Latest / Mechanical Engineering Made Simple / (#127) Defeating Resonance and Structural Shock

Transcript

(#127) Defeating Resonance and Structural Shock · 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
• 0:08
  below. Glad to be here for this one.
• 0:10
  It's a it's a topic that keeps a lot of engineers up at night.
• 0:13
  Oh absolutely, I mean I want you to just imagine standing next to
• 0:17
  a like a massive 3 ton industrial power generation
• 0:22
  turbine. Right, just a massive piece of
• 0:24
  machinery. Yeah, and it's spinning at
• 0:26
  10,000 revolutions per minute. Sounds like a jet engine.
• 0:29
  The floor beneath your boots is humming and you know everything
• 0:33
  seems perfectly fine. But it's usually not.
• 0:36
  Exactly because hidden inside the mathematics of that machine,
• 0:40
  there's this microscopic miscalculation in the stiffness
• 0:43
  of a single, like a rubber mount, and it's causing a
• 0:47
  resonant frequency to align perfectly with the rotor spin.
• 0:50
  Which is pretty much a nightmare scenario.
• 0:52
  Total nightmare. You can't see it yet, but the
• 0:54
  amplitude of the vibration inside that steel housing is
• 0:57
  like doubling every few seconds. And in exactly 2 minutes, that
• 1:01
  three ton piece of precision engineering is going to
• 1:04
  violently tear itself off its mouths, shatter its casing, and
• 1:09
  just send hypersonic shrapnel through a concrete wall.
• 1:12
  I've actually seen the aftermath of failures like that, and it
• 1:15
  is. It's not pretty.
• 1:17
  So today for this deep dive, we are conquering the mechanics of
• 1:22
  shock and vibration to make sure you never design that turbine.
• 1:25
  That's the mission. We're going to break down the
• 1:27
  biggest pain points and real world issues engineers keep
• 1:30
  running into with this subject. We'll call out the common
• 1:33
  modeling and measurement misconceptions that ruin
• 1:36
  designs. And there are a lot of them.
• 1:38
  So many, and then we'll walk through a step by step approach
• 1:41
  to avoid these traps and answer the most common questions about
• 1:45
  optimizing designs in practice. Right, because it's all about
• 1:48
  giving you the tools to actually optimize a design, not just on
• 1:51
  paper, but, you know, in the messy physical world.
• 1:54
  Exactly. But I guess to optimize the
• 1:56
  design, you first have to know exactly what kind of dynamic
• 1:59
  enemy you're fighting. Yeah, you really can't defeat an
• 2:01
  enemy you don't properly classify.
• 2:03
  And in the realm of structural dynamics, that enemy usually
• 2:06
  takes one of two primary forms, vibration or shock.
• 2:10
  OK, so let's break those down. Well, while they share the same
• 2:14
  underlying physics of, you know, mass and stiffness, they are
• 2:18
  fundamentally different beasts in how they deliver destructive
• 2:22
  energy. How so?
• 2:23
  So vibration is an oscillation. It's this continuous alternating
• 2:28
  transfer of energy within a mechanical system.
• 2:31
  Like a pendulum or something. Yeah, exactly.
• 2:34
  If we strip it down to its absolute simplest form, just
• 2:37
  think of a solid steel mass hanging from a coiled spring.
• 2:40
  OK. Visualizing that.
• 2:41
  You pull that mask down and let it go.
• 2:44
  The energy shifts from the potential energy stored in the
• 2:47
  stretched metal of the spring into the kinetic energy of the
• 2:51
  heavy moving mass as it accelerates upward.
• 2:54
  Right, and then back into potential energy as the spring
• 2:56
  compresses at the top of the stroke.
• 2:58
  Exactly. It's a constant, ongoing
• 3:00
  conversation between kinetic and potential energy A.
• 3:02
  Conversation. I like that.
• 3:04
  And we typically model this continuous conversation using
• 3:07
  three elements. You've got a mass to store
• 3:09
  kinetic energy, spring to store potential energy, and a damper
• 3:13
  to represent the energy permanently lost the environment
• 3:16
  through friction or heat. So that's vibration, but shock
• 3:20
  is totally different. Completely different shock does
• 3:23
  not give the structure time for a conversation.
• 3:25
  Just a shout. Yeah, it's a sudden, non
• 3:28
  periodic and severe excitation. An impact, an explosive blast, a
• 3:34
  package dropping off a forklift, or a sudden step in velocity.
• 3:38
  So the event is completely over before the structure even has a
• 3:41
  chance to fully react. Right.
• 3:42
  The external force spikes and vanishes, leaving the structure
• 3:45
  to wildly deal with this massive instantaneous injection of
• 3:49
  residual energy. Wow.
• 3:51
  OK, so let's unpack this taxonomy of tremors.
• 3:53
  Let's focus on vibration first because, you know, even within
• 3:56
  that category there are subdivisions that constantly
• 3:59
  trip people up in the field. Oh yeah, we have deterministic
• 4:01
  vibration and random vibration. And then like layered on top of
• 4:05
  that, we have stationary versus non stationary environments.
• 4:08
  And the mathematical approach you take to save your design
• 4:11
  depends entirely on getting those initial categorizations
• 4:15
  right. So what's deterministic
• 4:17
  vibration? A deterministic vibration
• 4:19
  follows an established, mathematically predictable
• 4:22
  pattern. If you know the past history of
• 4:24
  the vibration, you can plug it into an equation and perfectly
• 4:27
  predict its exact amplitude and direction at any designated
• 4:31
  future time. Like whether that's 10 seconds
• 4:33
  from now or 10 years from now. Exactly.
• 4:35
  The classic example is the vibration radiating from a
• 4:38
  rotating motor, a reciprocating engine, or like a helicopter
• 4:43
  rotor. It's periodic.
• 4:44
  It repeats. Yeah, the math is clean, but
• 4:46
  Ranjo vibration is the exact opposite.
• 4:48
  Which is what a lot of people actually deal with.
• 4:50
  Yeah, it's entirely unpredictable and it's exact
• 4:53
  moment to moment instantaneous magnitude.
• 4:55
  You just cannot write a simple sine wave equation for it.
• 4:58
  So how do you define it? You can only define it in
• 5:01
  statistical terms. You use probability
• 5:03
  distributions to indicate the likelihood of certain
• 5:05
  acceleration magnitudes occurring.
• 5:07
  OK, so give me an example of that.
• 5:09
  Think about the environment. A rocket experiences breaking
• 5:12
  through the atmosphere at supersonic speeds.
• 5:14
  Oh yeah, that's got to be intense.
• 5:16
  The acoustic energy from the rocket clume and the turbulent
• 5:20
  aerodynamic boundary layer tearing over the metal skin.
• 5:23
  It creates this broadband chaotic rumble.
• 5:27
  You have no idea what it's doing at any exact millisecond.
• 5:30
  Right. You don't know exactly what the
• 5:32
  skin will be doing at millisecond #400 but you can
• 5:35
  predict the probability that the stress will exceed the yield
• 5:38
  limit of the titanium over a 10 second window.
• 5:42
  That statistical approach that leads right into the stationary
• 5:45
  versus non stationary distinction.
• 5:46
  Right, it does. A stationary vibration is 1
• 5:48
  where the fundamental statistical properties like the
• 5:51
  mean value, the overall variance, or the root mean
• 5:54
  square. They don't change over time.
• 5:56
  So the environment is chaotic, but it's constantly chaotic in
• 5:59
  the exact same way. Exactly.
• 6:01
  A non stationary vibration on the other hand means those
• 6:04
  underlying statistical properties are actively
• 6:06
  shifting. So sticking with a rocket.
• 6:08
  Yeah, imagine a launch vehicle accelerating from Mach 1 to Mach
• 6:11
  5. The dynamic pressure on the
• 6:13
  vehicle is constantly changing. The vibration profile at minute
• 6:17
  1 is entirely different from the vibration profile at minute 3.
• 6:21
  Precisely. You know, from what we see in
• 6:22
  failure analysis reports, there is a massive misconception here
• 6:26
  that completely screws up designs right out of the gate.
• 6:30
  Oh, I know exactly what you're going to say.
• 6:31
  Because the math for a predictable sine wave is
• 6:35
  something we all learn in like high school physics, engineers
• 6:38
  have this incredibly dangerous tendency to treat all vibration
• 6:42
  as if it is stationary and deterministic.
• 6:44
  It's the trap of mathematical convenience.
• 6:46
  Yeah, they design their brackets and mounts to survive these
• 6:49
  clean, predictable, beautiful 50 Hertz sine waves, but the
• 6:54
  physical reality of the environment is actually throwing
• 6:56
  broadband chaotic, non stationary random vibration at
• 7:01
  them. We're taught harmonic motion on
• 7:03
  day one of physics. You know, simple sine and cosine
• 7:06
  waves. So when an engineer looks at a
• 7:08
  complex operational environment, the psychological temptation is
• 7:11
  to approximate it as a combination of steady sine
• 7:14
  waves. Because it makes the spreadsheet
• 7:16
  look cleaner. Yeah, it does.
• 7:17
  But designing an aerospace component to survive a
• 7:20
  continuous single frequency 50 Hertz sine wave is completely
• 7:24
  different from designing it to survive a random broadband
• 7:27
  rumble that hits every frequency between 10 and 2000 Hertz
• 7:31
  simultaneously. And fluctuates in overall
• 7:33
  intensity. Right.
• 7:35
  If you treat random vibration as deterministic, your fatigue
• 7:39
  cycle calculations will be wildly inaccurate.
• 7:42
  You'll severely underestimate the cumulative damage happening
• 7:45
  at the microscopic level of the metal.
• 7:47
  And your part will fracture in the field long before its design
• 7:50
  lifespan. It's crazy.
• 7:51
  I picture deterministic stationary vibration like a
• 7:55
  metronome sitting on a piano. TikTok TikTok.
• 7:59
  It's never changes pace and you know exactly where the pendulum
• 8:02
  arm will be 10 minutes from now. That's a great way to look at
• 8:05
  it. But non stationary random
• 8:06
  vibration is like taking a shopping cart with a wobbly
• 8:09
  wheel, filling it with bowling balls and driving it over a
• 8:12
  cobblestone St. while simultaneously accelerating and
• 8:15
  occasionally hitting a pothole. It is absolute chaos, right?
