Latest / Mechanical Engineering Made Simple / (#56) Industrial Reactor Lifecycle From Micro to Macro

Transcript

(#56) Industrial Reactor Lifecycle From Micro to Macro · Mechanical Engineering Made Simple

0:00
Design Junkie Family, you did real work this year.
0:03
You helped push this podcast into the top 20% on Spotify for
0:06
2025. Engineers know progress is never
0:09
an accident. It takes repetition, curiosity,
0:12
and a community that shows up. To everyone who rated,
0:15
subscribed and stayed in grind with us, welcome home.
0:18
You are part of this workshop now and this milestone is yours
0:21
as much as mine. And if you are new here, step in
0:23
and join the crew. Every great build starts with
0:25
the first cut. Your input matters, your rating
0:27
matters, and if you enjoy the podcast, show your support with
0:30
a like, a follow or a comment. Let's keep designing, learning
0:33
and leveling up together. Welcome to mechanical
0:36
engineering made simple. Welcome everyone.
0:38
Today we are strapping in for a a massive deep dive that sits
0:42
right at the heart of the chemical and process industries.
0:45
We are investigating the entire life cycle of industrial
0:48
reaction systems. We're going from the microscopic
0:50
physics that, you know, govern how 2 molecules meet all the way
0:54
up to the massive multi ton vessels that contain these
0:57
processes and of course the mechanical integrity that keeps
1:00
them safe. Our mission today is explicitly
1:02
focused on giving you the shortcuts to expertise in
1:05
industrial process engineering. We've pulled together a huge
1:09
stack of sources covering everything from fluid to
1:11
mechanics, and we're talking simple Newtonian fluids all the
1:15
way to complex slurries, all the way through reaction kinetics,
1:19
advanced mixing, and the critical engineering standards
1:21
that govern how you build the vessel itself.
1:23
And we are really focusing on that transition, that moment
1:26
where a process jumps from, you know, optimizing a small batch
1:30
in the lab to achieving consistent, efficient and most
1:33
importantly, safe operations at the production scale.
1:37
It's a leap that requires a complete shift in engineering
1:40
focus. And that shift is all about
1:42
understanding the link between scales.
1:44
I think the key knowledge nugget we want you to walk away with
1:46
today is this. True expertise comes from seeing
1:50
the relationship between local microscopic phenomena.
1:52
Like what exactly? Things like how fast a tiny Eddy
1:55
dissipates its energy, or how a single solid particle interacts
1:59
with a fluid boundary layer and connecting that to the resulting
2:03
global outcomes. So things like final product
2:05
yield or energy usage. Or even catastrophic vessel
2:08
failure. If you can connect those dots
2:10
from the micro to the macro, you've mastered the process.
2:14
OK, let's unpack this. We have to start small because
2:16
the entire philosophy of a scale up absolutely has to begin in
2:20
the lab. If you rush the R&D, you're
2:23
going to pay for it tenfold later on the plant floor.
2:26
That is the core principle of process development.
2:28
You must must perform extensive investigative work at the
2:33
laboratory scale first. And why is that so non
2:35
negotiable? Because in the lab you can
2:38
easily and rapidly investigate a huge number of parameters,
2:41
temperature, concentration, agitation speed, different
2:44
solvents, and you can fail cheaply and quickly.
2:47
This iterative process at the small scale leads to rapid
2:50
process optimization before you commit serious capital to steal
2:54
and concrete. And the scaling strategy itself
2:56
needs to be disciplined. The sources are really clear on
2:59
this. You should never jump straight
3:01
from a beaker to a 50,000 gallon reactor.
3:03
No, absolutely not. You should utilize several
3:05
vessel sizes. You know, lab bench scale, then
3:08
a pilot plan and only then move to full production.
3:10
The full scale plant isn't for R&D then.
3:13
It's role should not be R&D. Experiments on the production
3:17
line should be incredibly selective.
3:19
They're primarily for the final confirmation of your parameters,
3:22
maybe fine tuning the last 3% of efficiency, or investigating a
3:26
problem that's already happening.
3:28
You need to know the physics works before you hit the big
3:30
green button. So to make that stepwise scale
3:33
up, reliable engineers lean on standardized equipment.
3:37
Let's talk about the concept of a standard mixing vessel.
3:40
Right. So this is usually defined as a
3:42
vessel where the liquid filled depth, which we'll call H is
3:45
equal to the tank diameter T. It's a one to one ratio.
3:49
Standardizing that geometry helps minimize the variables
3:52
when you're scaling up. It does, but a quick design note
3:55
for anyone actually specifying these vessels, the overall
3:58
vessel height should be at least 30% greater than that maximum
4:02
fill depth. And what's that extra headroom
4:03
for? It's essential for coping with
4:05
surface splashing, especially at high agitation speeds, and also
4:09
for accommodating what's called gas holdup.
4:12
If you're sparging air into a fermentation, for instance,
4:15
those gas bubbles will expand the liquid volume.
4:18
You need space for that. And we're even seeing companies
4:21
get clever with manufacturing a standardization through modular
4:25
vessels. Oh yeah, this is a great
4:26
efficiency play. They use standardized
4:29
cylindrical sections and then interchangeable bases like a
4:32
tors spherical head which lets them create a huge variety of
4:36
vessel configurations using standardized, pre certified and
4:40
easy to swap components. So it's like building with
4:42
industrial scale Legos. It saves on custom fabrication
4:45
costs. It does, and it greatly speeds
4:47
up pilot testing when you can quickly assemble and disassemble
4:49
different tank geometries. All right, let's move inside the
4:52
lab and talk tools for a mixing engineer trying to nail down
4:56
these parameters. What's in the Essential Toolkit
4:58
for measurement? Well, the basics are absolutely
5:01
critical. A handheld optical tachometer is
5:03
essential for checking and calibrating your electronic
5:06
agitator speed readouts. You can't just trust the dial A
5:09
gut check exactly. You also need accurate averaging
5:12
voltmeters to read data from things like strain gauges and a
5:16
precise temperature controlled recirculation system to ensure
5:19
thermal stability. Rheology and reaction kinetics
5:22
are so sensitive to temperature. But the real complexity I think
5:26
comes in measuring the power input, specifically the torque.
