Random Vibration

FEA LEARNING CENTER

Random Vibration

Living in the Noise

By Joseph P. McFadden Sr.

McFaddenCAE.com

Companion document to the FEA Learning Center

in the Abaqus INP Comprehensive Analyzer

The real world doesn't vibrate at a single frequency.

Ride in a car and the vibration coming through the seat isn't a clean sine wave. It's a mix of everything — road texture, tire tread, engine firing, suspension dynamics — all layered on top of each other, changing every second. Strap a satellite to a launch vehicle and it experiences the same kind of broadband random excitation from acoustic noise, aero-buffit, and engine thrust oscillation.

You can't describe that environment with a single frequency and amplitude. You describe it with a power spectral density — a PSD — which tells you how much vibration energy exists at each frequency across the entire band. And random vibration analysis tells you how your structure responds to that broadband input.

This is the analysis that qualifies satellites for launch. That qualifies automotive electronics for decades of road exposure. That tells you whether your PCB will survive the vibration environment or accumulate enough fatigue cycles to crack a solder joint.

The Why — What Random Vibration Captures That Others Don't

Harmonic response tells you what happens at each individual frequency. Shock tells you what happens during a transient event. Random vibration tells you what happens when the structure lives in broadband noise — continuously excited at all frequencies simultaneously, with the energy level varying across the frequency band.

The response is statistical, not deterministic. And this is where random vibration fundamentally differs from every other analysis type — so let me be very deliberate about what that means.

In a static analysis, you apply a load and get a stress. That stress is a single, definite number at each point. The part experiences exactly that stress, right now, at this load. In a harmonic analysis, you get an amplitude at each frequency — a definite peak value per cycle. In a transient shock analysis, you get a time history — you can point to a specific instant and say "at 1.3 milliseconds, the stress was 47 Megapascals."

Random vibration gives you none of that.

Because the input is random — broadband noise with energy at all frequencies simultaneously — the response at every point in the structure is also random. The stress at any location is fluctuating continuously, up and down, every instant different from the last. There is no single stress value to report. There is no "the stress is 47 Megapascals." There is only "the stress fluctuates, and here is the probability of it being within a given range at any instant in time."

What Abaqus gives you as output is the RMS stress — the root mean square — at every point in the model. And here's what you must understand: for random vibration with a zero-mean response, which is the case for virtually all structural vibration, the RMS value is the standard deviation of the stress. It's not a peak. It's not an average. It's the statistical spread of the fluctuating response.

This statistical nature is what makes random vibration analysis both powerful and easy to misinterpret. The stress contour plot you see in CAE is not a stress that the part experiences at any particular moment. It's a probability envelope. And reading it correctly requires understanding the Gaussian distribution that underlies it.

The What — Psd, Grms, And What They Really Mean

Before I throw PSD numbers and GRMS levels at you, let me explain what those numbers actually mean. Because if you don't understand the units, the values are just noise.

Start with a fundamental problem. Vibration oscillates. The acceleration goes positive, then negative, then positive again — it swings back and forth around zero. If you try to describe the intensity of that vibration by taking a simple average of the acceleration over time, the positive and negative values cancel each other out and the average is zero. Or very close to it. A device shaking violently at 5 G peak has an average acceleration near zero. That average tells you nothing useful.

So we need a different measure of intensity — one that captures the magnitude of the oscillation regardless of whether the acceleration is positive or negative at any given instant.

The solution is to square the signal first. Squaring makes every value positive — a positive acceleration squared is positive, and a negative acceleration squared is also positive. Now instead of values canceling, they all contribute. Take the mean of those squared values, and you have the mean-square acceleration — a number that faithfully represents the average intensity of the vibration. Take the square root of the mean-square, and you're back in the original units — G's.

That is what RMS means. Root. Mean. Square. Square the values to eliminate the sign, average them to get the mean power, take the square root to return to physical units. It's the same reason electrical engineers use RMS voltage for AC power — the average of a sine wave is zero, but the RMS tells you its effective value.

