Twenty Years of Bone Lattice Fatigue Data, and Still No Failure Criterion

Bone-inspired lattices have been a darling of metamaterials research for a while now. They mimic trabecular bone's open-cell structure, promising stiffness and strength at a fraction of the weight. But there's a catch: fatigue. Twenty years of S-N curves, porosity maps, and strut-level simulations, and still no consensus on how to predict when these parts will actually break.

If you're an engineer tasked with qualifying a bone-like lattice for a cyclically loaded implant or aerospace bracket, you've probably felt the frustration. The literature gives you data, but not a decision. This article lays out the landscape, compares the main failure criteria, and gives you a path forward without overselling any single method.

The Fatigue Criterion Void: Why Two Decades of Data Hasn't Settled It

What a Failure Criterion Is (and Isn’t)

A failure criterion is a rule that says *when* a lattice bone implant stops being safe. Not when it cracks — when it *might* crack, given the loads, the cycles, and the geometry. That sounds simple. It isn’t. I have watched engineers stare at S-N curves for weeks, trying to squeeze a verdict out of scattered points. The data sits there, abundant and stubborn, yet no single equation wins.

The odd part is—we have criteria for solid metals. Von Mises, Goodman, Soderberg. They work because the material is continuous. But a bone lattice is not a block of metal. It’s a network of struts, some thinner than a human hair, with voids between them. Stress concentrates at strut junctions. Strain localizes in the thinnest members. And the fatigue life depends on which strut is misaligned, which node has a micro-porosity, which surface was rough from the print bed.

That gap between lattice geometry and continuum assumptions is the void. Data piles up, but nobody agrees on what to do with it.

The Gap Between Lattice Geometry and Continuum Assumptions

Most fatigue criteria treat a material as if it were a smooth, homogeneous blob. You apply a stress tensor, you get a life estimate. But a lattice is a structure pretending to be a material. Its effective stiffness changes with unit-cell shape, strut diameter, and relative density. Worse, the local stress at a strut surface can be five times the nominal stress you calculated from the global load. That’s not a minor correction — it’s a different world.

Consider a simple cubic lattice. One strut carries the load in compression; the adjacent strut bends. The bending creates tensile stress on one side, compressive on the other. Fatigue cracks love that tension side. A continuum criterion sees only average strain. It misses the bending entirely. So your prediction says 10 million cycles. The test specimen fails at 300,000. That hurts.

The catch is that some teams try to patch this with a stress concentration factor. They multiply the nominal stress by 2.5, call it a day. Then the next lattice design has a different strut angle, and the factor is wrong again. No amount of experimental data fixes that — because the data is specific to each geometry, each print orientation, each surface finish.

Two decades of fatigue tests have produced more curves than conclusions. The community shares data, but not a shared way to interpret it.

— design review note, orthopedic implant lab

Who Feels This Void Most: Implant and Aerospace Engineers

Implant engineers feel it first. A spinal cage or a femoral stem must survive 10–20 years of walking, stair climbing, and occasional falls. The safety factor is razor-thin. Use a criterion that’s too conservative, and the implant is over-designed — stiff, heavy, less bone-friendly. Use one that’s too optimistic, and a patient faces revision surgery. I have seen both outcomes in conference presentations, each team blaming the other’s material characterization.

Aerospace engineers inherit a different pain. They don’t need 20-year lives; they need 20,000 flight cycles with confidence intervals. Their lattices are often sandwich cores, loaded in shear and compression. The fatigue data exists, but the certification bodies require a validated criterion. Without one, you fall back on full-scale testing — expensive, slow, and impossible for every geometry change.

So the void persists. Two decades of data, and still no failure criterion. The tools to create one are there: digital image correlation, micro-CT of crack paths, finite element models that resolve individual struts. The missing piece is agreement on what the criterion should look like — stress-based, strain-based, energy-based, or defect-driven. That choice shapes every downstream decision, from test design to regulatory approval.

The next chapter digs into those four roads. But first, know this: picking a criterion is not a theoretical exercise. It’s a bet on where cracks actually start.