5:30
This isn't a simple multi meter job.
5:32
We're talking strain gauges and slip rings.
5:34
That's right. Measuring the torque transmitted
5:37
through that mixing shaft is crucial because it directly
5:39
translates to the power being dissipated into the fluid.
5:43
Strain gauges, which measure the tiny mechanical deformation
5:46
caused by torsion, are typically watered through slip rings or
5:50
more modern inductive couplings to get the signal off the
5:53
rotating shaft. And what's the biggest pitfall
5:55
in setting that up? Where do people go wrong?
5:58
Calibration. It is absolutely non negotiable
6:00
and it's extremely challenging to do right.
6:02
You have to perform an external calibration using known
6:05
quantifiable torques without accidentally applying any
6:08
bending loads to the shaft. And bending loads will just
6:10
confuse the signal. Completely.
6:12
It leads to significant errors in your power input calculation.
6:16
And placement matters too. Where do these gauges actually
6:20
go? If you're just measuring the
6:23
total power draw, a single gauge set can be placed high on the
6:26
shaft above the liquid level. But if you're doing detailed
6:30
process optimization, especially in a system with multiple
6:34
impellers. Say 1 near the top and one near
6:36
the bottom. Exactly.
6:37
Maybe a feed impeller near the top and a gas dispersing
6:40
impeller at the bottom. In that case you need multiple
6:43
gauges. They're positioned strategically
6:46
to measure the torque contributions from each
6:48
individual impeller, which lets you understand how much power is
6:51
being consumed in different parts of the tank.
6:53
That deeper understanding of localized power is key.
6:57
Shifting gears slightly, how do engineers quantify the quality
7:00
of mixing? It feels like a very qualitative
7:03
thing, but they use blending time or mixing time.
7:06
It does feel qualitative, but we have some great quantitative
7:09
methods. One of the oldest and still one
7:11
of the most effective is the acid base with indicator method.
7:13
OK. So you're basically injecting a
7:15
dye that also chemically reacts. Precisely, you introduce a small
7:19
pulse of acid or base into the vessel which already contains an
7:22
indicator, something like bromphaphenol blue which changes
7:26
color from blue to yellow, or phenolphthalene which goes from
7:30
pink to colorless. And the visual endpoint, the
7:33
last little swirl of color to disappear.
7:36
That tells you something important.
7:37
It gives you the point of last mixing, and that location is
7:41
often the perfect spot to strategically position a
7:43
conductivity or temperature probe for future process
7:46
control. You found the worst spot in the
7:49
tank. So once that time which we can
7:51
call Theta is determined, engineers use the dimensionless
7:55
mixing time N Theta. N Theta is one of the key
7:58
scaling parameters in mixing. It's the product of the measured
8:01
mixing time Theta and the impeller speed N It represents
8:05
the total number of impeller revolutions required to achieve,
8:08
say, a 95% state of uniformity. And if you can keep that
8:11
dimensionless number constant as you scale up.
8:14
You have a very solid basis for transferring the process to a
8:17
larger vessel with a good chance of success.
8:19
And this leads us to a crucial point about homogeneity.
8:23
Perfect 100% mixing is exponentially difficult to
8:27
achieve. We actually have formulas to
8:30
calculate the time required for higher degrees of mixedness.
8:33
We do. For example, if you figure out
8:36
the time it takes to get to 95% homogeneity, you can calculate
8:40
the time it takes to get to 99% homogeneity just by multiplying
8:44
that 95% time by 1.54. The practical implication of
8:48
that small multiplier 1.54 is massive, isn't it?
8:52
It's. Huge.
8:53
It tells a designer that achieving that final 4% of
8:56
homogeneity going from 95 to 99 requires over 50% more agitation
9:01
time and energy. And that difference is critical
9:03
for costing the power input and maybe avoiding unnecessary
9:06
energy consumption if 95% is actually good enough for the
9:10
product spec. It also reflects the physical
9:12
reality. As you get closer and closer to
9:14
100%, the remaining segregated pockets of fluid becomes smaller
9:18
and harder to reach, requiring dramatically more energy
9:21
dissipation to finally break them up.
9:23
Now, a big challenge when taking all these measurements is
9:26
intrusiveness. Any instrument you put in the
9:28
tank is going to fundamentally change the flow.
9:31
That's the Heisenberg uncertainty principle of
9:33
chemical engineering. Any probe, whether it's a
9:36
conductivity probe measuring local solids concentration or an
9:40
RTD measuring temperature, is intrusive.
9:43
It's a physical obstacle that can alter local flows and, well,
9:47
skew the very data you're trying to collect.
9:50
So how do you even check if your measurement probe is fouling
9:53
your own results? One way is to check for flow
9:55
separation. For example, if you're using a
9:58
conductivity probe to measure solid particles in a conducting
10:01
liquid, you can rotate the probe slowly.
10:04
If the measured solids concentration changes as you
10:06
rotate it, you know flow separation is happening because
10:09
of the probes physical presence. And if you confirm flow
10:12
separation is happening, what do you do?
10:14
The measured value of the solids concentration will generally be
10:17
too low because the probe is creating a wake of low
10:20
concentration fluid behind it. So in this scenario, standard
10:24
engineering practice is to take the highest measured
10:26
concentration value you recorded during that test as the most
10:30
representative value, while acknowledging the inherent
10:33
limitations of the instrument. That flows nicely into our next
10:36
big topic, the dynamics of mixing.
10:39
The challenge sharpens immediately when the fluid
10:42
itself is complex. I mean non Newtonian or when the
10:45
speed of mixing competes directly with the speed of a
10:48
chemical reaction. And we have to start with non
10:50
Newtonian fluids. For engineers used to simple
10:52
things like water, oil or ethanol, where viscosity is just
10:56
a constant number, non Newtonian fluids are a different beast.
10:59
Right here we define the apparent viscosity MU a not as a
11:03
constant, but as a function of the shear rate itself.
11:05
So the fluid's thickness literally changes depending on
11:08
how aggressively you stir it. Precisely.
11:11
We classify them in different ways, but the sources highlight
11:14
time independent fluids. For these, the shear stress at
11:17
any point depends only on the instantaneous shear rate.