• 8:20
  But if random vibration is totally unpredictable and
• 8:23
  changes moment to moment like that shopping cart, how do we
• 8:26
  even begin to mathematically model it without just blindly
• 8:30
  guessing and hoping the metal holds well?
• 8:32
  We have to abandon the quest for absolute certainty and embrace
• 8:35
  the mathematics of probability. OK, we stop asking the
• 8:39
  impossible question. Like what will the exact
• 8:42
  displacement of this bracket be at exactly 3.2 seconds?
• 8:46
  Because you just can't know. You can't.
• 8:49
  Instead, we ask what is the probability that the
• 8:51
  displacement will exceed 2mm during this entire 5 minute
• 8:55
  flight window? And how do we do that?
• 8:57
  To do this, we rely heavily on something called the ergodic
• 9:00
  hypothesis. The.
• 9:01
  Ergodic hypothesis. Yeah, it's a fundamental concept
• 9:04
  in statistical mechanics. It states that for a stationary
• 9:07
  random process, taking time averages from a single very long
• 9:11
  sample record will give you the exact same statistical
• 9:14
  properties as taking an ensemble average across multiple
• 9:17
  different sample records at a single instant in time.
• 9:20
  Wait, let's freeze right there. The ergotic hypothesis.
• 9:22
  That sounds like something out of a theoretical physics
• 9:25
  textbook. What does that actually mean for
• 9:28
  an engineer trying to measure vibration on a physical
• 9:32
  prototype? Think of it like a casino.
• 9:34
  OK, I'm with you. Suppose you want to know the
• 9:37
  statistical probability of a coin landing on heads.
• 9:40
  You have two ways to figure this out.
• 9:42
  The first way is to gather an ensemble.
• 9:45
  You get 1,000,000 people, give them each a coin, and have them
• 9:48
  all flip it at the exact same moment.
• 9:51
  You take a snapshot of that single millisecond, count the
• 9:54
  heads, and get your probability. That's an ensemble average.
• 9:57
  And the second. Way.
• 9:58
  The second way is to take just one person, give them one coin,
• 10:02
  and have them flip it 1,000,000 times over the course of a year.
• 10:06
  You average those results over time.
• 10:07
  And the hypothesis says. The ergodic hypothesis says that
• 10:11
  for a stationary random environment, those two methods
• 10:14
  yield the exact same statistical answer.
• 10:17
  Oh I see. In engineering terms, it means
• 10:19
  we don't have to launch 10,000 rockets simultaneously to
• 10:22
  understand the vibration environment.
• 10:24
  That would be expensive. Very We can just put a sensor on
• 10:28
  one rocket, record the data over a sufficiently long period of
• 10:31
  time, and use that single time history to confidently predict
• 10:35
  the statistical probability of failure for every future rocket.
• 10:39
  That is incredibly practical, but to even begin applying those
• 10:42
  advanced statistics, we have to build mathematical models of the
• 10:46
  physical structures themselves to see how they react to that
• 10:49
  input. Exactly.
• 10:50
  Which takes us out of the environment and directly into
• 10:52
  the hardware. We have to map the unpredictable
• 10:55
  input to a mathematical model of our system, but building those
• 10:58
  models is where theoretical traps are just everywhere.
• 11:01
  Oh, they really are. And the foundation of all
• 11:03
  structural dynamics begins by defining the degrees of freedom.
• 11:07
  Right SDOF and MDOF. Yes, a single degree of freedom
• 11:10
  system, or SDOF is the simplest possible mathematical
• 11:14
  abstraction. It means you only need one
• 11:16
  single independent coordinate to completely define the
• 11:19
  geometrical position of the entire system at any given
• 11:22
  instant in time. Like a rigid block constrained
• 11:25
  to move only perfectly up and down on a single coil spring.
• 11:28
  Exactly. That's an SDOF model.
• 11:31
  You only need to know it's vertical your coordinate.
• 11:34
  But a physical structure like the wing of an airplane.
• 11:36
  It's a continuous system. It's made of billions of
• 11:39
  molecules, meaning it has an infinite number of degrees of
• 11:42
  freedom. It can bend, twist, sheer and
• 11:46
  warp in countless directions simultaneously.
• 11:48
  Which is impossible to calculate.
• 11:50
  Because the math for infinite degrees of freedom is unsolvable
• 11:53
  and and practical engineering we discretize the continuous
• 11:57
  structure. We chop it up into a multiple
• 11:59
  degree of freedom model or an MDOF system using a finite
• 12:03
  number of lumped masses, springs and dampers.
• 12:06
  What fascinates me is how universal the physics are once
• 12:09
  you break it down into those lumped elements.
• 12:11
  Like the mass translates Florida State between something moving
• 12:14
  in a straight line and something spinning in a circle.
• 12:18
  There are direct analogies here. Yeah, the equations of motion
• 12:20
  are structurally identical, only the variables wear different
• 12:23
  clothes. I love that phrasing.
• 12:24
  In translational linear motion, the core elements are force,
• 12:27
  mass, and linear displacement. If you transition to rotational
• 12:31
  motion, say analyzing the vibration of a ship's propeller
• 12:35
  shaft, the mathematical analog to linear force is torque.
• 12:39
  OK, torque makes sense. And mass.
• 12:41
  The analog to physical mass is the mass moment of inertia,
• 12:45
  which accounts for how that mass is distributed around the axis
• 12:48
  of rotation, and the analog to linear displacement is angular
• 12:52
  displacement, usually measured in radians.
• 12:54
  If you deeply understand the differential equations governing
• 12:58
  how a simple block on a spring oscillates up and down, you
• 13:01
  inherently understand the mathematics of how a massive
• 13:04
  flywheel on a twisting steel shaft oscillates back and forth.
• 13:08
  So whether we are talking about a linear spring or a twisting
• 13:11
  shaft, what does this actually mean for calculating the
• 13:14
  baseline pulse of the system? I'm talking about the natural
• 13:17
  frequency. The UN damped natural frequency.
• 13:20
  It is the single most critical calculation in all of mechanical
• 13:23
  vibration analysis, period. It's the frequency at which the
• 13:26
  system naturally wants to oscillate if you strike it once
• 13:29
  and let it ring, assuming absolutely no damping or energy
• 13:32
  loss in the universe. And for our listeners, following
• 13:36
  along with the physics, the math here is foundational Omega n =
• 13:40
  sqrt K / m. Exactly that is the angular
• 13:43
  natural frequency Omega. And it is determined entirely by
• 13:47
  the ratio of the system's spring stiffness K to its mass M.
• 13:52
  Yep. Just K&M.
• 13:54
  Let's zoom in right here, because this exact equation,
• 13:57
  Omega n = sqrt K / m is where one of the most insidious
• 14:02
  multimillion dollar traps in mechanical engineering is hiding
• 14:05
  in plain sight. It's the static versus dynamic
• 14:08
  trap. It's the dangerous assumption
• 14:10
  that static deflection always equals dynamic stiffness.
• 14:13
  This is a classic catastrophic error that engineers make
• 14:16
  constantly, particularly when working with elastomers,
• 14:19
  isolators, or rubber mounts. Walk us through how this
• 14:22
  happens. Let's see, an engineer is
• 14:24
  designing a suspension system for a delicate piece of
• 14:26
  electronics inside a military vehicle.
• 14:30
  They need to know the stiffness K of a rubber isolation mount so
• 14:34
  they can plug it into that equation and calculate the
• 14:36
  natural frequency of their system.
• 14:38
  Right, so they need the K value. So what do they do?
• 14:40
  They take the rubber mount into the lab, place a heady static
• 14:43
  weight on it, Measure with a ruler how many millimeters the
• 14:46
  rubber compresses, and divide the force of the weight by that
• 14:50
  displacement to get K. And then they just plug that
• 14:53
  static stiffness into the dynamic equation.
• 14:55
  And by doing that, they have just compromised the entire
• 14:58
  design. Completely compromised.
• 14:59
  Yeah, rubber, synthetic elastomers and many other
• 15:02
  complex materials do not behave the same way under a rapidly
• 15:05
  oscillating dynamic load as they do under a slow static resting
• 15:09
  load. Why is that?
• 15:10
  At a molecular level, rubber is made of long tangled polymer
• 15:14
  chains. When you push on them slowly,
• 15:16
  those chains have time to slide past each other and reorient.
• 15:19
  That gives you a low static stiffness.
• 15:21
  OK, that makes sense. But when you strike them rapidly
• 15:24
  with a dynamic vibration, those long columnar chains do not have
• 15:28
  the time to disentangle. They lock up.
• 15:31
  Oh wow. They resist the rapid motion
• 15:33
  with significantly more force. This means the dynamic stiffness
• 15:37
  of a rubber mount can be vastly different, often an order of
• 15:41
  magnitude higher than its static stiffness.
• 15:43
  An order of magnitude. Easily, if you use the static
• 15:46
  stiffness from your lab ruler measurement in your dynamic
• 15:49
  calculation, your predicted natural frequency will be
• 15:52
  entirely wrong. You will design the system
• 15:55
  thinking your electronics are perfectly isolated from the low
• 15:58
  frequency rumble of the vehicle's engine.
• 16:00
  But in reality, your actual dynamic natural frequency is
• 16:04
  much higher, and it might be perfectly aligned with the
• 16:06
  destructive forces of the environment.
• 16:08
  The analogy I use for this trap is testing a mattress.
• 16:12
  Oh, let's hear it. Imagine you want to test how
• 16:14
  stiff a memory foam mattress is. If you gently sit on the edge of
• 16:18
  the bed, the phone sinks in slowly.
• 16:21
  The polymer chains have time to move.
• 16:22
  Right, low static stiffness. You measure that deep sinkhole,
• 16:26
  calculate your K value, and you think, OK, this is a very soft
• 16:29
  mattress. That is your static stiffness.
• 16:32
  But if you climb up to the top rope of a wrestling ring and
• 16:35
  drop an elbow onto that exact same mattress?
• 16:38
  It's not going to end well. Not at all.
• 16:40
  The material reacts completely differently to that sudden
• 16:42
  dynamic impact. The air can't escape the foam
• 16:45
  fast enough, the polymers lock up and the mattress pushes back
• 16:49
  with the rigidity of a wooden board.
• 16:50
  That's a great visual. If you calculated the safety of
• 16:53
  your jump based on gently sitting down, you are going to
• 16:56
  shatter your collarbone. Exactly.