Now extend that concept across frequency.

A PSD — power spectral density — breaks the vibration signal apart by frequency and tells you how much energy exists in each narrow frequency band. The vertical axis is G-squared per Hertz. G-squared because we've already done the squaring step — each value represents the mean-square acceleration per unit bandwidth at that frequency. Per Hertz because the energy is normalized to a one-Hertz-wide band so that you can compare measurements made with different frequency resolutions.

The PSD is a map of vibration energy versus frequency. Peaks in the PSD show you where the environment concentrates its energy. Flat regions show broadband content. Rolloffs show where the energy dies out. It tells you not just how intense the vibration is, but where in the frequency spectrum that intensity lives.

And the overall GRMS level? That's the single number that collapses the entire PSD curve into one measure of total vibrational energy. Mathematically, GRMS is the square root of the area under the PSD curve — integrate G-squared per Hertz across all frequencies, and you get G-squared. Take the square root, and you get GRMS. It's the RMS acceleration of the entire broadband signal.

Think of it this way. A truck floor vibrating at 1.04 GRMS — that's the MIL-STD-810 highway carrier profile — has a total vibrational energy represented by that single number. You can feel it. It's a noticeable buzz through the floorboards, a constant hum that makes a coffee cup slowly migrate across a dashboard. A forklift at 0.1 GRMS is barely perceptible to a human — a subtle tremor you might not notice unless you're looking at an accelerometer readout. A launch vehicle at 14 GRMS is violent — everything is shaking hard enough that bolted joints can loosen and unsupported wiring can chafe through insulation in seconds.

But GRMS alone does not tell you the full story. Two PSD profiles can have the same GRMS — the same total energy — but concentrate it at completely different frequencies. A profile with all its energy at 5 Hertz will excite a completely different set of structural modes than a profile with its energy at 500 Hertz. Same GRMS. Very different damage. The PSD shape matters as much as the overall level.

And here's the critical framing for everything that follows. The PSD is the input. It characterizes the environment — what the outside world does to the mounting points of your system. It's the vibration at the base of your shaker table, the acceleration at the vehicle floor, the motion at the forklift carriage, the energy transmitted through the conveyor structure. It describes the poking. It says nothing about how your structure responds to that poking.

That's what the analysis is for.

Real-world Psd Environments

NASA-STD-7001, also known as GEVS — General Environmental Verification Standard — is the most widely referenced PSD specification for spacecraft components. The acceptance level runs from 20 to 2,000 Hertz with a plateau around 0.04 G-squared per Hertz, giving an overall level of about 6.8 GRMS. The qualification level is 6 dB higher — four times the power spectral density — at about 13.6 GRMS. That's roughly double the vibration intensity, applied for twice the duration, and it's the level that proves your design has margin.

Automotive PSD profiles are typically lower in level but cover a different frequency range — 5 to 500 Hertz, with overall levels of 1 to 3 GRMS. The key difference is duration. Spacecraft vibration qualification lasts 60 to 120 seconds per axis. Automotive vibration lasts for the lifetime of the vehicle — thousands of hours. The damage accumulates over millions of cycles, making fatigue the primary failure mode rather than peak stress.

But aerospace and automotive are not the only worlds that care about random vibration. If you design rugged handheld devices, mobile computers, barcode scanners, wearable terminals, vehicle-mounted tablets, machine vision systems, or any electronics that live in warehouses, distribution centers, delivery vehicles, and logistics infrastructure — you are in the random vibration business whether you realize it or not. And the environments these devices face are surprisingly diverse. Each one has a different PSD character, a different frequency range, and a different failure mechanism.

Let me walk through them, because understanding these environments is the difference between a product that survives the real world and one that fails in ways you never tested for.

Truck And Trailer Transportation

Every product gets shipped. And during shipping, the product lives on a truck or trailer that vibrates continuously for hours or days.