Four Roads to Prediction: Stress, Strain, Energy, or Defects

Stress-Based Criteria: S-N Curves and the Goodman Diagram

Stress-based fatigue prediction leans on the oldest tool in the fatigue handbook: the S-N curve. You cycle a specimen at a fixed amplitude, count cycles to failure, and plot the points. For bone lattices, the appeal is immediate — the test is cheap, the data exists, and engineers already trust the format. The Goodman diagram extends this by folding mean stress into the picture, mapping a safe zone between alternating and average loads.

That sounds fine until you realize the lattice struts rarely see pure uniaxial stress. Bending dominates, especially in gyroid or diamond topologies. So you convert everything to an equivalent stress using von Mises or the principal stress at the strut surface. The catch is that equivalent stress hides the local gradient — the very thing that drives crack initiation in a 200-micron strut. I have watched teams burn two weeks fitting a Goodman curve to data that scattered across a factor of three, only to admit the geometry was the missing variable.

Stress criteria are fast. They're also blunt. They assume the material is homogeneous, which a lattice is not by definition.

Strain-Based Approaches: Coffin-Manson and Local Plasticity

Strain-based methods flip the question. Instead of asking what load the strut carries, you ask how much it deforms before the first microcrack appears. The Coffin-Manson equation ties plastic strain amplitude to reversals-to-failure, and for lattices it often fits better than S-N curves because the struts yield locally long before the bulk structure looks damaged.

The tricky part is measuring strain inside a lattice. Strain gauges are too large; digital image correlation works on the surface but misses interior struts. Most teams fall back on finite element models that extract strain at the strut's outer fiber, then feed that number into Coffin-Manson. The assumption is that surface strain equals the driving force for crack nucleation. That holds for solid metals, less so for lattices where the strut's as-built surface roughness — from laser powder bed fusion, say — already acts as a field of micro-notches.

The payoff is better accuracy at high strain amplitudes. The price is a model that demands calibration per geometry, per build orientation, per machine. Not a one-time cost.

Energy Density and Damage Mechanics

Energy-based criteria skip the stress-strain argument entirely. They calculate the strain energy density accumulated per cycle, then compare it to a critical value. The logic is thermodynamic: fatigue failure is the dissipation of energy into new surface area — cracks. For lattices, this approach captures the contribution of multiple struts failing at once, something local stress or strain criteria struggle with.

Damage mechanics takes this further, introducing a scalar variable that grows from zero to one as the lattice degrades. Each cycle increments that variable based on the current local state. The model can handle stiffness reduction, load redistribution between struts, and even the sudden cascade when the first strut snaps.

That's the good news. The bad news is that damage parameters are notoriously hard to identify without extensive testing. And the evolution law is often chosen for mathematical convenience, not physical fidelity.

Often, the dull step fails first.

Odd bit about science: the dull step fails first.

Odd bit about science: the dull step fails first.

Odd bit about science: the dull step fails first.

Then again, sometimes the prettiest strut is the weakest link.

“You can fit an elephant with enough parameters — but the elephant still won't tell you when it breaks.”

— comment overheard at a metamaterials workshop, 2022

Defect-Driven Criteria: Considering Porosity and Strut Notches

The fourth road starts with the worst truth about additive manufacturing: every strut has defects. Gas pores, lack-of-fusion voids, surface roughness peaks — all of them concentrate stress. Defect-driven criteria treat fatigue as a crack growth problem from an initial flaw size, using Kitagawa-Takahashi diagrams or Murakami's √area parameter to estimate the threshold below which a flaw is harmless.

For bone lattices, this is both the most physically honest and the most demanding approach. You need to know the defect population — its size distribution, location, and orientation — which requires X-ray computed tomography on every batch, or at least on statistically representative coupons. The method shines when you have a clear critical strut, like the thinnest ligament in a unit cell. It stumbles when defects are distributed and interact.

What usually breaks first is the assumption that a single defect governs failure. In a lattice, load can redistribute to neighboring struts, blunting the effect of one pore. But then the neighbor becomes overloaded, and the failure jumps.

So which do you choose? The answer is not in the data. The data has been sitting there for twenty years. The answer is in what you can measure, what you can simulate, and what risk you're paid to take. Start by asking which defect population you actually see in your parts. That narrows the field faster than picking the prettiest equation.