• 16:58
  But I have to push back on this whole modeling philosophy.
• 17:01
  We keep talking about a mass and a spring.
• 17:04
  Why do we even rely on single degree of freedom models with
• 17:08
  basic isolated masses and springs when a real world
• 17:12
  machine like a naval submarine or commercial jetliner has
• 17:16
  millions of complex interconnected parts?
• 17:19
  It feels like we're trying to model a hurricane using a single
• 17:22
  drop of water. I get that it sounds totally
• 17:24
  reductive, but we use these highly simplified single degree
• 17:28
  of freedom models because they represent the fundamental
• 17:31
  independent building blocks of something called modal analysis.
• 17:35
  Modal analysis. Yeah, even an incredibly complex
• 17:38
  structure with millions of parts, like a submarine pressure
• 17:41
  hall, can be mathematically decoupled into a series of
• 17:45
  individual, completely independent single degree of
• 17:48
  freedom systems. Really just broken down into
• 17:51
  simple springs and masses. Exactly.
• 17:54
  Each one of those simplified systems represents one specific
• 17:57
  normal mode of vibration. The submarine has a mode where
• 18:00
  it bends like a banana. It has another mode where it
• 18:02
  twists like a towel. It has another mode where it
• 18:05
  breathes in and out like a lung. And we can model each of those
• 18:08
  independently. By understanding the simple mass
• 18:10
  spring math, we can perfectly model each of those individual
• 18:13
  modes. The total chaotic complex
• 18:16
  physical response of the entire submarine to an underwater
• 18:19
  explosion is mathematically just the superposition, the linear
• 18:23
  sum of all those simple models reacting simultaneously.
• 18:26
  That is wild. However, the stake are
• 18:29
  incredibly high. Very high.
• 18:31
  If you miscalculate the natural frequency of even one of those
• 18:34
  fundamental modes because you use static stiffness instead of
• 18:38
  dynamic stiffness, you leave a Georgia in your defenses, and
• 18:41
  that gap opens the door to the most destructive phenomenon in
• 18:45
  mechanical engineering. The real world Wrecking Ball
• 18:48
  resonance. Resonance.
• 18:50
  Once you miscalculate that natural frequency, resonance
• 18:53
  will find the flaw in your math and tear the metal apart.
• 18:56
  Resonance is the phenomenon that haunts structural engineers.
• 19:00
  It occurs in forced vibration when the frequency of the
• 19:04
  external dynamic force acting on your system, whether that is the
• 19:07
  wind pushing on a bridge or the spinning of a motor, exactly
• 19:10
  matches the internal natural frequency of your system.
• 19:13
  Something terrifying happens physically.
• 19:15
  The phase relationship aligns. Perfectly.
• 19:17
  The phase relationship between the input force and the
• 19:20
  structure's velocity aligns perfectly.
• 19:22
  This means that every single cycle, the external force is
• 19:25
  pushing the structure exactly when the structure already wants
• 19:28
  to move in that direction. The energy just compounds.
• 19:30
  The energy from the force builds up cycle after cycle,
• 19:34
  accumulating inside the system. The amplitude of the physical
• 19:37
  vibration grows continuously, limited only by the amount of
• 19:41
  damping or energy dissipation in the system.
• 19:44
  And if there isn't enough damping?
• 19:45
  If the damping is low, the structure will literally shake
• 19:48
  itself to pieces. The ultimate historical case
• 19:51
  study for this is the Tacoma Narrows Bridge, collapsed in
• 19:54
  1940, right? The footage of that bridge
• 19:57
  twisting like a wet noodle before snapping and falling into
• 20:00
  the river is burned into the brain of every first year
• 20:03
  engineering student. The Tacoma Narrows Bridge is the
• 20:06
  perfect visceral example of what happens when massive amounts of
• 20:09
  external energy align with the natural frequency of a large
• 20:13
  structure. Though it was slightly more
• 20:14
  complex than basic resonance, right?
• 20:16
  Technically, yes, the bridge collapse was a complex aero
• 20:20
  elastic flutter phenomenon driven by wind vortices, but the
• 20:23
  core underlying principle of destructive energy accumulation
• 20:27
  is identical to the classic resonance.
• 20:29
  The wind flowing over the bridge deck created alternating
• 20:31
  vortices of low pressure that pushed the bridge up and down.
• 20:34
  And as the bridge twisted, it changed the aerodynamics, which
• 20:38
  amplified the vortices, which amplified the twisting.
• 20:42
  The energy kept feeding into the torsional natural frequency of
• 20:45
  the bridge. Because the steel and concrete
• 20:47
  structure had very little inherent damping to dissipate
• 20:51
  that energy. Right, so the amplitude of the
• 20:53
  twisting grew until the structural yield limits of the
• 20:56
  suspension cables and the deck girders were exceeded.
• 20:59
  It is the Wrecking Ball that destroys bridges, shatters
• 21:02
  turbine blades and cracks building foundations during
• 21:05
  earthquakes. And we measure exactly how sharp
• 21:07
  or severe this resonance is using a mathematical parameter
• 21:11
  called the quality factor or Q. Yes, Q is the parameter that
• 21:15
  quantifies the frequency selectivity, or the sharpness of
• 21:18
  the resonance peak. It tells you how much the input
• 21:20
  vibration will be amplified when you hit that exact resonant
• 21:23
  frequency. And what's the math for Q?
• 21:26
  Mathematically, Q = 1 divided by the quantity of two times Zeta.
• 21:30
  Zeta being the damping ratio. Exactly.
• 21:32
  Zeta is the damping ratio, which is the fraction of critical
• 21:35
  damping being present in the system.
• 21:37
  Let's define critical damping real quick before we move on,
• 21:39
  because that is an important baseline.
• 21:41
  Critical damping is the exact specific amount of damping
• 21:45
  required to return a displaced system to its resting
• 21:48
  equilibrium position as fast as possible without any oscillation
• 21:52
  at all. So if you push a swinging door
• 21:54
  and it slowly closes and stops exactly at the center latch
• 21:57
  without swinging past. It that door is critically
• 22:00
  damped, but most structural systems like airplane wings or
• 22:03
  steel buildings are incredibly under damped.
• 22:06
  Their Zeta value is far less than one.
• 22:08
  So Zeta is a very small number. The denominator in our Q
• 22:12
  equation 2 times Zeta is also a very small number.
• 22:15
  And dividing one by a very small number means the Q factor
• 22:18
  becomes very large. And a large Q factor is bad
• 22:21
  news. Very bad.
• 22:22
  A large Q factor physically manifests as a massive towering
• 22:25
  amplification of the vibration precisely at lesnans.
• 22:29
  If you have AQ of 50, an input vibration of 1G of acceleration
• 22:33
  will be amplified into 50 GS of acceleration on your component.
• 22:37
  I have always found the nomenclature of the quality
• 22:39
  factor to be deeply, darkly ironic.
• 22:42
  Oh, completely. In everyday human life, if
• 22:44
  something has a high quality factor, you assume it's a
• 22:47
  beautifully made, durable product.
• 22:50
  But in structural dynamics, if your multi $1,000,000 satellite
• 22:53
  component has a high quality Q factor, it basically means it is
• 22:57
  a highly efficient energy accumulator with a very high
• 23:00
  probability of vibrating itself into space dust.
• 23:04
  Yeah, you really don't want a high Q.
• 23:06
  So to survive this Wrecking Ball, we have to optimize the
• 23:09
  design by adding damping. What are the physical mechanisms
• 23:13
  we can use? In practice, we generally
• 23:15
  classify energy dissipation into 3 distinct types of damping.
• 23:19
  First is viscous damping. Fluid friction.
• 23:21
  Right. Imagine the shock absorber on
• 23:23
  your car, which is essentially A piston pushing through a
• 23:25
  cylinder filled with thick oil. As the piston moves, the fluid
• 23:29
  is forced through tiny orifices, generating fluid shear forces.
• 23:33
  Turning kinetic energy into heat.
• 23:35
  Exactly. It converts the kinetic energy
• 23:37
  of the motion into thermal energy, heating up the oil.
• 23:41
  The defining mathematical characteristic of viscous
• 23:43
  damping is that the damping force is directly proportional
• 23:46
  to the velocity of the motion. So the faster you try to push
• 23:49
  the piston, the harder the fluid pushes back.
• 23:52
  The second type is Coulomb damping, which is also known as
• 23:54
  dry friction. This is completely different
• 23:57
  from a fluid. Yes.
• 23:58
  Coulomb damping is the sliding of 2 dry, unlubricated surfaces
• 24:02
  against each other. Imagine a heavy steel block
• 24:05
  sliding across a concrete floor. At a microscopic level.
• 24:09
  Both the steel and the concrete are not smooth.
• 24:12
  They are covered in tiny jagged peaks and valleys called
• 24:16
  asperities. As the surfaces slide, these
• 24:19
  microscopic peaks crash into each other, cold Weld together
• 24:23
  and then shear apart. And breaking those micro welds
• 24:26
  dissipates energy as heat. The defining characteristic here
• 24:30
  is that unlike viscous damping, the Coulomb damping force is
• 24:34
  constant. It is completely independent of
• 24:36
  how fast the block is sliding or how far it slides.
• 24:39
  It only depends on the weight and friction.
• 24:41
  Right. It only depends on the normal
• 24:42
  force, how heavy the block is, and the kinetic coefficient of
• 24:45
  friction between the two specific materials.
• 24:48
  And the third one, which seems to cause the most mathematical
• 24:51
  trouble is structural damping. This doesn't involve fluids or
• 24:55
  sliding parts. This comes from within the solid
• 24:58
  material itself. Structural damping, which is
• 25:00
  also called hysteretic damping, is the energy dissipated by
• 25:04
  internal friction within the molecular structure of a
• 25:07
  material as it dynamically bends, flexes, and strains.
• 25:10
  So every time you bend a piece of metal, the atomic lattice is
• 25:13
  deforming. Exactly.
• 25:14
  When the metal strains under a load, it absorbs potential
• 25:18
  energy. When the load is removed and the
• 25:20
  metal relaxes back to its original shape, it releases that
• 25:23
  energy. But not all of it.
• 25:25
  No, not all of it. A tiny fraction of that energy
• 25:28
  is lost as internal heat due to the molecular friction.