MIL-STD-810 Method 514.8 defines several truck transportation PSD profiles based on decades of measured data. The US highway common carrier profile — representing a standard tractor-trailer on interstate roads — covers roughly 5 to 500 Hertz. The vertical axis is the most severe, with a flat plateau around 0.015 G-squared per Hertz from 10 to 40 Hertz, rolling off at higher frequencies. The overall vertical level is about 1.04 GRMS. The longitudinal axis is lower at about 0.74 GRMS, and the lateral axis is the mildest at about 0.20 GRMS.

The composite wheeled vehicle profile — representing military trucks and field transportation over unprepared roads — is substantially more severe. This profile was compiled from measurements across many military trucks and trailers, combined into a worst-case envelope, and then doubled for conservatism. The energy is concentrated at low frequencies, below 100 Hertz, with significant content as low as 5 Hertz. Overall levels are roughly 3 to 5 GRMS depending on the specific vehicle category. If your device might end up in a military logistics chain or on rough service roads in developing regions, this is the profile to design for.

For commercial parcel delivery, the ISTA 3-series protocols — from the International Safe Transit Association — define PSD profiles based on extensive measurements of real delivery vehicles. The over-the-road trailer profile has an overall level of about 0.53 GRMS. A pickup and delivery vehicle — the last-mile van that brings the package to a loading dock or doorstep — runs about 0.46 GRMS. These levels are lower than the military profiles but the durations are much longer, and the cumulative fatigue damage from days of continuous transport can exceed the damage from a short qualification test at higher intensity.

The critical insight for product designers: truck vibration energy peaks in the 3 to 30 Hertz range. This is where the vehicle body bounces on its suspension springs. If your product has a fundamental frequency in this range — and cradle-mounted tablets and vehicle-mounted computers often do — truck transportation will excite it at resonance for hours.

Forklift Handling

This is one of the most overlooked vibration environments, and for warehouse electronics it may be the most important one.

Forklifts operate on concrete and asphalt warehouse floors that are not perfectly smooth. Expansion joints, floor patches, dock plates, ramp transitions, and surface irregularities all generate continuous vibration as the forklift traverses the facility. Published measurements show the dominant vibration energy occurs at very low frequencies — 2.5 to 5 Hertz — corresponding to the forklift's suspension and tire dynamics. The energy decreases rapidly above 10 Hertz and is negligible above 120 Hertz. The recommended test spectrum runs from 1 to 120 Hertz with six breakpoints to capture the low-frequency peak and high-frequency rolloff.

The overall GRMS levels are modest — typically 0.05 to 0.15 GRMS — much lower than truck transport. But the dominant frequencies are very low, which means the displacements are large even though the accelerations are small. A 0.1 GRMS signal at 3 Hertz produces far more displacement than the same level at 300 Hertz. That displacement translates to relative motion in cradle mounts, connector pins, charging contacts, and docking stations.

Several factors affect forklift vibration intensity. Higher speed increases vibration roughly proportionally. Gas-powered forklifts vibrate more than electric forklifts due to engine harmonics and rougher power delivery. Asphalt floors produce higher vibration than smooth concrete due to surface irregularities. And counterintuitively, lighter loads increase vibration — the additional mass of a heavy pallet lowers the system's natural frequency and damps the response. Heavier pallets actually smooth the ride.

For a vehicle-mounted terminal bolted to a forklift dashboard, or a scanner holstered on the operator, the forklift environment is sustained, repetitive, and relentless — eight hours a day, five or six days a week, for the service life of the device. The vibration level at any instant is low. The cumulative exposure is enormous. Fatigue of solder joints, connector pins, display flex cables, and cradle latching mechanisms is the primary failure mode — not overstress, but accumulated damage from billions of low-amplitude cycles.

And here's the complication that most test standards don't capture. Forklift vibration is not purely Gaussian. The floor transitions, dock plate crossings, and ramp entries produce intermittent transient spikes — brief high-G events embedded in the low-level random background. Research shows kurtosis values at the fork tine tips ranging from 1 to nearly 14 — far from the Gaussian value of zero. That means the standard three-sigma assumption underestimates the peak stresses the device actually sees. For critical applications, consider using higher sigma multipliers or kurtosis-controlled vibration testing.

