promoting massive physical mixing and aggressively ripping
47:18
heat away from the hot wall faster than a smooth insulating
47:21
laminar layer ever could. It is a harsh physical trade
47:25
off. You get significantly better,
47:27
more aggressive cooling, but you pay a massive penalty and
47:30
pressure drop, aerodynamic drag, and the required fan power to
47:33
overcome that artificially increased friction.
47:35
It all comes completely back to that physical balance.
47:38
Heat transfer versus pumping power.
47:40
Theoretical efficiency versus actual usable power output.
47:44
Before we wrap up, I want to leave you with something to
47:46
really chew on that builds on exactly everything we've
47:48
discussed today. We've talked extensively about
47:51
optimizing fixed physical channel spacing, using the Bijan
47:55
number for a static electronics cooling, finding that perfect
47:58
unchanging geometry where the physical asymptotes cross.
48:02
But what if future mechanical designs weren't physically
48:04
static? What do you mean?
48:05
Like physically shape shifting heat sinks inside a computer.
48:08
Exactly. Imagine a mechanical system with
48:12
shape shifting responsive cooling channels.
48:15
A physical system that bio mimics biological vascular
48:18
systems exactly like the capillaries in your own body.
48:21
Channels that dynamically physically expand and contract
48:24
in real time to maintain the perfect optimal local Bijan
48:28
number as the heat load fluctuates across a microchip.
48:31
That sounds like science fiction.
48:32
It does, but think about it. If one core of a processor
48:35
suddenly runs incredibly hot, the micro channels directly over
48:38
it physically dilate to increase local mass flow while
48:41
simultaneously constricting over cooler areas to save fan power.
48:45
That physically dynamic, constantly adapting architecture
48:48
is the absolute next frontier of thermal fluid design.
48:51
So what does this all mean? It means that whether you are
48:54
mathematically calculating unmeasurable entropy to find
48:57
your physical maximum power limit, designing a titanium jet
49:00
engine compressor that won't violently stall on a dense sea
49:03
level take off, or trying to pack just one more physical
49:06
microchip into a server farm without melting the silicon,
49:10
relying on simplified, pristine 1D textbook assumptions will
49:13
completely trap you. You have to design for the
49:16
messy, 3 dimensional, highly chaotic, off design physical
49:19
reality. The X-ray machine might be
49:22
broken, but with the right mathematical relationships and a
49:24
deep, uncompromising physical intuition, you can still figure
49:27
out exactly what's going on inside the machine.
49:29
If you got something useful out of today's episode, hit follow
49:32
and drop a like. It helps grow the show and keeps
49:34
the engineering knowledge flowing.