• 25:31
  If you plot the stress versus the strain over a full cycle of
• 25:35
  vibration, the curves going up and coming down do not perfectly
• 25:39
  overlap. They format hysteresis.
• 25:41
  Loop. They form an ellipse called a
• 25:42
  hysteresis loop. The area inside that loop
• 25:45
  represents the energy permanently lost to heat per
• 25:47
  cycle. And here is where another
• 25:50
  massive modeling misconception completely derails the
• 25:54
  optimization process. Engineers often assume that
• 25:57
  structural damping inside a piece of steel behaves
• 26:01
  mathematically just like viscous damping inside an oil dashpot
• 26:05
  across all frequencies. This is a critical era that
• 26:08
  leads to catastrophic high frequency failures.
• 26:10
  Because they're completely different mechanisms, right?
• 26:13
  With true viscous damping the oil dashpot, the energy
• 26:16
  dissipated per cycle increases linearly with the frequency of
• 26:19
  the vibration. The faster the oscillation, the
• 26:22
  more energy the fluid shears absorb.
• 26:25
  But solid materials don't work that way.
• 26:27
  Now, decades of experimental physical evidence show that
• 26:29
  structural hysterisis loss, the energy dissipated internally by
• 26:33
  solid materials like steel or aluminum, is largely independent
• 26:37
  of the frequency. It depends on the amplitude.
• 26:39
  Exactly. The internal atomic friction
• 26:41
  depends on the amplitude of the displacement, how far you bend
• 26:44
  the metal, not on how many times per second you are bending it.
• 26:47
  So if an engineer forces the mathematical model to treat
• 26:50
  solid steel like it's a fluid dashpot, what happens to their
• 26:54
  physical prototype? If you treat structural damping
• 26:57
  as if it were a viscous fluid in your finite element analysis
• 27:00
  software, the underlying equations will automatically
• 27:04
  predict that the damping force gets stronger and stronger as
• 27:07
  the frequency goes higher. That sounds dangerous.
• 27:09
  It is you look at the colorful stress plots on your computer
• 27:13
  simulation and think excellent at 1000 Hertz.
• 27:16
  My structure is heavily damped and totally safe from resonance.
• 27:20
  But in physical reality, the structural damping of the steel
• 27:23
  that is constant, it didn't increase with frequency.
• 27:26
  Your simulation will show a massive safety margin, but your
• 27:29
  physical prototype on the shaker table will experience explosive
• 27:33
  fatigue failure at those high frequencies because the actual
• 27:36
  physical damping was vastly lower than your viscous
• 27:39
  mathematical model predicted. So how do you fix it in the
• 27:42
  math? To fix this, you cannot use a
• 27:44
  simple viscous coefficient. We have to introduce a specific
• 27:48
  structural damping parameter, often denoted mathematically as
• 27:51
  a complex stiffness modified fire, to force the equations to
• 27:55
  keep the damping force invariant with frequency.
• 27:58
  I see. But wait.
• 27:59
  Let me look at this from a high level design perspective.
• 28:01
  If adding damping prevents this Wrecking Ball resonance and
• 28:05
  absorbs all this destructive energy by turning it into heat,
• 28:09
  why don't we just slather every single structural component in
• 28:14
  heavy rubber, viscoelastic goo and fluid dash bots and call it
• 28:18
  a day? Why isn't maximum damping always
• 28:20
  the ultimate goal? That is the instinctive
• 28:22
  reaction. But structural dynamics is
• 28:25
  always a game of trade-offs. trade-offs always.
• 28:27
  First, there is the fundamental thermodynamic issue.
• 28:31
  Damping does not magically erase energy from the universe.
• 28:34
  It's strictly converts kinetic energy into thermal energy.
• 28:37
  Right heat. If you slather a highly dynamic
• 28:39
  dynamic system in thick discolastic rubber and subjected
• 28:42
  to continuous vibration, the internal molecular friction will
• 28:46
  generate massive amounts of localized heat.
• 28:48
  And rubber isn't great at shedding heat.
• 28:50
  Because rubber is a terrible thermal conductor, the heat
• 28:53
  cannot escape. The core temperature of the
• 28:55
  rubber will spike, it will melt, its molecular bonds will
• 28:58
  degrade, and the isolator will physically disintegrate.
• 29:00
  So you basically trade a vibration problem for a thermal
• 29:03
  failure problem. Exactly.
• 29:05
  And the second issue is about high frequency transmissibility.
• 29:09
  Oh right, pumping actually hurts you at higher frequencies.
• 29:12
  Precisely in the context of a vibration isolation mount,
• 29:16
  adding damping lowers the peak response exactly at the resonant
• 29:20
  frequency. That is exactly what you want.
• 29:22
  OK, but. But at frequencies well above
• 29:24
  resonance in the isolation region of the spectrum, a highly
• 29:28
  damped mount will actually transmit more vibration force to
• 29:31
  the foundation than a lightly damped mount.
• 29:34
  That's totally counterintuitive. It is the heavy damping stiffens
• 29:38
  the dynamic response at high speed, acting like a rigid
• 29:41
  bridge that passes the high frequency chatter directly into
• 29:44
  your delicate equipment. You are intentionally trading
• 29:47
  low frequency safety at residence for high frequency
• 29:50
  punishment across the rest of the spectrum.
• 29:53
  It is an incredibly delicate optimization problem that
• 29:56
  requires knowing exactly what frequencies you're trying to
• 29:59
  defeat. Which means we are flying blind
• 30:02
  until we know the frequencies. Before we can even begin to
• 30:05
  apply the right physical fixes like viscous dampers, tuned mass
• 30:09
  isolators, or stiffness modifiers, we have to actually
• 30:12
  measure the physical problem. We need hard data from the real
• 30:16
  world. Data is everything.
• 30:18
  But the data acquisition phase is where the most insidious,
• 30:22
  totally invisible traps are hidden.
• 30:24
  Let's look at how we measure this stuff.
• 30:26
  But if you mess up the measurement, you are doomed.
• 30:28
  Data acquisition is the critical translation bridge between the
• 30:32
  messy physical world of moving metal and the pristine
• 30:35
  mathematical world of digital computers.
• 30:37
  And if that bridge is flawed? If that bridge is structurally
• 30:40
  compromised during the measurement phase, every single
• 30:43
  simulation, calculation, and optimization you do afterward is
• 30:46
  completely worthless. You are optimizing a fantasy.
• 30:50
  So there are four major horsemen of bad data that engineers
• 30:53
  constantly run into when trying to record vibration.
• 30:56
  Let's ride through them and unpack the physics of why they
• 30:58
  ruin data #1 clipping. Clipping happens when the
• 31:01
  electrical gain on your instrumentation, specifically
• 31:04
  your signal amplifier or your analog to digital converter, is
• 31:08
  set too high for the physical environment.
• 31:10
  OK, said too high. Right.
• 31:12
  A physical vibration sensor, like an accelerometer, produces
• 31:16
  a tiny electrical voltage signal proportional to the
• 31:19
  acceleration. If the physical shock is
• 31:22
  incredibly violent, it produces a massive voltage spike if.
• 31:26
  That spike is too big. If that voltage spike exceeds
• 31:29
  the maximum electrical limit of your recording device, say your
• 31:33
  system can only handle 5 volts and the sensor sends 8 volts.
• 31:37
  The system literally cannot. Record it so it just chops it
• 31:40
  off. The top and bottom peaks of the
• 31:42
  analog waveform get violently chopped off flat at the 5 Volt
• 31:45
  line. A picture is like trying to
• 31:46
  forcefully walk a six foot tall man into a room with a 5 foot
• 31:50
  ceiling. His head and his feet are just
• 31:52
  violently sheared off. The room cannot contain the
• 31:55
  reality of his height. That is exactly what happens
• 31:58
  electrically, and the mathematical consequences for
• 32:00
  your analysis are dire. First, by chopping off the
• 32:04
  extreme physical peaks of the vibration, you are artificially
• 32:07
  lowering the overall root mean square, or RMS value of your
• 32:11
  recorded data. So your software will tell you
• 32:13
  the vibration environment is significantly less severe than
• 32:16
  it actually is in reality. Exactly.
• 32:18
  But the second consequence is even worse.
• 32:21
  Those flat, chopped off peaks look like perfect sharp square
• 32:25
  waves to your digital analyzer. And why is that bad?
• 32:28
  If you understand Fourier theory, you know that a
• 32:31
  perfectly sharp square wave can only be created mathematically
• 32:34
  by summing an infinite series of extremely high frequency sine
• 32:38
  waves. Oh no.
• 32:39
  So the act of cloning the data instantly injects a massive
• 32:42
  amount of fake artificial high frequency noise into your data
• 32:46
  spectrum. You will spend weeks trying to
• 32:48
  design A damper to fix a high frequency vibration that does
• 32:52
  not exist anywhere in the physical world.
• 32:54
  Right, it only exists as an electrical scar created by your
• 32:57
  bad game setting. All right, so if turning the
• 32:59
  game too high chops the head off our data and injects fake
• 33:03
  frequencies, the obvious play it safe engineers fix is just to
• 33:07
  turn the electrical gain way down, right?
• 33:09
  Make the voltage tiny so it never clips.
• 33:12
  But I'm guessing that introduces the second horseman.
• 33:15
  It introduces a totally different nightmare
• 33:17
  instrumentation noise. This is when your gain is set
• 33:20
  far too low. Every electronic system in the
• 33:23
  world, the copper cables, the amplifiers, the silicon sensors,
• 33:27
  has an inherent electrical noise floor.
• 33:30
  This is caused by the random thermal movement of electrons
• 33:33
  inside the circuitry. And if your gain is too low?
• 33:36
  If you turn your gain down too low, the tiny voltage signal
• 33:39
  generated by the actual physical vibration of your machine
• 33:42
  becomes smaller than the random thermal noise of your cables.
• 33:45
  The real signal is completely buried in a sea of electrical
• 33:48
  fuzz. Yep.
• 33:49
  If you base your structural design on this deeply buried
• 33:51
  data, you might end up designing a heavy steel stiffener to
• 33:55
  survive the random thermostatic of your amplifier rather than
• 33:58
  the actual mechanical vibration of your engine.
• 34:01
  OK, we avoid clipping and we avoid the noise floor, but the
• 34:04
  third horseman is the big one. It is the ghost maker aliasing.