Conveyor Systems And Sortation Equipment

Inside a modern distribution center, products don't just sit on shelves. They move — continuously, at speed, through an interconnected network of conveyor belts, diverters, sorters, merges, and accumulation zones. And every transition is a vibration event.

Belt conveyors generate low-frequency vibration from belt sag between rollers, roller eccentricity, and drive motor harmonics. The dominant frequencies depend on belt speed and roller spacing. A typical belt running at 1 to 2 meters per second over rollers spaced 300 millimeters apart produces a fundamental excitation around 3 to 7 Hertz, with harmonics extending to 50 or 60 Hertz. Overall GRMS levels on the belt surface are typically 0.1 to 0.3 GRMS — similar to forklift levels.

But the transient events are the real concern. Every time a package transfers from one conveyor section to another — at merge points, diverter gates, and sortation chutes — the item experiences a brief shock impulse as it crosses the gap between belt sections or changes direction. These transfers happen dozens or hundreds of times per trip through a large fulfillment center. Each one is a low-level shock — perhaps 2 to 10 G for a few milliseconds — but they accumulate. And for a handheld device or scanner riding in a tote on the conveyor, the repeated impacts at transfer points stress the same locations every time.

Roller conveyors — where the product rides directly on unpowered rollers — produce a distinctive vibration signature dominated by the roller passage frequency. A package moving at 0.5 meters per second over rollers with 75-millimeter pitch crosses a roller every 150 milliseconds, giving a fundamental around 6.7 Hertz. But the contact is not smooth — each roller transition produces a small impact as the leading edge of the package drops onto the next roller. The result is a broadband excitation with a comb-like structure — peaks at the roller passage frequency and its harmonics.

High-speed sortation systems — tilt-tray sorters, cross-belt sorters, sliding shoe sorters — introduce lateral accelerations as items are diverted from the main conveyor path. Tilt-tray and cross-belt sorters can generate lateral G levels of 1 to 3 G during the divert action, with a duration of 50 to 200 milliseconds. These are really repeated shocks rather than random vibration, but they contribute to the cumulative fatigue environment.

For fixed infrastructure electronics — scanner arrays mounted above conveyor lanes, vision triggers, weigh-in-motion sensors, print-and-apply labelers — the vibration source is the conveyor structure itself. The steel frame transmits motor vibration, belt tension fluctuations, and package impacts throughout the structure. A scanner mounted on a gantry over a high-speed conveyor sees continuous low-level vibration punctuated by package-arrival impacts, hour after hour, for years.

Machine Vision Gantry Systems And Tunnel Scanners

Machine vision systems and multi-sided scan tunnels — the kind used in automated parcel sorting, airport luggage handling, and high-speed logistics lines — have a vibration environment all their own.

These systems typically consist of a rigid steel gantry or tunnel structure spanning a conveyor line, with cameras, laser scanners, barcode readers, and illumination arrays mounted at precise positions and angles. The vibration concern is not damage to the electronics in the traditional sense — it's optical stability. Even microscopic vibration of a camera or scanner relative to its target can degrade image quality, reduce barcode read rates, and cause dimensional measurement errors.

The vibration sources are multiple. The conveyor belt running through the tunnel transmits motor vibration and package impacts into the gantry structure through the floor and through shared mounting frames. Adjacent conveyors, sorters, and material handling equipment contribute background vibration through the building floor slab. In airport baggage handling systems, the environment is compounded by aircraft engine run-ups, ground support equipment, and the massive air handling systems in the terminal building — all transmitted through the structure.

The frequency content is typically 5 to 200 Hertz, with the most problematic energy in the 10 to 60 Hertz range where gantry structural resonances tend to fall. A steel gantry spanning 2 to 3 meters with camera brackets has bending and torsional modes in exactly this range. If a conveyor motor harmonic coincides with a gantry mode, the camera mount amplifies the vibration and image quality degrades even though the source vibration is small.