What the Right Criterion Must Do: Your Comparison Checklist

Accuracy vs. Simplicity: The Eternal Trade-off

A failure criterion that predicts every fracture to the decimal is worthless if it takes three weeks to compute. I have watched teams polish a high-fidelity model for months, only to abandon it because the design loop closed twice daily. The right method earns its keep by matching the physics that actually dominates your build—not by capturing every microstructural quirk. Stress-based criteria are cheap and familiar, but they miss the progressive damage that bone lattices exhibit. Strain-based methods capture more, yet they demand a strain field you may not trust. Energy criteria feel elegant, but they hide the failure location. Defect-based approaches are honest, yet brutally data-hungry.

The catch is that accuracy has a price.

Then again, simplicity has a hidden cost too. A criterion that ignores build-specific defects will look fine in your validation samples and fail on the production floor. The question is not which method is truest—it's which error you can live with. Wrong in the conservative direction? Wrong in the wrong direction? That distinction changes everything. A simple criterion with a generous safety factor may cost you mass and material, but a complex one that mispredicts by 15% could cost you a certification.

Data Requirements: How Much Calibration is Enough?

Every criterion needs calibration. Some need a handful of tests; others need a statistical army. Stress-life approaches typically require S-N curves across multiple stress ratios and build orientations—doable, but that's already a dozen specimens per condition. Strain-based methods push you into cyclic strain-controlled tests, which are slower and more expensive. Energy criteria need hysteresis data, and defect-based models demand CT scans or surface roughness distributions from every new batch of powder. The real question: does your supplier change powder suppliers mid-project? Because that invalidates your calibration overnight.

We fixed this once by embedding a verification coupon in every build.

Not the whole lattice—just a small fatigue witness specimen. When the batch shifts, the coupon tells you before your critical part fails. That's the kind of pragmatic move that turns calibration from a one-time chore into a live sensor. Most teams skip this. They calibrate once, pray, and move on. The data requirement is not just about how many tests you run—it's about how long that calibration stays valid. If your process drifts, your criterion drifts with it, silently.

Practicality in a Design Loop: Speed and Integration

A criterion that lives in a spreadsheet will get more use than one that requires a supercomputer. The design loop for bone lattices is iterative: you tweak strut diameters, re-run the simulation, check fatigue life, repeat. If the fatigue check takes longer than the simulation itself, engineers will skip it. That's the brutal reality. The best criterion is not the most accurate—it's the one that returns an answer before the coffee cools.

What usually breaks first is not the math; it's the handoff.

The person who writes the fatigue analysis is rarely the person who builds the CAD model. If your criterion can't be embedded into the simulation pipeline—as a user-defined material model, a post-processing script, or a lookup table—it will die in a PowerPoint deck. I have seen this happen more times than I can count. The analysis gets done once, for the final design, and never again. That's not a fatigue criterion; that's a post-mortem report.

Transferability is not a luxury—it's the difference between a research tool and an engineering standard.

— field note, lattice design review

Transferability Across Builds and Materials

Here is where most criteria stumble. You validate on one machine, one powder lot, one build angle—and then the vendor sends a new heat. Does your criterion still hold? Stress-life curves are notoriously sensitive to surface roughness and residual stress, both of which shift with every process tweak. Strain-based methods fare slightly better, but only if the strain fields are geometry-independent. Energy criteria are promising here because they scale with the hysteresis loop, which tracks the actual damage process rather than the nominal load.

The tricky bit is that defect-based models are the only ones that genuinely transfer—if you know the defect population.

And you rarely do. A build that looks identical under an optical microscope can hide subsurface pores that change the fatigue life by a factor of three. The pragmatic answer is not to demand perfect transferability but to build a criterion that flags when it's losing validity. That means embedding uncertainty bounds, running periodic verification coupons, and being honest about the envelope of your calibration. Wrong order? Change the test plan first, then the criterion.

Your next action is simple: take three fatigue data sets from your own lab, apply each criterion, and see which one fails gracefully when the data is thin. That exercise takes a day. It will tell you more than any literature review. Do it before your next design review.