• 34:08
  Aliasing is a fundamental mathematical failure of the
• 34:11
  digital sampling theorem, and it is responsible for massive
• 34:14
  engineering failures. Let's.
• 34:16
  Break this one down slowly. The physical world is continuous
• 34:20
  analog data, but computers can only understand discrete digital
• 34:23
  numbers. When we convert the continuous
• 34:26
  analog voltage from our sensor into a digital signal, we take
• 34:29
  discrete point in time snapshots or samples at specific time
• 34:33
  intervals, which we call delta T.
• 34:35
  OK, point in time snapshots. The Nyquist Shannon sampling
• 34:38
  theorem dictates a hard mathematical law.
• 34:40
  For this process to accurately capture a frequency, you must
• 34:45
  sample at a rate that is at least twice as fast as the
• 34:47
  highest frequency present in your analog signal.
• 34:50
  And for our listeners visualizing the math, the
• 34:52
  Nyquist frequency limit FA equals 1 / 2 times delta T.
• 34:56
  Exactly FA equals 1 / 2 times delta T.
• 35:00
  What physically happens in the data if an engineer fails to
• 35:03
  respect that mathematical speed limit?
• 35:05
  If a violent high frequency vibration exists in the physical
• 35:08
  world, say at 10,000 Hertz, but your computer is only taking
• 35:12
  samples at 8000 times a second, the digital sampling process
• 35:17
  cannot capture the wave fast enough to define its true shape.
• 35:20
  So what does the computer do? Instead, the mathematics of the
• 35:23
  digital conversion fold that high frequency energy backward
• 35:27
  into the lower frequency part of your data spectrum.
• 35:30
  The 10,000 Hertz physical vibration masquerades as a low
• 35:33
  frequency digital vibration. I always explain aliasing to
• 35:37
  people by having them. Imagine watching an old western
• 35:39
  movie where a Stagecoach is driving away.
• 35:42
  The movie camera is essentially taking discrete samples, 24
• 35:45
  frames per second. The wagon wheel effect, right?
• 35:49
  If the wooden spokes on the Stagecoach wheel are spinning so
• 35:52
  fast that the rate they pass the camera lens exceeds the cameras
• 35:55
  frame rate, a terrifying optical illusion happens.
• 35:59
  The wheel suddenly looks like it has stopped, and then it looks
• 36:01
  like it is spinning slowly backward even though the
• 36:04
  Stagecoach is moving forward. It's the perfect visual analogy.
• 36:07
  The high speed physical rotation is aliased into a low speed
• 36:11
  visual representation because the camera couldn't sample the
• 36:14
  light fast enough. Now map that exact illusion onto
• 36:18
  a high stakes and engineering pain point.
• 36:20
  You are a structural engineer looking at your newly acquired
• 36:24
  digital data from a prototype test and you see a massive,
• 36:28
  terrifying energy spike at exactly 50 Hertz.
• 36:31
  OK, you spend the next six weeks and perhaps hundreds of
• 36:34
  thousands of dollars in fabrication costs designing
• 36:36
  heavy steel stiffeners and tuned mass stampers to kill a 50 Hertz
• 36:41
  rumble. You install the fix, you run the
• 36:43
  machine, and the machine immediately tears itself apart.
• 36:46
  Because the 50 Hertz rumble never existed physically.
• 36:49
  The real physical vibration was actually screaming at 10,000 and
• 36:52
  50 Hertz, but because you sampled the data too slowly, the
• 36:55
  high frequency energy alias down and appeared as a digital ghost
• 36:58
  on your screen at 50 Hertz. You just optimized a multi
• 37:01
  $1,000,000 physical structure to fight a phantom electrical
• 37:05
  illusion. Which is horrifying.
• 37:07
  A phantom problem that cost millions.
• 37:10
  But wait, I have to challenge the premise of this entire
• 37:12
  problem. We are living in an era where
• 37:14
  data storage is incredibly cheap.
• 37:17
  Terabytes of solid-state storage are practically free.
• 37:21
  Processing power is astronomical.
• 37:23
  I see where this is going. Why don't we just mandate that
• 37:25
  every single sensor samples at a billion times a second?
• 37:29
  Wouldn't over sampling to Infinity completely outrun the
• 37:32
  Nyquist limit and eliminate aliasing while also fixing our
• 37:36
  4th Horseman quantization error? In a purely theoretical,
• 37:40
  frictionless universe, infinitely faster sampling is
• 37:43
  better, but practically trying to sample structural data at a
• 37:47
  billion times a second creates immediate, insurmountable
• 37:51
  physical bottlenecks. Really like what?
• 37:53
  First, analog to digital converters that can run at
• 37:56
  gigahertz speeds while maintaining high bit resolution
• 37:59
  are incredibly expensive. They draw massive amounts of
• 38:01
  power, and they generate immense heat.
• 38:03
  You cannot stick a liquid cooled supercomputer on the wing of a
• 38:06
  fighter jet. That's fair.
• 38:08
  Second, while the hard drives might be cheap, the transmission
• 38:11
  bandwidth is not. If you are trying to push
• 38:14
  billions of data points per second through a telemetry
• 38:17
  antenna from a rocket tearing through the ionosphere, or
• 38:20
  through a sonar cable from a moving submarine, you will
• 38:23
  instantly saturate your transmission lines and lose the
• 38:26
  data stream entirely. So you literally can't get the
• 38:29
  data out. No you can't.
• 38:30
  And 3rd as you mentioned, there is the quantization error.
• 38:33
  Right, let's unpack the 4th Horseman quantization error.
• 38:36
  Quantization error is the digital round off noise
• 38:39
  inherently caused by the analog to digital converters bit
• 38:43
  resolution. It is the staircase effect.
• 38:45
  The staircase effect. Yeah, and analog voltage is a
• 38:48
  perfectly smooth continuous curve, but a 16 bit digital
• 38:53
  converter only has exactly 65,536 discrete microscopic
• 38:58
  voltage steps it can use to represent that curve.
• 39:01
  It is like trying to trace a smooth ramp using Lego bricks.
• 39:04
  That's exactly it. The true analog voltage flowing
• 39:07
  from the sensor almost never lands perfectly flat on one of
• 39:10
  those discrete Lego steps, so the computer is forced to round
• 39:13
  the value up or round it down to the nearest available step.
• 39:16
  Which means it's constantly slightly wrong.
• 39:18
  Right. This tiny continuous rounding
• 39:22
  error introduces a permanent underlying floor of white noise
• 39:26
  into your data. If you try to sample at a
• 39:29
  billion times a second, you are forcing the computer to generate
• 39:32
  billions of microscopic round off errors every second,
• 39:36
  drastically raising the noise floor.
• 39:38
  So the standard practice is an infinite sampling.
• 39:41
  The standard professional engineering practice is not to
• 39:44
  rely on infinite digital sampling.
• 39:46
  It is to use a physical analog low pass anti aliasing filter
• 39:51
  placed directly in the circuit board right before the digital
• 39:54
  converter. So you use an analog circuit as
• 39:56
  a bouncer at the club. It physically scrubs away and
• 39:59
  blocks any high frequencies above the Nyquist limit from
• 40:02
  entering the room before they can even reach the digital
• 40:04
  converter and cause ghosting. Brilliant.
• 40:06
  It's the only reliable way. OK, so let's assume we survived
• 40:09
  the gauntlet of the Four Horsemen.
• 40:11
  We set our games perfectly to avoid chopping heads off with
• 40:14
  clipping. We stayed well above the thermal
• 40:16
  noise floor. We used our analog anti aliasing
• 40:19
  filters to kill the ghosts and we have enough bit death to
• 40:22
  minimize the staircase effect. We have clean data.
• 40:25
  We have successfully catured, clean, uncorrupted, pristine
• 40:29
  digital data. We are still just staring at a
• 40:32
  massive, unintelligible, jagged, squiggly line of voltage over
• 40:36
  time on a computer screen. Time history data.
• 40:39
  Staring at that chaotic time history data doesn't actually
• 40:42
  tell me how to physically fix the brackets on my engine.
• 40:45
  How do we decode that squiggly line?
• 40:48
  You have hit the exact transition point between data
• 40:51
  acquisition and data analysis. Staring at a complex time
• 40:55
  history trace is like looking at a bowl of mixed paint and trying
• 40:59
  to figure out exactly what colors went into it.
• 41:01
  It's impossible to tell by just looking at it.
• 41:04
  It rarely tells an engineer the specific physical frequencies
• 41:06
  that are causing the damage. To extract actionable
• 41:09
  information, we have to fundamentally shift our
• 41:11
  perspective. We have to move out of the time
• 41:14
  domain and translate the data into the frequency domain.
• 41:17
  And we do this using a mathematical algorithm called
• 41:19
  the Fast Fourier Transform or FFT.
• 41:22
  Exactly. I love the concept of the
• 41:24
  Fourier transform. It feels like real world magic.
• 41:27
  I think of it like unbaking a cake.
• 41:29
  Unbaking a cake. Yeah, you have this incredibly
• 41:31
  complex baked time history cake that represents the total
• 41:35
  vibration. The FFT algorithm takes that
• 41:38
  cake, unbakes it, and prints out a receipt telling you exactly
• 41:41
  how much flour, sugar, and eggs went into it.
• 41:43
  I like that. Or in our engineering case, it
• 41:46
  tells you exactly how much 10 Hertz energy, 50 Hertz energy,
• 41:49
  and 1000 Hertz energy combined to create that complex, chaotic
• 41:54
  rattle. Unbaking the cake is an
• 41:56
  excellent way to visualize the theorem that Fourier developed
• 41:59
  over 200 years ago. The FFT algorithm mathematically
• 42:03
  decomposes the complex jagged waveform into a vast spectrum of
• 42:07
  individual perfect sine waves, each with its own specific
• 42:11
  frequency and its own specific amplitude.
• 42:13
  But this is where engineers must be incredibly careful.
• 42:15
  Right, absolutely, because you have to choose the right
• 42:17
  mathematical translation tool based on the specific type of
• 42:21
  vibration environment you identified back in step one.
• 42:23
  OK, let's revisit that. What if the data is
• 42:25
  deterministic? If your data is deterministic,
• 42:28
  like the steady predictable hum of a diesel generator, you use a
• 42:32
  simple line spectrum. When you run the FFT on
• 42:35
  deterministic data, the output will show distinct sharp
• 42:39
  vertical spikes at specific discrete frequencies.