Overall GRMS levels at the gantry mounting points are typically low — 0.02 to 0.1 GRMS. But the amplification at resonance can be dramatic. A Q of 20 at a gantry resonance means the camera mount sees 20 times the floor vibration at that frequency. If the floor has 0.01 G-squared per Hertz at 40 Hertz and the gantry has a mode at 40 Hertz with Q of 20, the camera mount experiences 400 times that spectral density at resonance — 4 G-squared per Hertz locally. That's enough to blur images and reduce read rates.

The design response is either to shift the gantry resonances away from the dominant conveyor excitation frequencies — by stiffening the gantry or adding mass — or to isolate the camera mounts with elastomeric bushings that decouple the optics from the structural vibration. Both approaches require knowing the gantry's natural frequencies, which means modal analysis. Random vibration analysis then predicts the RMS displacement at the camera mount, which translates directly to image blur. If the three-sigma displacement exceeds the depth of field or the pixel pitch of the imaging system, read rate will suffer.

The Common Thread

Every one of these environments — truck transport, forklift handling, conveyor systems, and machine vision gantries — shares three characteristics that set them apart from aerospace and automotive vibration.

First, the levels are lower. We're talking 0.05 to 1.0 GRMS, not the 7 to 14 GRMS of a launch vehicle. But the durations are vastly longer — hours, days, years. Fatigue dominates over peak stress.

Second, the frequency content is concentrated at the low end — 2 to 50 Hertz. This is the regime of suspension dynamics, roller passage, belt sag, and structural modes. Devices designed to survive high-frequency launch vibration may be completely uncharacterized for the low-frequency environments where they actually spend their service life.

Third, the vibration is not purely Gaussian. Floor transitions, package impacts, transfer points, and dock plate crossings inject transient spikes into the random background. Standard Gaussian-based qualification tests may not replicate the peak stresses from these non-stationary events. Kurtosis-controlled testing and mixed random-plus-transient profiles are emerging as better representations of these real-world environments.

From Input To Response — How Your Structure Filters The Environment

Everything I've described so far — the truck PSD, the forklift spectrum, the conveyor vibration, the gantry excitation — is the input. It's what the environment does to the base of your structure. The PSD that appears in the spec or the test standard describes the motion at the mounting points, not the response of the structure itself.

And this distinction is everything.

Your structure is not a rigid block that experiences the input vibration uniformly. It's a dynamic system with its own natural frequencies, mode shapes, and damping. It has a transfer function — a frequency-dependent gain that tells you how much the structure amplifies or attenuates the input at each frequency.

At frequencies well below the first natural frequency, the structure moves essentially as a rigid body with the base. The response equals the input. No amplification, no attenuation — one-to-one transmission.

At a natural frequency, the response amplifies. Dramatically. The amplification factor is Q — the transmissibility at resonance. For a lightly damped structure with 2 percent damping, Q is 25. That means the response acceleration at resonance is 25 times the input acceleration at that frequency. In PSD terms, the amplification is Q-squared — 625 times the input power spectral density at that frequency. A modest input of 0.01 G-squared per Hertz at a frequency that happens to coincide with your structure's resonance becomes 6.25 G-squared per Hertz at the response point. The structure is selectively amplifying the input energy at its own preferred frequencies.

At frequencies well above the natural frequency, the response attenuates. The structure can't keep up with the rapid oscillations, and the input energy rolls off. This is the isolation region — the same physics that vibration isolators exploit.

So the response PSD at any point in your structure looks very different from the input PSD. The input might be flat — broadband energy spread evenly across the frequency band. But the response will have sharp peaks at every natural frequency where the structure resonates, with valleys between them. The response spectrum is the input spectrum reshaped by the structure's own dynamics.