11:20
At that exact point. It doesn't matter how long it's
11:23
been stirred, just how fast it's being stirred right now.
11:25
And in the lab, engineers need proxy fluids to mimic these
11:29
often nasty industrial conditions.
11:31
They do. We use test fluids like
11:32
carboxymethyl cellulose or CMC, but its solutions tend to be
11:37
cloudy, which immediately makes them unsuitable for optical
11:40
visualization techniques like laser Doppler anemometry or
11:43
particle image velocimetry. OK, So what else is there?
11:46
Then you have carbopole. It's useful for creating fluids
11:49
that have what's called a yield stress, but it's rheology is
11:52
extremely sensitive to changes in pH and for a classic shear
11:56
thinning fluid. Where the viscosity goes down as
11:58
you start it faster. Exactly.
12:00
The go to is often nitrosol or hydroxyethyl cellulose.
12:05
It produces fluids that exhibit that classic shear thinning
12:08
behavior, which is really common in things like polymerization
12:11
and food products. And the complexity of these
12:14
fluids leads to a major design challenge, especially in large
12:18
fermentations or polymer processing, something called
12:21
cavern formation. Cavern formation is the classic
12:24
sign that you have insufficient power input in a highly viscous
12:29
shear, thinning or yield stress fluid like in xanthan gum
12:33
fermentations. What does that look like?
12:35
The impeller spins, but because the fluid has structural
12:37
integrity, the mixing energy only creates a well mixed cavern
12:41
of high shear, low viscosity material right around the
12:45
impeller. The vast majority of the
12:47
vessel's contents just sit there, stagnant and unmixed.
12:50
Which is an enormous operational failure.
12:52
You think you're mixing the whole tank, but you're really
12:54
just turning a hole in the middle of.
12:55
It it creates dead zones where you get huge temperature or
12:59
nutrient gradients. So engineers have developed
13:01
models to predict the diameter of this cavern DC.
13:05
These models typically operate under the assumption that the
13:07
stress imparted by the impeller at the cavern boundary is
13:11
exactly equal to the fluids yield stress Tall, wide.
13:14
It's basically a force balance. It's a force balance you're
13:17
trying to match the mechanical force of your agitation against
13:20
the internal yield struck that viscous goo.
13:23
And we see extreme versions of this in industrial processes
13:26
like pulp suspension rheology. Oh, absolutely.
13:30
The sources point out that pulp suspensions can exhibit solid
13:33
like behavior. They act like a rigid plug even
13:36
when they are 90% water by mass, we're talking at only 10% mass
13:40
concentration. That is just so
13:42
counterintuitive. It is, but the mechanical
13:45
behavior is completely dominated by the interlocking fiber
13:48
matrix. To get it to move, you have to
13:50
overcome the suspensions yield stress tall Y, which increases
13:54
dramatically with mass concentration and crucially in
13:58
these pulp systems, the macro flow, the bulk movement often
14:01
begins before the fiber flocks are fully disrupted.
14:04
So you're moving the whole mass, but you haven't yet achieved the
14:07
micro dispersion you need for an effective chemical reaction like
14:10
bleaching. Exactly.
14:11
It's a huge challenge. Let's move to the chemical side,
14:14
then the mixing reaction Nexus, specifically micro and mezzo
14:19
mixing. This is where we learn that
14:21
those global metrics, like a 10 minute blend time, can sometimes
14:25
be completely meaningless. This is the world of selectivity
14:28
and yield In fast chemical reactions, particularly what are
14:31
called competitive consecutive schemes, where your desired
14:34
product R is prone to reacting immediately with more reactant B
14:39
to form an undesirable by product S The local mixing time
14:43
constants control the final yield, not the overall vessel
14:46
blend time. Because if the mixing is slow,
14:48
the reactant you just added B finds a huge local concentration
14:52
of A or R right at the feed pipe, and it leads to immediate
14:54
overreaction and terrible selectivity for your desired
14:57
product R. And this is a fundamental law
15:00
for homogeneous reactions. Mixing effects can only reduce
15:03
the yield below the ideal theoretical maximum.
15:05
They can never improve it past that point.
15:08
The moment you introduce a slow mixing step into a fast
15:10
reaction, you start losing valuable product.
15:13
To analyze this problem properly, we have to talk about
15:16
turbulent scales, starting with the smallest, most fundamental
15:20
time scale, the Kolmogorov time scale.
15:22
Right. The Kolmogorov timescale, Tau K
15:26
represents the time required for the smallest possible Eddy in
15:29
the flow to dissipate its energy.
15:31
It's calculated using the kinematic viscosity nu and the
15:34
rate of dissipation of turbulent kinetic energy per unit mass,
15:37
which is epsilon. This tiny timescale dictates the
15:41
rate of true molecular mixing, where molecules finally get to
15:44
touch. So we have a tiniest scale micro
15:46
mixing which is happening below that Kolmogorov scale and is
15:49
highly sensitive to viscosity. Where does meso mixing fit into
15:53
this picture? Meso mixing occurs at the
15:56
intermediate scales of turbulence.
15:58
It's controlled primarily by the initial scale of segregation LS,
16:02
which is determined by the engineering design of your
16:04
injection point, like the diameter of your feed pipe.
16:07
And that's where most of the reaction happens.
16:08
It's where the bulk of the reaction occurs before you get
16:10
to full molecular contact. And unlike micro mixing, meso
16:14
mixing is less affected by viscosity.
16:16
This analysis gives us the critical litmus test for
16:19
reaction control. The mixing dam Kohler number D
16:24
AM. D AM is a magnificent
16:26
dimensionless number. It's defined as the ratio of the
16:29
mixing time constant Tau M to the reaction time constant Tau
16:33
R. It tells you very simply whether
16:36
mixing or reaction is the limiting step.
16:38
So if mixing is much faster than the reaction, DAM is small and
16:42
the reaction is what we call rate limited.
16:44
Correct, but where does the mixing start?
16:46
Becoming the dominant controlling variable, the part
16:49
that dictates your yield. Yeah, where's the tipping point?