• 42:42
  It tells you exactly how fast the generator is spinning and
• 42:45
  exactly what harmonics it is generating.
• 42:47
  Right, But if the environment isn't a steady hum, what if the
• 42:51
  data is random vibration? The wobbly shopping cart full of
• 42:55
  bowling balls rolling over cobblestones.
• 42:57
  Yeah, what then? If you try to use a simple line
• 42:59
  spectrum to analyze random vibration, the output is
• 43:02
  entirely useless because random vibration is chaotic.
• 43:06
  The amplitudes are constantly shifting millisecond by
• 43:08
  millisecond, and the energy is smeared continuously across a
• 43:12
  massive broadband of frequencies.
• 43:14
  There are no clean spikes. There are no clean distinct
• 43:17
  spikes to look at for random data.
• 43:19
  You must abandoned the line spectrum and translate the data
• 43:22
  using power spectral density or PSD.
• 43:25
  Let's really breakdown the mechanics of PSD, because the
• 43:27
  acronym alone sounds deeply intimidating to a lot of junior
• 43:30
  engineers. What is it actually calculating?
• 43:33
  Power spectral density is a statistical tool.
• 43:36
  Mathematically, it calculates the limiting mean square value
• 43:40
  of the vibration, which relates directly to the physical energy
• 43:43
  or power contained in the signal, and divides that energy
• 43:47
  by the frequency resolution bandwidth.
• 43:49
  So you're looking at energy distribution?
• 43:51
  You are essentially looking at how the total destructive energy
• 43:55
  of the vibration is distributed across the entire frequency
• 43:58
  spectrum. As the measurement time
• 44:00
  approaches Infinity, you have to stop thinking about exact
• 44:03
  amplitudes. Right, because it's chaotic.
• 44:06
  Instead of a line spectrum saying there is exactly 2 GS of
• 44:10
  peak acceleration precisely at 50 Hertz, APSD plot tells you
• 44:15
  there is a specific density of continuous acceleration energy
• 44:18
  concentrated in the frequency band around 50 Hertz.
• 44:20
  Yeah, the total energy. Is the area under the curve of a
• 44:23
  PSD plot. Gives you the overall root mean
• 44:25
  square energy of the entire chaotic environment.
• 44:28
  But just like measuring the data HEAD traps, the mathematical
• 44:31
  translation process itself has traps, right?
• 44:34
  What are the processing errors engineers make when they run
• 44:36
  these FFTS and PS DS? The most common and deceptive
• 44:40
  processing error is leakage error.
• 44:42
  Leakage error. What's up?
• 44:44
  When you command a computer to perform an FFT, the underlying
• 44:47
  mathematics inherently assume that the finite 10 second sample
• 44:52
  of data you gave it repeats itself infinitely forever in
• 44:55
  both directions. It pretends your small clip is
• 44:58
  the whole universe. Exactly.
• 44:59
  But if the jagged voltage at the exact end of your 10 second data
• 45:03
  sample does not perfectly line up with the voltage at the exact
• 45:06
  beginning of the next hypothetical repeated sample, it
• 45:09
  creates an artificial jagged discontinuity.
• 45:13
  And the math hates a sudden jump.
• 45:14
  The math algorithm panics at the sudden discontinuity, and it
• 45:17
  tries to fix it by leaking the true energy from the actual
• 45:21
  frequency into all the neighboring frequencies.
• 45:24
  A sharp, distinct peak of energy gets sneered out into a wide,
• 45:27
  shallow mound. So you lose your resolution.
• 45:30
  How do you fix that? Engineers have to actively
• 45:32
  counter this mathematical artifact by applying a digital
• 45:35
  window to the data before running the FFT.
• 45:38
  The A window, like a Hanning or flat top window, is a
• 45:41
  mathematical curve that forces the very ends of the data sample
• 45:45
  to taper smoothly down to 0, eliminating the discontinuity
• 45:49
  and stopping the energy from leaking.
• 45:50
  OK, so that fixes leakage. And beyond leakage, there's the
• 45:54
  fundamental, almost philosophical balancing act
• 45:57
  between random errors and bias errors when you are setting up
• 46:01
  your PSD parameters. Yes, this is the ultimate
• 46:04
  compromise in digital signal processing.
• 46:06
  It is almost like the Heisenberg uncertainty principle for
• 46:09
  vibration analysis. Also, when you process a PSD,
• 46:12
  you want the plot to be smooth and accurate.
• 46:15
  To reduce random error, which is the statistical variance that
• 46:18
  makes your PSD plot look like a jagged, noisy, unreadable
• 46:21
  mountain range. You need to average a massive
• 46:24
  number of data samples together over a very long period of time.
• 46:27
  So longer averaging means smoother data.
• 46:29
  Right, but to reduce bias error, which is a lack of resolution
• 46:33
  that causes the algorithm to artificially blunt and
• 46:35
  underestimate the true destructive height of shark
• 46:38
  resonance peaks, you need to set a very narrow frequency
• 46:41
  resolution bandwidth. The mathematical reality is that
• 46:44
  you cannot have both infinite averaging time and an infinitely
• 46:48
  narrow resolution bandwidth. You have a finite amount of
• 46:51
  recorded data. You are forced to balance the
• 46:54
  two errors based on the physical realities of what you are trying
• 46:57
  to find. So if you need a long averaging
• 47:00
  time to smooth out the random error, I have to ask the
• 47:03
  difficult messy reality question.
• 47:05
  Hit me, What does this all mean for non stationary data?
• 47:09
  If the physical environment is constantly changing, like our
• 47:12
  rocket accelerating through the atmosphere and experiencing
• 47:14
  maximum dynamic pressure, how can you possibly take an average
• 47:19
  over a long time period without completely blurring out and
• 47:22
  hiding the exact momentary extreme events we are trying to
• 47:25
  study? That is the ultimate paradox of
• 47:27
  non stationary data analysis and it is incredibly difficult to
• 47:30
  solve. Because if you average the whole
• 47:32
  launch. If you average the rocket's
• 47:33
  vibration over a full 62nd time period to get a smooth curve,
• 47:37
  you completely smooth out and erase the violent 3 second
• 47:40
  transient peak of maximum aerodynamic pressure that might
• 47:43
  actually tear the payload apart. The math hides the danger.
• 47:47
  And if you average too short? If you average over too short of
• 47:50
  a time period, say just half a second, your statistical
• 47:54
  confidence drops to near 0, the random error skyrockets, and
• 47:58
  your spectrum just looks like meaningless static.
• 48:01
  So how do you win? The modern computational
• 48:03
  solution is to use a running average, creating what is called
• 48:06
  a waterfall plot or a spectrogram.
• 48:09
  A waterfall plug. You slice the non stationary
• 48:11
  time history data into very short overlapping pseudo
• 48:14
  stationary segments. You compute an instantaneous
• 48:18
  frequency spectrum for each tiny slice, and you stack them
• 48:22
  sequentially in a 3D plot with time on the Z axis.
• 48:26
  That sounds computationally heavy.
• 48:27
  It requires immense computational power, but it
• 48:30
  allows an engineer to visually see how the frequency content
• 48:34
  physically shifts dynamically through time as the rocket
• 48:37
  accelerates, without blurring the critical extreme peaks into
• 48:41
  statistical oblivion. That is how we decode a
• 48:43
  constantly changing environment. OK, so we've translated the
• 48:46
  continuous hums, the random rattles, and the shifting
• 48:48
  environments. We know how to unbake the cake
• 48:51
  using FFTS and PFTS, but what happens when the event isn't
• 48:55
  continuous? You mean shock?
• 48:57
  Right. While vibration analysis looks
• 49:00
  at ongoing continuous environments where energy builds
• 49:03
  up over time, shock analysis requires an entirely different
• 49:06
  engineering mindset. The event is over, the explosion
• 49:09
  has happened, the package has hit the concrete before the
• 49:12
  internal structure even realizes what hit it.
• 49:15
  How do we model the sudden violence?
• 49:17
  Mechanical shock is defined purely by its suddenness and its
• 49:20
  severity. It is a transient excitation.
• 49:23
  Transient. Right when we analyze continuous
• 49:26
  vibration, we care deeply about the steady state response, how
• 49:30
  the structure behaves after the forces have settled into a
• 49:33
  rhythm. In shock analysis, there is no
• 49:35
  rhythm. So what do we care about?
• 49:36
  We care about the primary response, which is the massive
• 49:39
  deflection that occurs precisely while the shock pulse is
• 49:42
  actively happening, and we care about the residual response,
• 49:45
  which is the free, unforced vibration of the structure
• 49:47
  ringing out like a bell after the external shock force has
• 49:50
  completely passed. And to standardize how we test
• 49:53
  for these complex chaotic real world impacts, we model these
• 49:56
  shocks using idealized classical physics shapes, right?
• 50:00
  Yes. To make laboratory testing
• 50:02
  repeatable and mathematical analysis possible, we distill
• 50:05
  complex real world impacts into classical shock forms.
• 50:09
  Like what The most extreme theoretical form is the
• 50:12
  acceleration impulse. We model this mathematically
• 50:16
  using the Dirac delta function. The delta function.
• 50:18
  Which describes an infinitely high acceleration occurring in
• 50:22
  an infinitesimally small fraction of time, delivering a
• 50:25
  massive specific velocity change instantly.
• 50:28
  In reality, nothing is truly instantaneous, but a hard hammer
• 50:32
  strike on an anvil gets very close.
• 50:34
  OK, in another form. Then there is the step velocity,
• 50:36
  where the foundation of a structure instantly changes
• 50:39
  speed, like a car crashing into a concrete wall makes sense.
• 50:42
  And in the massive realm of industrial drop testing, making
• 50:45
  sure your laptop survives falling off a desk, we
• 50:48
  frequently use the Half Sign Pulse.
• 50:50
  This is a smooth, rounded impact profile that mathematically
• 50:53
  looks exactly like the top half of a sine wave.
• 50:56
  It models an object hitting a slightly compliant surface, like
• 50:59
  a cushioned floor. But just looking at a time
• 51:02
  history plot of 1/2 sign drop pulse, just seeing that it was a
• 51:06
  50G impact that lasted for 11 milliseconds doesn't actually
• 51:10
  tell an engineer if the delicate silicon microchips on the
• 51:13
  internal circuit board will fracture.