This is why two structures exposed to the same input PSD can have completely different stress levels. A PCB with a first bending mode at 200 Hertz picks up energy from the input at 200 Hertz and amplifies it. A stiffer PCB with a first mode at 800 Hertz ignores the 200-Hertz content entirely and amplifies the 800-Hertz content instead. Same input. Different response. Different stress. Different fatigue life.

The overall GRMS of the response is also different from the GRMS of the input. Because the structure amplifies energy at resonance and attenuates it elsewhere, the response GRMS depends on where the natural frequencies fall relative to the input energy. If the structure's resonances align with the peaks of the input PSD, the response GRMS can be several times higher than the input. If the resonances fall in valleys or outside the PSD band, the response GRMS can be lower.

This is the whole point of random vibration analysis. You give Abaqus the input PSD — the environment. The modal analysis gives Abaqus the structure's natural frequencies and mode shapes — the transfer function. And the solver multiplies the input through the transfer function at every frequency, at every point, to compute the response PSD. The RMS values you see in the contour plot are the square roots of the areas under those response PSD curves — the GRMS of the response at each location in your model.

The input PSD is the question. The structure's dynamics are the filter. The response PSD is the answer. And that answer varies from point to point across the structure because different locations participate differently in different mode shapes. A point at the center of a PCB — where the first bending mode has maximum displacement — will have a much higher response than a point near a screw hole where that same mode has a node. Same input. Same structure. Different locations. Different response spectra. Different RMS stress.

Understanding this input-to-response chain is what separates someone who runs an analysis from someone who understands what the analysis is telling them.

The How — Random Vibration In Abaqus

Like SRS, random vibration analysis in Abaqus is a two-step perturbation procedure.

Step one: modal analysis. Extract natural frequencies and mode shapes covering at least the PSD frequency range. If the PSD goes to 2,000 Hertz, extract modes to at least 3,000 or 4,000 Hertz. Check that cumulative effective mass exceeds 90 percent of total mass in each excitation direction.

Step two: random response. Define the PSD input, specify modal damping, and the solver computes the RMS response at every point in the model. The output is statistical — RMS stress, RMS displacement, RMS acceleration.

Damping is critical. In random vibration, damping controls the amplification at resonance — and since the broadband input excites every resonance simultaneously, the damping directly affects the overall response level. Too little damping gives overly conservative results. Too much damping gives unconservative results.

Two percent damping — that's a damping ratio of 0.02 — is typical for machined aluminum structures with minimal joints. Three percent is a common default for assembled structures with bolted connections. Five percent is appropriate for structures with many joints, potting, gaskets, or intentional damping treatments like vibration isolators. When in doubt, use less damping — it's more conservative. And note: you damp a system. You do not dampen it. Engineers damp. Towels dampen.

The perturbation limitation applies fully. Contact elements cannot be used. Material nonlinearity is not captured. Large deformations are not considered. If your structure has contact interfaces, model them as tied constraints. This is the same restriction as modal, SRS, and harmonic — it's inherent to all perturbation procedures.

Interpreting The Output — What Rms Really Means

This is the section that matters most, and it's the one most engineers get wrong. So let's take our time.

When Abaqus finishes a random vibration analysis and you look at the stress contour plot, you're looking at the RMS stress field. At every node, at every element integration point, the software has computed the root-mean-square of the fluctuating stress response to the broadband PSD input.

What does that number physically represent?

Imagine you could freeze time and measure the stress at one point in your structure at one random instant. Then unfreeze time, let it vibrate for a while, and freeze it again. Measure again. Do that a thousand times. You'd get a different stress value every time — because the response is random. But if you plotted all thousand measurements as a histogram, you'd get a bell curve. A Gaussian distribution. The center of that bell curve is zero — the mean stress is zero because the vibration oscillates symmetrically about the at-rest state. And the width of that bell curve — how spread out the values are — is the RMS stress.

The RMS is the standard deviation of the response. And once you know the standard deviation of a Gaussian distribution, you know everything about the probability of the response falling in any range.

Here's the probability table that every structural dynamics engineer should have memorized.