16:52
Studies show that the critical changes in yield, meaning the
16:55
process shifts from being reaction controlled to mixing
16:58
controlled, occur when that mixing dam color number falls
17:01
between 0.01 and 0.05 five. That is the operational danger
17:06
zone. If your reaction kinetics push
17:08
you into that narrow band, you must use specialized engineering
17:12
solutions. Knowing that 0.05 threshold is
17:15
an incredible technical shortcut, So what are the
17:18
practical engineering fixes when you're designing A reactor for a
17:21
mixing sensitive reaction in that critical range?
17:24
You have to ensure rapid incorporation of the reactant.
17:28
That means moving away from a single feed pipe and using
17:31
techniques like multiple injection points distributed
17:33
around the high shear. Impeller zone.
17:35
You could also design A rapid recycle loop.
17:37
Yes, that draws fluid from the bulk and injects the reactant
17:41
directly into that high shear zone right near the impeller
17:44
tip. You want to get it dispersed as
17:45
fast as humanly possible. The sources also point to using
17:49
inline static mixers for pre mixing, often right before the
17:52
main reaction vessel. They are an excellent solution
17:55
for this. You'd use a static mixer, maybe
17:57
with a length to diameter ratio of four to ensure intimate high
18:01
intensity mixing over a very short residence time, maybe just
18:05
one second. This guarantees the reactants
18:07
are fully mixed before they even enter the main reactor,
18:10
mitigating those large scale segregation problems.
18:13
Now let's pivot to another type of complex system, liquid liquid
18:16
dispersions. Like creating an emulsion.
18:20
This involves a constant dynamic battle between dispersion,
18:23
breaking drops apart, and coalescence, where drops
18:26
recombine. And the key concept here is the
18:29
maximum stable drop size D Max. This is the size at which the
18:33
disruptive forces from turbulence are perfectly
18:35
balanced by the cohesive forces holding the drop together.
18:38
That's the interfacial tension and the drops internal
18:41
viscosity. So if a drop is bigger than D
18:44
Max, the turbulent energy rips it apart.
18:47
If it's smaller, it survives. But the dynamics are complicated
18:51
by coalescence, the drops sticking back together.
18:54
They are. Coalescence depends on 2
18:57
independent factors that have to happen in sequence. 1st is the
19:00
collision frequency which is on the city.
19:02
This is simply how often the drops crash into each other and
19:05
it goes up with agitation rate and how much dispersed phase you
19:08
have. OK, so that's the first step.
19:09
What's the second? The second is the coalescence
19:11
efficiency Lambda. This is the probability that
19:14
when they do collide, the thin film of continuous phase fluid
19:18
separating the 2 drops will drain and ruptor, allowing the
19:21
drops to merge into one. And here is the truly
19:23
fascinating and kind of counterintuitive part, the role
19:27
of local energy dissipation, epsilon.
19:30
It creates this conflicting effect that designers have to
19:32
manage. On one hand, the collision
19:35
frequency bagela increases with local energy dissipation.
19:39
More power means more turbulence and more crashes.
19:42
It's simple. But on the other hand.
19:44
On the other hand, the coalescence efficiency Lambda
19:48
actually decreases with increased epsilon.
19:50
Wait, so crashing them together harder makes them less likely to
19:54
stick together? Why is that?
19:56
Because the duration of the contact is reduced for two drops
19:59
to merge, the contact time during the collision has to be
20:02
longer than the critical film drainage time, the time it takes
20:06
for that fluid between them to drain away.
20:08
So higher turbulence means they just bounce off each other too
20:11
quickly. Exactly.
20:12
The higher relative velocities lead to shorter contact times,
20:15
making film rupture less likely before the drops just fly apart
20:19
again. And the drainage time itself is
20:21
sensitive to the chemistry. Mass transfer between the phases
20:24
can drastically shorten that critical film drainage time.
20:28
If you have a salute moving from one drop to the other, it
20:31
creates local surface tension gradients that destabilize the
20:34
film, making coalescence much more efficient even at high
20:38
turbulence. To handle these highly dynamic,
20:41
constantly changing systems, liquid liquid dispersions
20:44
crystallization polymerization engineers have to rely on
20:48
advanced modeling tools called population balance equations, or
20:53
PBE's. PBE's are essentially a detailed
20:56
computational accounting system for the entire particle or drop
20:58
distribution within a reactor. They track the evolution of the
21:02
drop size distribution over time.
21:04
And what are the four simultaneous events that a PBE
21:06
tracks? It tracks convection, so drops
21:08
moving in and out of the control volume, and it tracks the
21:11
processes of birth and death. Birth happens either by
21:14
breakage, where one large drop splits into two smaller ones, or
21:18
by coalescence, where two small drops combine.
21:20
And death is just the reverse of that.
21:22
Exactly. Death occurs when drops are lost
21:25
from a certain size class due to those same breakage or
21:28
coalescence events. PB ES are incredibly powerful
21:32
for predicting these distribution changes, especially
21:35
when you couple them with detailed CFD flow information.
21:39
And finally, in these liquid liquid systems, you can get the
21:42
dramatic event of phase inversion when the whole system
21:45
flips, say from an oil and water emulsion to a water and oil one.
21:50
Phase inversion is a very system specific phenomenon and it's
21:53
strongly influenced by the type of surface active agents you
21:56
have. The critical take away from the
21:58
research is that coalescence, the sticking together of the
22:01
dispersed drops, typically dominates as the controlling
22:03
mechanism that triggers the inversion.
22:05
So if you want to delay or prevent phase inversion, you
22:09
need to minimize the factors that promote coalescence.
22:12
Exactly. Anything that influences those
22:14
film drainage rates, like reducing the interface mobility,
22:18
increasing the viscosity of the continuous phase or the contact
22:21
time. These are the critical controls
22:23
for managing or preventing that catastrophic shift.
22:26
OK, to master these complex dynamic systems, we need eyes
22:30
inside the reactor, which leads us directly to advanced flow
22:33
diagnostics. We need to be able to measure
22:35
turbulence and flow patterns without actually touching the
22:38
fluid. And the non invasive standards
22:40
here are particle image velocimetry or PIV, and laser
22:45
Doppler anemometry or LDA. They both use lasers and tiny
22:49
seed particles to gather information, but they provide
22:52
different kinds of data. LDA is also called the gold
22:55
standard for turbulence measurement.