Trade-offs at a Glance: Accuracy, Cost, and Confidence

A Side-by-Side Comparison Table

Four roads, four price tags. Stress-based S-N curves are the old workhorse—cheap to run, easy to fit into standard fatigue software, but they ignore the porous architecture that makes bone lattices what they're. Strain-based criteria handle large deformations better, yet they drift when your lattice locally buckles. Energy-based methods capture damage accumulation more faithfully; they also demand nonlinear FEA runs that eat weekends. Defect-driven approaches look at actual strut flaws, which feels honest, but every scan and reconstruction adds cost and operator judgment. That sounds fine until you price out a full campaign.

Here is the blunt trade-off matrix. Accuracy climbs as you move from stress to strain to energy to defect-based—but so does expense. Stress-based screening gets you an answer in days; defect-based validated predictions can take months. Confidence is messier. A low-cost criterion gives you false confidence when the loading is multi-axial or the lattice has manufacturing scatter. High-fidelity methods reduce that risk, but only if your defect statistics are trustworthy. Garbage in, gorgeous output.

Shortcuts cost a day.

Flag this for materials: shortcuts cost a day.

Most teams skip this: cost is not just money. It's calendar time, compute time, and the patience of your review board.

When Simplicity Wins: Quick Screening

For early design sweeps, stress-based criteria still dominate. I have seen labs rank ten lattice topologies in a week using nominal stress amplitudes, no FEA beyond a single unit cell. Wrong for final certification, right for killing bad ideas fast. The catch is that screening criteria bias you toward thicker struts and stiffer topologies—conservative, predictable, and sometimes exactly the wrong shape for your load case. Use them to sort, not to promise.

Strain-based criteria fit a middle niche. They handle large deformations and soft lattices where stress concentrates vanish into geometric nonlinearity. The accuracy gain is modest, the cost bump is modest, and the interpretation stays intuitive. For lattices above 15% relative density, strain criteria often straddle the sweet spot between cost and confidence. Below that, defects start to matter more than any continuum measure.

When High-Fidelity Is Worth the Expense

Energy-based and defect-driven methods earn their keep when fatigue life predictions feed into medical devices or load-bearing aerospace brackets. That sounds like a cliché, but the logic is concrete: one unexpected strut failure in a femoral implant revision costs more than the entire simulation budget. Defect-based approaches, especially those seeded from micro-CT scans of as-built parts, catch the real killers—surface roughness, lack-of-fusion voids, notched strut necks. Energy-based criteria capture how damage redistributes across the lattice as struts crack and shed load. Both are overkill for a concept review.

The odd part is—when teams do invest in these expensive criteria, they often discover the manufacturing scatter dominates everything.

Your scan resolution, build orientation, and post-processing all shift the fatigue life more than any criterion choice. So the high-fidelity path only pays off if you freeze the process first.

The best criterion is the one that survives contact with your actual build quality. Not the one that looks prettiest in a journal.

— a fatigue engineer, mid-project, after three unexpected failures

What usually breaks first is the assumption that your nominal geometry represents what sits on the build plate. Defect-based methods force you to confront that gap. Energy methods let you track its consequences. Both cost you time—but they convert unknown unknowns into measurable margins. That's the real currency.

Choose based on which failure mode you can afford to miss. Then pick the cheapest criterion that catches it.

From Choice to Practice: Steps to Implement a Criterion

Step 1: Identify Your Load Case and Critical Location

Start with the part, not the math. Walk the bone-lattice component through its real service cycle and ask where it actually breaks. Fatigue cracks don't care about your aesthetic unit cell—they nucleate at strut junctions, near build defects, or where the stress field necks down. I have seen teams spend weeks calibrating a criterion against uniform compression data, then watch the hip implant fail at a screw hole nobody modeled. Map the hot spot physically before you open any solver.

Wrong order kills the whole effort.

That sounds fine until you realize the load case is rarely a single amplitude. Physiological loading on a femoral scaffold jumps between walking, stair climbing, and occasional impact. Pull the duty cycle from your application, bin it into equivalent blocks, and decide whether your criterion handles sequence effects or just ignores them. Most do the latter. That's acceptable only if your safety factor eats the error.