At one sigma — one times the RMS — 68.27 percent of the time, the instantaneous stress falls within plus-or-minus one RMS of zero. That means roughly two-thirds of the time, the stress is within this band. But one-third of the time, it's outside it. That's too frequent for design.

At two sigma — two times the RMS — 95.45 percent of the time, the stress falls within this range. Only about 4.5 percent of the time does the stress exceed two sigma. That's better, but for structural qualification, 4.5 percent exceedance over millions of cycles is still a lot of high-stress events.

At three sigma — three times the RMS — 99.73 percent of the time, the stress stays within this range. Only 0.27 percent of the time — roughly one instant in 370 — does the stress exceed three sigma. This is the standard design level for structural qualification. When engineers say "design to three sigma," they mean: take the RMS stress, multiply by three, and compare that value to your material allowable.

Let me make this concrete with numbers. Suppose the RMS stress at a critical solder joint is 5 Megapascals.

At one sigma, the stress exceeds plus-or-minus 5 Megapascals about 31.73 percent of the time. Nearly one-third of every second, the stress is above 5 or below minus 5 Megapascals.

At two sigma — 10 Megapascals — the stress exceeds this level about 4.55 percent of the time. During a 60-second test, that's roughly 2.7 seconds spent above 10 Megapascals.

At three sigma — 15 Megapascals — the stress exceeds this level only 0.27 percent of the time. During a 60-second test, that's about 160 milliseconds total. This is your design stress for that location: 15 Megapascals. Compare that to the solder fatigue strength.

Now — and this is critical — that three-sigma level assumes Gaussian statistics. Real vibration environments are not perfectly Gaussian. They can have higher-than-expected peaks, a property measured by a quantity called kurtosis. A Gaussian distribution has a kurtosis of 3. Real environments — especially those with transient events mixed into the random background — can have kurtosis of 5, 7, or higher. Higher kurtosis means fatter tails in the probability distribution, which means more frequent excursions beyond three sigma than the Gaussian model predicts.

For critical applications, some organizations use four-sigma or even five-sigma design levels. At four sigma, only 0.006 percent of the time does the stress exceed the threshold. At five sigma, it's less than one in a million. But three sigma remains the standard for most structural qualification.

The Contour Plot Is Not A Stress Map

I want to say this one more time because it's the source of so many misinterpretations.

When you look at a random vibration stress contour in Abaqus CAE, the colors do not represent a stress state that the part experiences at any moment. The part never looks like that contour plot. At any given instant, the actual stress distribution is different — some locations higher, some lower, some reversed in sign — all fluctuating randomly.

What the contour shows is the one-sigma envelope. It shows you where the fluctuations are most intense. Red areas are locations where the stress fluctuates the most — the standard deviation is highest. Blue areas are locations where the stress barely fluctuates. The pattern tells you where to worry, but the actual design stress is the contour value multiplied by your sigma factor — typically three.

This also means you cannot simply look at the RMS stress contour and conclude "my part is fine because the RMS stress is below yield." An RMS stress of 80 Megapascals means a three-sigma stress of 240 Megapascals. If your yield strength is 250 Megapascals, you're much closer to failure than the contour plot suggests.

Fatigue — The Accumulation That Peak Stress Misses

Even if the three-sigma stress is below yield, the part may still fail — by fatigue.

The structure is vibrating continuously for the duration of the test — 60 seconds for acceptance, 120 seconds for qualification, thousands of hours for automotive service life. Every cycle of vibration is a stress cycle. And those cycles accumulate damage.

The number of cycles per second is approximately the dominant response frequency — the natural frequency where the resonance peak is highest. If your board resonates at 400 Hertz and the test lasts 60 seconds, that's 24,000 stress cycles. During a 120-second qualification test, it's 48,000 cycles. For an automotive component vibrating at 200 Hertz for 10,000 hours of road life, it's 7.2 billion cycles.

The stress amplitude varies from cycle to cycle — some cycles are small, near one sigma. Some are large, near three sigma or beyond. The Gaussian distribution tells you the probability of each amplitude level. Fatigue damage accumulates at all levels, weighted by the S-N curve of the material.