22:57
Why is it preferred over PIV for that specific task?
23:00
It really comes down to scale and speed.
23:02
LDA uses a very small sensor volume, just a few micrometers
23:06
in diameter, and has an extremely fast response time.
23:10
This high spatial resolution lets it resolve turbulent scales
23:13
as small as two to three times the Kolmogorov scale, which is
23:16
essential for getting accurate localized turbulence energy
23:18
dissipation measurements. And PIV on the other hand gives
23:21
you the big picture. PIV gives you a simultaneous
23:24
snapshot of the velocity field over a large 2 dimensional
23:27
plane, but conventional 2 DPIV only measures the projections of
23:32
the 3D velocity vectors onto that light sheet.
23:34
If the velocity component that's normal to the light sheet is
23:37
significant, PIV can give you inaccurate results.
23:40
A kind of visual distortion? A parallax error.
23:43
Exactly a parallax error, especially near the edges of the
23:46
image. Which means if a designer needs
23:48
the full 3 dimensional flow field, 2 DPIV just isn't going
23:52
to cut it. You need stereoscopic PIV or
23:55
three DPIV. This setup uses multiple cameras
23:58
to view the particle movement from different angles, which
24:01
allows you to measure that crucial third velocity component
24:03
that's normal to the light sheet and minimize that parallax
24:06
error. It's much more complex, often
24:08
requiring high pulse rate lasers to capture time resolved
24:11
measurements that show the flow changing second by second.
24:14
So moving from measuring existing flow to predicting new
24:17
flow, computational fluid dynamics, or CFD, is how we
24:21
design these complex impellers and vessels before cutting any
24:24
steel. But modeling turbulence requires
24:26
huge computational effort. Turbulence is messy and
24:29
inherently transient, so we have to use models to approximate its
24:32
effects. The workhorse models are the
24:35
Reynolds average, Navier, Stokes or RANS models.
24:38
These are time averaged, which means they simplify the chaotic
24:41
nature of the flow. And for many applications, RANS
24:44
models are sufficient. Right, they are, particularly
24:46
for general flow field prediction.
24:48
Certain RANS variance, like the talk bar model, offer
24:52
surprisingly good accuracy for specific complex applications
24:55
like predicting flow and challenging unbaffled stirred
24:59
vessels. They offer a good balance of
25:01
speed and acceptable accuracy. But if we need to capture the
25:04
genuine time dependent fluctuations, the truly
25:07
transient flow behavior that dictates things like micro and
25:11
meso mixing RANS fails us. That's where you have to step up
25:14
to large. Eddie simulation or LESLES is
25:17
fundamentally A transient model. It filters out the smallest
25:20
eddies, modeling their effect with a sub grid scale model, but
25:24
it explicitly captures the random fluctuations of the
25:27
medium and large eddies. So why would a designer choose
25:29
LES? Yes, even with the massive
25:32
computational cost. Because it's necessary for
25:34
capturing unstable phenomena that are critical to
25:37
performance. For instance, simulating the
25:39
periodic shedding of hairpin vortices behind the tabs in a
25:42
static mixer. Those vortices are the key
25:45
mixing mechanism, and a steady state RANS model would
25:48
completely fail to even see them.
25:51
It's the difference between a working process and an
25:53
underperforming 1. Speaking of trade-offs, when
25:56
you're modeling the impeller itself, there are different
25:58
levels of geometric rigor, trading speed for accuracy.
26:02
The simplest, fastest approach is the velocity data approach.
26:06
This is a steady state technique where you don't even model the
26:09
impeller, you just use experimental data to prescribe
26:12
the time average velocities in the impeller outflow region as a
26:15
boundary condition, and the CFD solver calculates the flow in
26:19
the rest of the vessel. Fast but not very detailed.
26:21
Right. If you want to see the intricate
26:23
details of the flow interaction between a blade and a baffle,
26:26
you need a full geometric model. The most common technique here
26:30
is the Multiple Reference frame, or MRF model.
26:33
It's still a steady state approximation, but it defines
26:37
the impeller region as rotating relative to the stationary tank,
26:40
offering a good compromise between fidelity and speed.
26:44
And the gold standard, the most rigorous approach.
26:47
That would be the sliding mesh monitor.
26:49
This is a genuinely time dependent method.
26:52
The simulation tracks the actual movement of the impeller blades
26:55
through the fluid grid, accurately capturing the time
26:58
varying interaction between the rotating impeller and stationary
27:01
elements like backles or internal coils.
27:03
So if you're tracking a transient process, like the time
27:06
it takes to lift the solid particle off the vessel floor,
27:09
or the flow of gas from a sparger, sliding mesh is
27:12
basically required. Exactly.
27:14
It's computationally expensive, but it gives you the most
27:16
accurate predictive power for those transient or high fidelity
27:20
problems. Let's apply this diagnostic and
27:22
modeling knowledge to some specialized industrial mixing
27:25
problems, starting with fermentation.
27:28
Bioreactors are primarily challenged by oxygen transfer.
27:32
Right In bioreactors, whether you're growing yeast, bacteria,
27:35
or mammalian cells, the goal is almost always maximizing the gas
27:39
liquid mass transfer coefficient KLA.
27:42
This is the rate at which oxygen moves from the gas bubble into
27:45
the liquid phase where the cells can actually use it.
27:48
And we have correlations that link KLA to the physical power
27:52
inputs. We do for standard aqueous
27:54
media. KLA is correlated to the power
27:56
dissipated per volume PG over V and the superficial gas velocity
28:00
VS. The relationship is typically
28:02
expressed as KLA being proportional to PG over V raised
28:05
to the power of 0.4 times VS raised to the power of 0.5.
28:10
That correlation tells us that both the sheer power input and
28:13
the gas flow rate are important, but that ratio shifts
28:16
significantly in high viscosity systems like those mycelial
28:19
fermentations we talked about earlier.
28:21
Oh, in high viscosity media standard radial flow impellers
28:25
like the classic Russian turbine, they fail
28:27
spectacularly. They just have poor circulation
28:30
capacity outside the immediate impeller zone.