• 51:14
  No it doesn't. To understand the actual
• 51:17
  physical damage otential, we have to translate that pulse
• 51:20
  using the SRS, the shark response sectrum.
• 51:24
  This is another area where I see massive confusion in the
• 51:27
  industry. It is arguably the most
• 51:29
  profoundly misunderstood tool in all of dynamic structural
• 51:33
  analysis. I see engineers misuse it
• 51:35
  constantly. Let's clear it up then.
• 51:36
  Let's establish A foundational rule.
• 51:38
  Right now. The shock response spectrum does
• 51:40
  not show the shock itself. It does not show the frequency
• 51:43
  content of the input pulse. It is not an FFT of the impact.
• 51:47
  So what does the SRS actually show?
• 51:49
  The SRS calculates the absolute maximum physical response of a
• 51:52
  theoretical series of single degree of freedom structures to
• 51:56
  that specific shock. Wait, a theoretical series?
• 51:58
  Yeah. Imagine you have a rigid steel
• 52:00
  base plate. On that base plate you mount a
• 52:03
  row of 100 tiny independent theoretical mass spring systems.
• 52:08
  Each one is tuned to have a completely different natural
• 52:11
  frequency. The first one is 10 Hertz, the
• 52:13
  next is 20 Hertz all the way up to 10,000 Hertz.
• 52:16
  OK, I'm picturing this row of springs.
• 52:18
  Now you subject that entire base plate to a violent physical
• 52:22
  shock, like an 11 millisecond half sign drop.
• 52:25
  As the shock hits, all those tiny springs compress and expand
• 52:28
  wildly. The SRS algorithm calculates and
• 52:32
  records the absolute maximum peak response, the maximum
• 52:35
  displacement velocity, or equivalent static acceleration
• 52:38
  that each individual spring mass system experienced either during
• 52:42
  the primary shock or during the residual ringing afterward.
• 52:45
  Just the maximum peak. Just the peak.
• 52:46
  You then plot those maximum peak responses on the Y axis against
• 52:50
  the natural frequencies of the systems on the X axis.
• 52:53
  That resulting curve is the shock response spectrum.
• 52:56
  The underlying mathematics for calculating this involves
• 52:58
  solving a complex convolution integral called the Duhamel
• 53:02
  integral, for every single one of those theoretical mass spring
• 53:05
  systems. This brings us directly to the a
• 53:08
  massive misconception that ruins shock testing and destroys
• 53:12
  millions of dollars of hardware. It is the error of confusing the
• 53:16
  Fourier spectrum with the shock response spectrum.
• 53:20
  I look at it this way, the Fourier Spectra is studying the
• 53:23
  punch. It tells you the mass, the
• 53:25
  velocity, and the exact frequency content of the steel
• 53:29
  Wrecking Ball swinging at a building.
• 53:31
  It analyzes the weapon. Excellent analogy.
• 53:34
  But the SRS is studying the bruise left behind.
• 53:37
  It tells you exactly how far the building's internal support
• 53:40
  beams physically bent and yielded when they got hit by
• 53:44
  that weapon. They're 2 completely different
• 53:46
  perspectives. That is an outstanding, highly
• 53:48
  accurate analogy. The four year spectrum breaks
• 53:50
  down the composition of the input energy.
• 53:52
  The SRS measures the resulting damage potential to the
• 53:55
  structure. Engineers constantly mix these
• 53:58
  two up because both plots have frequency on the X axis.
• 54:01
  Just grab the wrong tool. An engineer will take a shock
• 54:03
  pulse from a test, run a standard FFT on it, look at the
• 54:06
  frequency content of the pulse itself, and then try to design
• 54:09
  their circuit board to avoid the frequencies present in the
• 54:12
  pulse. But the board doesn't care about
• 54:14
  the input frequencies. Exactly.
• 54:16
  The internal structure doesn't care about the mathematical
• 54:20
  Fourier spectrum of the input, it strictly cares about how far
• 54:23
  its own springs are physically forced to deflect.
• 54:27
  The SRS curve tells you exactly what the maximum equivalent
• 54:30
  static acceleration will be for a structure of any given natural
• 54:33
  frequency. If you design your hardware
• 54:35
  based on the Fourier spectrum of the shock instead of the SRS of
• 54:38
  the shock. You are optimizing the mass and
• 54:41
  stiffness for the wrong parameters entirely, and your
• 54:43
  board will snap in half upon impact.
• 54:46
  But let me challenge the entire foundational concept of the SRS.
• 54:50
  If the entire shock response spectrum relies on
• 54:53
  mathematically simulating a bunch of incredibly simplified,
• 54:56
  independent single degree of freedom mass spring systems
• 55:00
  sitting on a theoretical base plate, how can that simplistic
• 55:03
  model possibly give us accurate, actionable design parameters for
• 55:07
  something as incredibly complex as a spacecraft satellite or a
• 55:11
  naval submarine? A submarine pressure hull isn't
• 55:14
  a simple coil spring. It seems wildly counterintuitive
• 55:18
  that such a simple model works for such complex structures, but
• 55:22
  it works Florida State because of the theory of modal
• 55:24
  superposition that we discussed earlier.
• 55:26
  Back in Section 2. Yes, a naval submarine is vastly
• 55:30
  complex, but mathematically its entire dynamic response to a
• 55:34
  shockwave can be broken down into a linear combination of its
• 55:38
  independent normal modes. Each one of those complex
• 55:41
  bending and twisting modes behaves mathematically exactly
• 55:45
  like one of those simplistic theoretical single degree of
• 55:49
  freedom mass spring systems. That's incredible.
• 55:52
  If you run a modal analysis and determine the natural
• 55:54
  frequencies of the submarines, critical internal decks and
• 55:57
  bulkheads, you can then look at the SRS plot generated by the
• 56:00
  underwater explosive shockwave. You simply find the frequencies
• 56:05
  of your bolt heads on the X axis of the SRS plot and you can
• 56:08
  immediately read the exact maximum peak acceleration each
• 56:12
  bulkhead will experience on the Y axis.
• 56:14
  Just reading it right off the chart.
• 56:15
  It gives you the absolute worst case boundary conditions for
• 56:18
  your structural design without having to run an incredibly
• 56:22
  computationally expensive full scale transient finite element
• 56:25
  analysis of the entire submarine for every possible explosion
• 56:30
  scenario. That is the immense power of
• 56:33
  translating the data properly. It simlifies the chaos.
• 56:37
  OK, we have thoroughly maed the theory.
• 56:39
  We've navigated the invisible data measurement Tras of
• 56:42
  aliasing and clipping. We've translated the time
• 56:45
  history data using PS DS for random vibration and the SRS for
• 56:49
  sudden shocks. We've covered a lot of ground.
• 56:51
  Now we can finally walk through the ultimate step by step
• 56:54
  approach to taking all this math and fixing the physical design
• 56:56
  in the real world. Let's talk about step by step
• 56:59
  optimization. This is where the mathematical
• 57:01
  rubber meets the physical Rd. All the theory we just discussed
• 57:04
  feeds directly into a rigid 4 step methodology for controlling
• 57:08
  shock and vibration. In practice, if you skip a step,
• 57:10
  you fail. OK, step one, identify.
• 57:13
  You must correct categorize the excitation.
• 57:16
  Is your input a deterministic hum from a motor, a random
• 57:20
  broadband rumble from turbulence, or a transient shock
• 57:23
  from a drop? You cannot mathematically apply
• 57:26
  a PSD analysis to a predictable engine vibration, and you cannot
• 57:29
  use a line spectrum FFT to understand 1/2 sign drop.
• 57:34
  If you misidentify the enemy, you will use the wrong weapon.
• 57:37
  Step 2. Step 2 is capturing clean,
• 57:41
  uncorrupted data from the physical environment.
• 57:43
  You must meticulously set your electrical gains properly to
• 57:47
  avoid clipping the high peaks or drowning the signal and the
• 57:49
  thermal instrumentation noise floor.
• 57:51
  And handle the aliasing. And you absolutely must ensure
• 57:54
  your hardware is using analog low pass filters to enforce the
• 57:58
  Nyquist limit and prevent high frequency aliasing from creating
• 58:01
  digital ghosts in your spectrum. Step 3.
• 58:03
  Analyze. Use the correct domain
• 58:05
  translation tool. If you are dealing with a steady
• 58:07
  state continue vibration, move from the time domain to the
• 58:11
  frequency domain using the fast Fourier transform.
• 58:13
  Use line Spectra to analyze deterministic rotating equipment
• 58:17
  and use power spectral density to analyze random aerodynamic
• 58:20
  turbulence. And watch out for errors.
• 58:22
  And when using PSD, carefully balance your bias and random
• 58:26
  errors using the appropriate mathematical windowing and
• 58:29
  averaging times. If the event is a sudden
• 58:32
  transient shock, completely ignore the FFT and use the shock
• 58:36
  response spectrum to understand the maximum damage potential to
• 58:39
  your specific modes. And finally Step 4, control.
• 58:44
  This is implementing the physical fix on the hardware.
• 58:48
  From all the engineering sources and case studies we review, we
• 58:51
  can group these physical fixes into 3 distinct buckets of
• 58:54
  action. Bucket one is reduction at the
• 58:56
  source. This is always the best, most
• 58:59
  direct and most efficient method.
• 59:01
  Fix the problem where it starts. Exactly.
• 59:03
  If a massive rotating cooling fan is vibrating and shaking
• 59:06
  your chassis, you don't add heavy structural supports, You
• 59:09
  balance the moving mass of the fan.
• 59:11
  You add or subtract tiny weights to the rotor blades until the
• 59:14
  eccentric centrifugal forces completely cancel each other
• 59:17
  out. Like balancing a tire on a car.
• 59:19
  Right. If a gear bearing is rattling
• 59:22
  violently, you tighten the tolerances and fix the
• 59:24
  clearances. You alter the operational
• 59:27
  parameters to stop the destructive vibration energy
• 59:30
  from ever being generated in the first place.
• 59:32
  But sometimes you simply cannot touch the source.
• 59:35
  You cannot ask a turbulent ocean to calm down, and you cannot ask
• 59:39
  the aerodynamic pressure on a rocket to stop.
• 59:43
  That leads us to bucket 2. Bucket 2 is isolation.
• 59:47
  You place a highly engineered dynamic barrier between the
• 59:50
  violent source and the delicate receiver.