Step 2: Calibrate with Physical Tests

No simulation replaces a test coupon that actually snaps. Machine a batch of representative lattice specimens—same strut diameter, same relative density, same print orientation—and run constant-amplitude fatigue tests at three or four stress levels. Plot the S-N cloud, fit your chosen criterion's parameters, and stare at the scatter. Bone lattices from powder bed fusion scatter like a shotgun; a 20% life spread at one load level is routine. That scatter is your calibration uncertainty, not a nuisance.

The catch is matching the test to the model's variables.

If your criterion is strain-based, you need extensometry, not just force readings. Energy-based? Track hysteresis loops. Defect-based? Count pores via micro-CT before testing, then correlate failure sites. Mixing units between your physical calibration and your simulation input creates a silent error that propagates straight into your prediction. We fixed this by logging every test artifact's geometry and defect population alongside the fatigue data, so the model and the measurement speak the same language.

Step 3: Build a Simulation Model

Now translate the criterion into a finite element routine. Mesh the lattice at a resolution that captures strut bending—coarse solid elements smear the strain gradients and flatter your life estimate. Run the load cycle, extract the metric your criterion demands, and map it onto a fatigue damage field. This step exposes the practical gap between the paper's formula and your software's capabilities. Some criteria need custom subroutines; others fit neatly into built-in fatigue modules. Budget for the coding time.

Most teams skip mesh sensitivity checks.

That hurts when the prediction flips sign between mesh densities. Run two meshes, compare the damage contour, and confirm your hot spot doesn't move. Also check your boundary conditions—clamping a lattice block rigidly at one end artificially stiffens the first layer of struts and shifts failure inward. Add a compliant interface if the real part sits in a polymer sleeve or metal endplate.

Step 4: Validate Against a Pilot Batch

Final gate: print a small pilot batch of actual components and test them under the full duty cycle, not just constant amplitude. Compare the predicted failure location and life against the physical outcomes. The first pilot usually misses—calibration coupons are cleaner than production parts, and your load case simplification hides sequence effects. That miss is gold. It tells you whether your criterion's error is a constant offset you can fold into a safety factor or a structural flaw in the model itself.

Validation is not a rubber stamp. It's the moment your criterion earns its keep or gets thrown back.

— fatigue engineer, design review

Don't chase a perfect match. A factor-of-two life prediction with correct failure location beats a factor-of-ten error that says the wrong strut breaks. Log the discrepancy, set a conservative knock-down factor, and move forward. The next component design gets better data because you documented where the prediction bent.

Then implement the criterion in your design workflow as a gating check, not a post-mortem tool. Run it early, rerun it after any geometric change, and feed the pilot results back into the calibration set. That closes the loop. Your next build inherits a slightly sharper criterion, and the fatigue void shrinks one step at a time.

Shortcuts cost a day.

When the Prediction Goes Wrong: Risks of Skipping the Criterion

Unexpected Fracture: The Cost of Overconfidence

The implant passes every static test. Then, on day 47 of a 90-day trial, it snaps. Not at the strut junction you modeled, not at the surface where you applied the load, but somewhere deep in the lattice where stress was supposed to be a third of yield. That's what skipping a fatigue criterion buys you: a failure no simulation warned about and a surgeon explaining to a patient why the revision is needed.

Flag this for materials: shortcuts cost a day.

I have watched this happen more than once. The team had invested months in geometry optimization, printed beautiful specimens, and validated stiffness to within 4% of the model. The fatigue data was the spreadsheet nobody opened. The catch is that static margins never translate into cyclic life for lattice structures. Micro-porosity clusters, surface notch effects, and ligament bending create stress concentrations that multiply with each load cycle.

Wrong criterion, same outcome. A team that picks the stress-based S-N curve from solid titanium bulk material feels safe. That feels like precision. But bulk data assumes a defect-free interior and polished surface. Your printed strut has a rough as-built finish and internal voids you have not characterized. The fatigue limit drops by 70–80% compared to bulk, and no one accounts for it until the fracture surfaces come back from the lab.