28:33
They create that centralized cavern we discussed.
28:35
So what's the engineering solution to overcome the fluid's
28:39
yield stress and get circulation going throughout the entire
28:42
tank? You switch to axial flow
28:44
hydrofoils like like a procum or a pitched blade turbine.
28:47
These are specifically designed to push liquid axially either up
28:51
or down, generating bulk circulation loops that can
28:53
penetrate that high viscosity bulk fluid and get things
28:56
moving. And the purpose of increasing
28:58
that pumping capacity isn't just to make the fluid move, it's to
29:02
enhance the overall oxygen uptake efficiency.
29:05
Absolutely. By improving the bulk
29:07
circulation, the axial impeller increases the size of that well
29:10
mixed low viscosity cavern around the impeller.
29:13
And researchers have shown that the critical factor determining
29:15
overall oxygen uptake, and crucially, cell survival, is the
29:20
frequency at which cells are circulated through that highly
29:22
oxygenated high shear cavern region and then back out into
29:25
the bulk. This concept of cell circulation
29:28
time distribution, the CTD is so important that engineers use
29:32
specific scaled down modeling techniques just to simulate
29:36
these inhomogeneities. They do.
29:38
In giant bioreactors, maybe 22 cubic meters in size, you get
29:42
significant dissolved oxygen gradients.
29:45
You can model this in the lab using a system of two continuous
29:48
stirred tank reactors and STR plus STR system.
29:52
One STR simulates the highly oxygenated impeller zone and the
29:56
other simulates the poorly oxygenated bulk zone.
29:59
So you're essentially forcing the cells to alternate between
30:02
good times and bad times, just as they would circulating in a
30:04
massive tank. And this modeling technique has
30:07
been validated by some really striking industrial results.
30:10
One study involving large scale E coli cultivation using
30:13
standard Russian turbines, the massive glucose concentration
30:17
gradients up to 57 milligrams per liter at the top versus only
30:21
27 at the bottom. The cells were starving at the
30:23
bottom of the tank and. What was the fix?
30:25
They switched the impeller type to Skaba impellers, which
30:28
provide a combined radial and axial flow pattern.
30:32
This single change, focusing on improving the circulation time
30:35
distribution rather than just total power input, largely
30:39
eliminated those glucose concentration gradients.
30:42
Which proves that the flow pattern dictates homogeneity and
30:45
ultimately the cells access to critical nutrients far more than
30:49
just raw horsepower. It's a powerful lesson, a
30:52
perfect example of using fluid dynamics to solve a biological
30:55
problem. That is a fantastic example.
30:58
Let's pivot now to industrial pulp bleaching, another system
31:02
completely dominated by non Newtonian flocculated behavior.
31:06
Pulp processing, where you're trying to separate cellulose
31:09
from lignin using chemicals like chlorine dioxide, is critically
31:12
dependent on mass transfer. Even at only 10% mass
31:15
concentration, that suspension yield stress has to be overcome
31:19
to ensure the bleaching chemicals can reach the fiber
31:21
walls. Here the kinetics are fast and
31:23
the delivery of the chemical is limited by the physical
31:26
dispersion of those fiber flocks.
31:28
And this leads us to a famous case study that perfectly
31:31
illustrates that newer, higher power technology doesn't always
31:35
guarantee better results if you ignore the fundamental flow
31:38
physics. This mill upgraded their old PEG
31:42
mixer to a modern high shear dynamic mixer expecting big
31:46
efficiency gains. The complete opposite chemical
31:50
use actually increased unexpectedly by 2 kilograms of
31:54
chlorine dioxide per ton of pulp.
31:57
They checked the machine. The shear was excellent, but the
31:59
yield was terrible. Thermal imaging eventually
32:02
revealed the problem. The cold chemical stream was
32:04
channeling, it wasn't mixing in. Exactly because the pulp stream
32:07
was non Newtonian and behaved like a rigid plug.
32:10
The chlorine dioxide stream, which was much colder, was just
32:13
channeling along the side of the pulp plug.
32:15
The powerful new mixer was just pushing this unmixed plug along
32:18
the pipe. The mixer never got a chance to
32:20
do its job because the reagents were poorly introduced upstream.
32:24
So the solution wasn't adjusting the new mixer, it was addressing
32:27
the method of addition. They solved it by changing the
32:30
addition method ahead of the mixer, ensuring the chemical was
32:33
more widely dispersed into the entering pulp stream before it
32:36
even saw the mixer. This process fix ultimately
32:39
resulted in a massive saving of 1.5 kilograms of chlorine
32:43
dioxide per ton over the original installation.
32:46
The lesson being always address your upstream mixing conditions,
32:51
especially with these high yield stress fluids.
32:53
It's a simple lesson, but an expensive one to learn the hard
32:56
way. OK, we have talked about the
32:58
science of mixing kinetics and flow in exquisite detail, but
33:02
all of that process engineering is completely irrelevant if the
33:06
physical container fails. Yeah.
33:08
Now we have to shift our focus entirely to the mechanical and
33:11
regulatory backbone that ensures operational safety, the design
33:15
for integrity. This is the critical transition.
33:17
We're moving from microscopic turbulence to macroscopic forces
33:20
and regulatory requirements, and the mechanical integrity of
33:24
large pressure vessels is governed by incredibly stringent
33:26
codes like the ASME Boiler and Pressure Vessel Code, Section 8,
33:30
Division One. What's the general philosophy
33:32
behind that code structure? Does every single vessel need
33:36
bespoke detailed finite element analysis?
33:40
No, not necessarily. Under Division One, the code is
33:43
structured with very high built in safety factors and it
33:46
provides established proven rules for standard construction
33:49
details. Things like nozzles, flanges,
33:52
and heads. Because of these conservative
33:54
safety factors, extensive detailed stress analysis is
33:58
often waved. But there's a catch, I assume.
34:00
The catch is the engineer must document that all potential
34:03
loadings, internal pressure, seismic loads, wind loads,
34:06
support loads and thermal loads have been thoroughly considered
34:09
in the design. You can't ignore anything.