• 59:52
  Like mounting a hammer. Yeah, if a massive heavy forging
• 59:56
  hammer is shaking the concrete floor of a manufacturing plant
• 59:59
  and destroying nearby precision milling machines, you mount the
• 1:00:02
  heavy hammer on massive pneumatic shock isolators to
• 1:00:05
  protect the building and the reverse.
• 1:00:07
  Conversely, if you have highly delicate server electronics
• 1:00:10
  inside a naval ship that is experiencing massive hull shock
• 1:00:12
  from crashing. Your waves.
• 1:00:14
  You mount the delicate electronics on specialized wire
• 1:00:17
  rope shock isolators to protect them from the ship.
• 1:00:19
  You physically break the transmission path of the energy.
• 1:00:22
  And if you can't balance the source and you don't have the
• 1:00:25
  space to isolate it, you are left with bucket 3 reduction of
• 1:00:29
  the response. This is where you are forced to
• 1:00:32
  physically alter the structure of the receiver itself.
• 1:00:36
  You can change its internal natural frequency to avoid the
• 1:00:39
  resonance Wrecking Ball by welding on stiffeners to
• 1:00:42
  increase the K value or bolting on mass to increase the mem
• 1:00:45
  value. Or adding damping.
• 1:00:47
  You can add energy dissipation by applying thick layers of
• 1:00:51
  viscoelastic damping materials to the housing to turn the
• 1:00:54
  kinetic energy into heat. Or you can attach an auxiliary
• 1:00:58
  mass dynamic vibration absorber or a tuned mass damper.
• 1:01:02
  That one is a brilliant engineering trick.
• 1:01:04
  It really is. You attach a secondary small
• 1:01:07
  mass spring system to your main structure, mathematically tuned
• 1:01:10
  to vibrate completely out of phase with your main structures
• 1:01:13
  movement. So they fight each other.
• 1:01:15
  As the main structure moves up, the tuned mass damper slams
• 1:01:18
  down, applying A counteracting inertial force that completely
• 1:01:22
  cancels out the motion of the primary structure.
• 1:01:24
  These practical control steps always lead to some common
• 1:01:27
  questions we get from the field regarding standardization and
• 1:01:29
  testing. Engineers want to know how to
• 1:01:32
  validate that these physical fixes actually work before they
• 1:01:35
  ship the product. The standards question.
• 1:01:38
  Yeah, what about relying on military standards like ISO TC
• 1:01:41
  108 and MIL STD 810? People constantly ask, should we
• 1:01:46
  just use the generic vibration test specifications published in
• 1:01:50
  these manuals for our components to save testing time and money?
• 1:01:53
  The definitive industry wide answer in modern structural
• 1:01:56
  engineering is absolutely not not at all no.
• 1:01:59
  While documents like MIL, STD 810 and the ISO standards
• 1:02:03
  provide excellent, rigorously proven guidelines and procedures
• 1:02:07
  for exactly how to operate a shaker table or how to mount a
• 1:02:10
  sensor, the modern standard practice is to completely avoid
• 1:02:13
  using their generic baseline test levels.
• 1:02:16
  But people used to do that. In the past, engineers would
• 1:02:18
  just pull a generic highway transport vibration curve from a
• 1:02:21
  dusty manual and test to that. But the modern mandate is
• 1:02:24
  tailoring. You must tailor your test to
• 1:02:26
  your specific measured physical environment.
• 1:02:29
  So you have to go and get the data.
• 1:02:30
  You must physically acquire raw data from the actual operating
• 1:02:34
  environment of your specific hardware.
• 1:02:36
  Put sensors on the actual Humvee driving through the actual mud.
• 1:02:40
  Analyze it using the exact steps we just outlined and derive a
• 1:02:44
  custom PSD test specification that mathematically replicates
• 1:02:48
  your true localized field conditions.
• 1:02:51
  Because generic tests are risky. Using generic pre published
• 1:02:54
  tests is a massive risk. They will either severely under
• 1:02:58
  test your part leading to catastrophic field failures
• 1:03:00
  because your specific environment was harsher, or they
• 1:03:03
  will severely over test your part forcing you to over design,
• 1:03:06
  over stiffen and add massive amounts of unnecessary waiting
• 1:03:09
  cost to survive a vibration that the part will never actually C
• 1:03:13
  in reality. Let me bring UA messy reality
• 1:03:15
  question about that third control bucket tuned mask
• 1:03:18
  Tamers. We talk about how you can
• 1:03:20
  theoretically tune a secondary auxiliary mask to erfectly
• 1:03:23
  counteract your rimary vibration, but paper is patient
• 1:03:26
  and steel isn't. That's the truth.
• 1:03:28
  How do we actually dial in a tuned damper in practice, Out in
• 1:03:32
  a freezing oil field or inside a hot engine compartment when
• 1:03:36
  variables like ambient temperature completely change
• 1:03:39
  the molecular stiffness of our rubber isolators?
• 1:03:43
  We already established an hour ago the dynamic stiffness is a
• 1:03:46
  massive trap. If the temperature drops, the
• 1:03:48
  rubber gets harder, the K value changes, and the math falls
• 1:03:51
  apart. That is the ultimate Crucible of
• 1:03:54
  mechanical engineering reality ruining your math.
• 1:03:57
  You calculate the perfect tuned mass damper on your pristine
• 1:04:01
  whiteboard. You install it in the field.
• 1:04:03
  It works perfectly at room temperature, and then a cold
• 1:04:05
  front rolls in. And then suddenly it is 20°
• 1:04:09
  colder, the polymer chains in the rubber spring of your damper
• 1:04:12
  stiffen up, the dynamic K value skyrockets, the natural
• 1:04:16
  frequency of the damper shifts away from the frequency you were
• 1:04:18
  trying to cancel, and your damper is now completely out of
• 1:04:21
  tune and entirely useless. Or worse.
• 1:04:24
  Or worse, it starts amplifying the main structures vibration.
• 1:04:27
  In practice, seasoned engineers handle this brutal reality by
• 1:04:30
  designing broadband dampers. Broadband dampers.
• 1:04:33
  Instead of tuning the absorber to an impossibly narrow shark
• 1:04:36
  frequency peak, they intentionally inject a specific
• 1:04:40
  amount of viscous or structural damping directly into the
• 1:04:43
  absorber itself. This added damping flattens out
• 1:04:47
  the mathematical response curve. OK.
• 1:04:48
  So it covers more ground. The damper becomes slightly less
• 1:04:51
  effective at the absolute perfect peak frequency, but it's
• 1:04:55
  cancellation effect is smeared out over a much wider frequency
• 1:04:59
  band. So even if environmental
• 1:05:01
  temperature shifts the rubber stiffness and moves the natural
• 1:05:04
  frequency left to right, the damper stays within its broad
• 1:05:07
  effective operating window. That's so smart.
• 1:05:10
  It is the classic engineering compromise, trading peak
• 1:05:13
  theoretical performance for robust, fail safe real world
• 1:05:17
  reliability. Trading whiteboard perfection
• 1:05:19
  for real world reliability. That is exactly what mechanical
• 1:05:22
  engineering made simple is all about.
• 1:05:24
  So what does this deep dive mean for you, the listener?
• 1:05:26
  The one actually designing these parts.
• 1:05:28
  It means that the chaotic, destructive mechanics of shock
• 1:05:31
  and vibration are not black magic.
• 1:05:34
  They are not random acts of destruction.
• 1:05:36
  They are governed by rigid physical laws that can be
• 1:05:39
  understood, measured and completely controlled as long as
• 1:05:43
  you respect the mathematics and avoid the invisible traps.
• 1:05:46
  Exactly. The physics are impartial
• 1:05:48
  whether you are identifying the difference between deterministic
• 1:05:51
  inputs and random chaos, making sure your dynamic stiffness
• 1:05:54
  calculations account for polymer locking, actively respecting the
• 1:05:58
  Nyquist limit with analog filters during your digital data
• 1:06:01
  acquisition to prevent aliasing, or ensuring you use the shock
• 1:06:05
  response spectrum instead of a Fourier transform to measure the
• 1:06:08
  actual structural bruising of a transient impact, the core
• 1:06:11
  principles hold true. By systematically rigorously
• 1:06:14
  moving through the four steps of identification, measurement,
• 1:06:17
  analysis, and control, you can mathematically tame any dynamic
• 1:06:21
  environment. I want to leave you with a final
• 1:06:23
  provocative thought to Mull over as you head back to your CAD
• 1:06:26
  models and finite element simulations.
• 1:06:29
  Throughout this entire deep dive we have been talking about
• 1:06:32
  passive structural systems, assive masses, assive springs,
• 1:06:36
  fluid dash pots, ieces of hysteretic rubber.
• 1:06:39
  Things that just sit there and react to the physics imposed
• 1:06:41
  upon them. Exactly.
• 1:06:43
  But the industry is moving rapidly toward the era of active
• 1:06:46
  vibration control. We are talking about integrating
• 1:06:49
  smart materials, piezoelectric sensors and high speed
• 1:06:52
  electromagnetic actuators directly into the load path that
• 1:06:56
  can mathematically adapt their damping properties and stiffness
• 1:06:59
  in real time based on live data. Active control is the future.
• 1:07:03
  Will the static passive masses and springs we painstakingly
• 1:07:07
  designed today become entirely obsolete in a decade?
• 1:07:11
  Imagine a near future where a multi $1,000,000 turbine can
• 1:07:14
  actively listen to its own internal frequencies, anticipate
• 1:07:18
  a chaotic random excitation from the environment, calculate the
• 1:07:21
  exact Fourier transform of a threat in microseconds, and
• 1:07:24
  physically fire opposing electromagnetic actuators to
• 1:07:27
  completely cancel out the motion before the Wrecking Ball of
• 1:07:30
  resonance even has a chance to form.
• 1:07:32
  It's an incredible thought. The predictable metronome and
• 1:07:35
  the wobbly shopping cart on the cobblestone St. might both be
• 1:07:38
  entirely defeated by mechanical structures that can
• 1:07:41
  mathematically think, anticipate, and violently ush
• 1:07:44
  back. If you got something useful out
• 1:07:46
  of today's Eisode, hit follow and drop a like.
• 1:07:49
  It helps grow the show and keeps the engineering knowledge
• 1:07:51
  flowing.