Miner's rule sums the fractional damage at each stress level. The Dirlik method is a widely used closed-form approach that estimates the probability density function of rainflow cycle amplitudes directly from the moments of the PSD response — without requiring a time-domain simulation. It's computationally efficient and works well for single-degree-of-freedom-dominated responses. For more complex multi-modal responses, time-domain reconstruction from the PSD with rainflow counting may be more accurate.

The key point is this: a part can be well below yield at three sigma and still fail by fatigue if the cycle count is high enough. The three-sigma check tells you whether the part survives the peak instantaneous stress. The fatigue assessment tells you whether it survives the accumulation of millions of moderate stress cycles. You need both.

Applying Rms To Displacement And Acceleration

The same statistical framework applies to every output variable, not just stress.

RMS displacement tells you the standard deviation of the position fluctuation. Three-sigma displacement is your design excursion — use it for clearance checks. Will the vibrating PCB contact the adjacent housing wall? If the three-sigma displacement exceeds the gap, you'll get intermittent contact.

RMS acceleration tells you the standard deviation of the acceleration fluctuation. Three-sigma acceleration at a mounting point, multiplied by the component mass, gives you the peak dynamic load on the attachment — the design load for your screws, adhesive, or solder joints.

RMS relative displacement between two points — say between a component lead and the PCB pad — is the driver for solder joint fatigue. If you can extract the relative displacement RMS, you can compute the strain range directly and feed it into a Coffin-Manson fatigue model.

In every case, the RMS value from Abaqus is the one-sigma level. Multiply by three for the standard design envelope. And remember: the contour is a probability map, not a snapshot.

Practical Wisdom

Test-analysis correlation is the gold standard. If you have vibration test data — measured accelerations at control points and responses at sensor locations — compare your analytical RMS predictions to the measured values. Agreement within 20 to 30 percent for the first several resonant peaks is typical for a well-correlated model. If the discrepancy is larger, the boundary conditions and damping are the first things to check.

Be careful with multi-axis excitation. Real environments are three-dimensional — vibration occurs simultaneously in all three axes. But qualification tests and most analyses apply the PSD one axis at a time. This is conservative for some configurations and unconservative for others. If your structure has cross-axis coupling — excitation in one direction causing significant response in another — a single-axis analysis may underpredict the total response.

And remember that random vibration analysis gives you steady-state statistical response. It doesn't capture startup transients, non-stationary environments, or combined random-plus-deterministic loading. For those, you may need time-domain simulation with a synthesized random time history as input.

Random vibration analysis is one of the four members of the linear perturbation family — along with modal, harmonic, and SRS — that share the same mathematical foundation and the same restrictions. For the deeper discussion of those restrictions, see the dedicated Learning Center discussion on perturbation limitations.

DSP tools for PSD computation, GRMS calculation, and fatigue life estimation from random vibration environments are available at McFaddenCAE.com.

This has been a Learning Center discussion on random vibration. I'm Joe McFadden. Thanks for listening.

About the Author

Joseph P. McFadden Sr. is a CAE engineer specializing in finite element analysis, modal analysis, materials behavior, and injection mold tooling validation. With nearly four decades of experience in structural simulation, he brings a holistic perspective to engineering education — connecting how systems respond to how people think and learn.

His work at McFaddenCAE.com includes the Abaqus INP Comprehensive Analyzer — a desktop tool for analyzing, visualizing, and extracting sub-assemblies from large FEA models without requiring an Abaqus license — along with DSP tools for SRS computation, jerk extraction, velocity change analysis, and energy balance verification.

The FEA Learning Center is an integrated educational platform within the Analyzer, providing guided discussions on structural dynamics topics with working example INP files. This document series is the companion written reference for those discussions.

The four-volume FEA Best Practices audiobook series — Building the Model, The System's Natural Character, When Things Collide, and Keeping the Simulation Honest — is available at McFaddenCAE.com.