34:11
I think the most fundamental distinction for any pressure
34:14
vessel engineer is understanding the difference between primary
34:17
stress and secondary stress. If you miss this, you miss the
34:21
core of pressure vessel design. Absolutely.
34:23
Primary stresses, which we label P, are caused by unrelenting
34:27
loads. Think of the pressure pushing
34:29
against the vessel wall or the dead weight of the fluid inside.
34:32
If a primary stress exceeds the materials yield strength, the
34:35
resulting deformation is not self limiting.
34:38
Meaning the vessel will just continue to deform until
34:41
catastrophic failure it ruptures.
34:43
Precisely, they are the ultimate safety measure.
34:45
Therefore, primary membrane stress PM must be kept extremely
34:50
always less than or equal to the allowable stress SE set by the
34:54
code for that material and temperature.
34:56
OK, now contrast that with secondary stresses labeled Q.
35:00
Secondary stresses are caused by loads that are self limiting or
35:03
relenting. The classic examples are thermal
35:06
stresses, or what we call discontinuity stresses, that
35:08
arise at a knuckle radius or where a nozzle joins the shell.
35:12
If the material yields under a secondary stress, the resulting
35:15
small amount of deformation relaxes the stress and the
35:18
process stops. The stress limits itself.
35:21
They're less dangerous, so they're allowed to be higher.
35:23
Yes, because they won't cause immediate catastrophic rupture
35:27
in a ductile material. The codes allow for a much
35:29
higher combined stress limit. The combined primary plus
35:32
secondary stresses PL. plus Q are often allowed to go up to
35:36
three times the allowable stress SE, reflecting that yielding in
35:39
these localized areas is acceptable provided it doesn't
35:42
lead to excessive ratcheting or fatigue over time.
35:45
Beyond just gross rupture from high primary stress, what are
35:50
some of the other major failure modes that engineers have to
35:53
actively guard against in their designs?
35:56
A critical one is corrosion fatigue.
35:58
This happens when two different damage mechanisms, corrosion and
36:01
fatigue from cyclic stress, happen at the same time.
36:04
So they combine to make things worse.
36:06
It's a synergistic failure. The corrosion process creates
36:09
tiny pits on the metal surface which act as perfect stress
36:13
risers. These pits then become the
36:16
initiation point for fatigue cracks to propagate under cyclic
36:19
loading, which can drastically reduce the components expected
36:22
fatigue life. Another major concern is thermal
36:24
stress. Thermal stress is classified as
36:27
a secondary stress because it arises when thermal expansion or
36:30
contraction is mechanically prevented.
36:32
This usually happens because of steep linear temperature
36:34
gradients across the vessel wall, say when you rapidly
36:37
quench a process. And since it's a secondary
36:40
stress, we know that yielding can relax it, but it still leads
36:43
to risk. Right.
36:45
While it won't cause immediate rupture in most ductile
36:48
materials, repeated cycling of high thermal stress can lead to
36:51
failure through low cycle fatigue or eventually cause
36:54
unacceptable levels of permanent deformation that might
36:57
compromise the vessels alignment or its ceiling effectiveness.
37:01
Let's look specifically at the mixers mechanical backbone,
37:05
starting with the rotating components like the motor in the
37:07
drive. Motor selection typically
37:09
defaults to a NEMA Electrical Design Type B, which provides
37:13
normal starting torque and is suitable for most mixer
37:15
applications. However, if the process requires
37:18
variable speed, say for starting up a highly viscous shear
37:22
thinning batch, that variable speed motor must be specifically
37:26
rated for operation across the entire intended frequency range.
37:29
Especially when operating in explosion proof environments.
37:32
Absolutely, you need to ensure safe and efficient power
37:35
delivery at all speeds. Then we need to reduce that high
37:38
motor down to the required impeller speed, which means gear
37:40
reducers. Gear reducers are essential and
37:43
their efficiency varies widely. High efficiency helical and
37:46
spiral bevel gears lose only about 1 to 2% of power per
37:50
reduction stage, but if you opt for cheaper or higher ratio worm
37:54
gears, your power losses can jump to four to 10% depending on
37:58
the ratio, which adds up fast and energy costs over the life
38:01
of the machine. And what about the rolling
38:03
elements that facilitate the rotation itself?
38:06
The bearings. Bearings are the contact point
38:08
between the stationary housing and the rotating shaft.
38:11
They have to withstand all the radial thrust and overturning
38:14
loads transmitted by the impeller, which are often
38:17
unpredictable dynamic loads. They consist of inner and outer
38:20
rings and rotation of tours entirely on the rolling
38:23
elements, balls or rollers which reduce friction drastically.
38:26
Now to the most critical structural component, the shaft
38:29
itself. When designing a shaft, is the
38:32
minimum diameter based on pure strength, shear and tensile
38:35
stress limits? While the strength calculations
38:37
are mandatory for a mixer, the critical design criterion is
38:40
almost always shaft stiffness. You must select a shaft diameter
38:44
that is large enough to meet the natural frequency criterion.
38:47
Why does frequency trump strength?
38:50
Because the shaft has to operate safely away from its critical
38:53
speed, if the operating speed gets too close to the shafts,
38:56
natural frequency vibration amplification occurs, leading to
39:00
massive dynamic loads that can destroy bearings, seals, and
39:03
even the vessel itself. Even a shaft that's strong
39:07
enough to handle static forces will rapidly fatigue and fail if
39:10
it operates near its natural frequency.
39:12
That dynamic load concern is what keeps designers up at
39:15
night. Let's move to sealing, which
39:18
keeps the process fluid where it belongs.
39:21
What are the key industrial choices for process mixing?
39:24
For high pressure, hazardous or high value fluids, the industry
39:27
standard is the mechanical seal, often specified as a double
39:31
cartridge seal. These are complex devices using
39:34
specialized face materials like carbon, tungsten carbide or
39:37
silicon carbide, and they're pressurized with a clean barrier
39:40
fluid to ensure 0 leakage to the environment.
39:43
And why the cartridge design specifically?
39:45
Cartridge seals are preferred because the entire assembly, the
39:48
faces, springs and barrier connections comes as a single
39:52
preset unit. This allows for easy removal and
39:55
replacement during maintenance, and critically, it lets you
39:58
pretest the seal under pressure on a workbench before it's
40:01
installed in the reactor, which greatly minimizes installation
40:04
errors. And on the complete opposite end
40:06
of the complexity scale, what's the simplest seal?
40:09
The lip seal. It's basically just an elastomer
40:12
material pressed against the shaft by a small spring.
40:15
Lip seals cannot hold any appreciable pressure.
40:18
They're used primarily in atmospheric tanks, or sometimes
40:21
just to keep dust and dirt out of the bearing housing.
40:24
They're cheap, but functionally very limited.
40:27
Finally, let's talk mounting structures, the beams and
40:30
structural steel that hold the mixer drive.
40:32
This is a frequently underestimated area of
40:35
mechanical design. The mounting supports must be
40:37
designed specifically for the dynamic loads created by the
40:40
mixer, not just it's static weight.
40:42
The forces generated by an unbalanced impeller flow
40:44
stabilities, or operating your critical speed are much, much
40:48
greater than the static weight of the gear drive.
40:50
So if an engineer only calculated for the static
40:52
weight, the resulting support beams would be far too
40:55
lightweight and would be prone to catastrophic vibration
40:58
failure. Precisely The dynamic nature of
41:00
mixer loads requires structural stiffening and moment
41:03
calculations that account for these flow induced forces.
41:07
It's a serious concern that is often missed by engineers who
41:10
are unfamiliar with agitated systems.
41:12
Let's tie this mechanical rigor back to the broad requirement
41:16
for regulatory compliance and manufacturing quality control.
41:19
For a pressure vessel to operate anywhere in the developed world,
41:22
it has to pass legislative muster.
41:25
The regulatory environment is extremely strict.
41:28
In the European Union, the design and manufacturer are
41:30
governed by the Pressure Equipment Directive or PED.
41:34
Other regions have similar national standards.
41:37
Compliance is mandatory, and it leads to critical certifications
41:41
like the CE Mark, which indicates the vessel meets
41:43
essential safety requirements. And how does the manufacturing
41:47
process ensure this quality is actually met?
41:50
It starts with the most critical fabrication step, welding.
41:53
Welding requires mandatory quality checks throughout
41:55
fabrication. You cannot just trust the welder
41:58
skill. Welding Procedure
42:00
Specifications, or WPS and their associated procedure
42:03
Qualification records or PQRS are required for all important
42:08
seams. These documents detail exactly
42:10
how the Weld must be performed for every method used.
42:13
MMA, mig, submerged arc or tig and.
42:16
Once the Weld is complete we move to non destructive testing
42:19
or NDT which verifies the structural integrity without
42:23
damaging the part. NDT ensures that any defect size
42:26
and the total area of defects are within code acceptable
42:28
limits. Two key methods are used for
42:31
checking longitudinal and circumferential seams and
42:33
nozzles, radiographic testing or RT, and ultrasonic testing or
42:38
UT. What's?
42:39
The practical difference between those two?
42:40
RT or radiography uses X-rays to generate a film image.
42:45
It's excellent at detecting volumetric defects like porosity
42:48
or slag inclusions. UT, or ultrasonic testing, uses
42:53
sound waves, and it's generally better for detecting planar
42:56
defects like cracks. UT is also superior for
43:00
accurately sizing the depth of a defect.
43:02
You often use both methods depending on the Weld geometry
43:05
and the type of defect you most expect to find.
43:08
After all the fabrication and NDT checks, the final mechanical
43:11
proof is the pressure test. Nearly all vessels operating
43:15
above atmospheric pressure are pressure tested.
43:17
The goal is to verify the resistance against static
43:19
stresses and strains under controlled conditions.
43:22
This test has to be documented. While the factor varies by
43:26
regulation, some highly specialized vessels might be
43:28
tested at pressures up to five times the maximum working
43:31
pressure. The most standard pressure
43:33
vessels use a smaller margin, like 1.3 times the maximum
43:36
pressure. And finally, the moment the
43:37
vessel leaves the shop, handling and transportation, you can
43:40
spend a million engineering hours designing the perfect
43:43
vessel, only to have it damaged before it's even installed.
43:46
This is a logistics problem that has to be solved by the design
43:49
engineer, not the shipping company.
43:51
Shipping saddles, the supports the vessel rests on during
43:54
transit, must be designed along with the vessel.
43:57
They have to be located precisely adjacent to major
43:59
stiffening elements like reinforcing rings or the dished
44:02
heads to ensure the load is distributed safely.
44:05
And during the erection process on site, there are local loads
44:09
to consider as well. Absolutely.
44:10
When you're using lifting trunions, those temporary
44:13
lifting lugs welded to the shell, the local loads they
44:16
impart on the vessel shell must be calculated meticulously.
44:20
If those local stresses are too high, they can cause buckling or
44:23
plastic deformation before the vessels even lifted into place
44:26
and operational. We have completed an enormous
44:29
journey today. We moved from optimizing the
44:31
local energy dissipation rate in a tiny xanthine fermentation
44:34
sample all the way to calculating the maximum stress
44:37
on from a high pressure nozzle under seismic loads.
44:40
It all feeds back into one fundamental design objective,
44:45
creating an industrial process that is efficient because it is
44:48
inherently safe and robustly engineered.
44:50
Indeed, we have seen that in many advanced chemical and
44:53
biological processes, especially these fast reactions or cell
44:57
cultures, the critical performance metric, whether it's
45:00
yield or cell viability, is less about the average velocity of
45:04
the fluid and much more about the instantaneous local
45:07
turbulent energy dissipation rate epsilon.
45:11
This raises a critical and I think provocative question for
45:14
the future of industrial design. Go on, In a world of increasing
45:18
process complexity, how much longer will macroscopic time
45:21
average? And models remain relevant when
45:24
the critical physics that govern failure and efficiency happen at
45:26
time scales far, far faster than we can typically measure or
45:30
model outside of highly specialized, computationally
45:33
intensive research settings. That's it for today, Design
45:36
junkies. If you're getting something out
45:37
of this, follow the podcast and leave a review.
45:39
It helps the whole family grow, learn and build Better Together.