That hurts.

Development Delays and Regulatory Scrutiny

The first sign of trouble is not a fracture—it's a delay. Your prototype fails at 200,000 cycles instead of the required 2 million. Now you redesign, reprint, re-test. Each iteration costs three weeks and a block of machine time. The budget bleeds quietly. Then the regulator asks for the fatigue rationale, and you hand them a plot of static compression tests. That doesn't end well.

“Fatigue prediction is not a refinement. It's the gate between prototype and patient.”

— observation from a senior biomechanics reviewer, not a formal study

Regulatory bodies increasingly expect a documented fatigue criterion with traceable assumptions. If your report says “factor of safety in yield strength,” the reviewer knows you avoided the cyclic question. The response letter will list the gap, request additional testing, and your timeline stretches by another quarter. The trade-off is brutal: investing in a criterion upfront costs days; skipping it costs months when a reviewer forces the issue.

The odd part is that the fix is often simple—a strain-life approach calibrated on as-built specimens, or a defect-tolerant model using your actual porosity distribution. But implementing it after the fact feels like an admission of sloppiness. It's not. It's engineering maturity, and it's cheaper before the rejection letter arrives.

The Silent Risk of Using Bulk Material Data

Most teams don't deliberately ignore fatigue. They default to bulk data because it's available, measured, and published. That's the silent trap. Bone lattice struts, typically 100–500 µm in diameter, have a surface roughness that dominates their fatigue behavior. The crack initiation site is not a material flaw—it's the geometric notch of the strut surface itself. Bulk data has no mechanism to capture that.

Consider the strain-life approach. If you plot total strain amplitude versus cycles to failure for solid titanium, then repeat for lattice struts at the same relative density, the curves diverge by an order of magnitude at low cycles. The damage mechanism differs: bulk fails by transgranular crack growth, lattice fails by local ligament ratcheting and micro-crack coalescence. You can't map one onto the other with a simple correction factor.

We fixed this in one project by printing dogbone specimens from the same batch as the lattice, testing them in axial fatigue, and then correlating the local strut strain from micro-CT digital image correlation. The effort took eight days. The alternative was accepting a 2× uncertainty in life prediction. I know which I sleep better with.

How to Redesign When Fatigue Fails

When fatigue fracture does occur, resist the urge to thicken every strut. That's the blunt instrument, and it increases stiffness, alters load-sharing, and can shift failure to the adjacent bone—worse for the patient. Instead, localize the problem. Map the fracture origin, measure the local strain history, and ask why that strut carried more load than modeled.

Often the answer is boundary condition error, not material weakness. The lattice connects to a solid frame, and the transition zone creates bending moments you idealized away. Redesign the transition, add a gradient of strut diameter, or change the node connectivity. One targeted modification beats a global thickening. The process is iterative, but each cycle teaches you what the criterion should have predicted—and you can fold that insight back into the model.

Do that. If you have no criterion, the fracture is a surprise. If you have a criterion, the fracture is a data point—and usually a fixable one. The next revision then has a credible fatigue rationale, the regulator has less to argue about, and the patient gets an implant with a known—not guessed—life.

Frequently Asked Questions on Bone-Lattice Fatigue Prediction

Can Bone’s Remodeling Be Modeled in Fatigue Tests?

Short answer: not yet, and pretending otherwise will sink your validation budget. Bone adapts—it repairs microcracks, redistributes load, and changes stiffness week to week. Your lattice specimen does none of that. The best you can do is mimic the *end state*: a trabecular-like architecture that carries load predictably. But fatigue tests run on a fixed geometry, fixed boundary conditions, and a fixed stress field. That disconnect matters more than most engineers admit. I have seen teams run 10⁶ cycles on a lattice, get a clean S-N curve, and then watch the same part fail at 10⁴ cycles in a cadaver test—because the surrounding tissue stiffened or softened the interface, shifting the load path entirely.

That hurts.

What you *can* model is the degradation envelope—the rate at which stiffness drops as cracks accumulate. It won’t capture biological repair, but it gives you a conservative lower bound. The catch is that conservative bounds often over-design, adding strut thickness that reduces porosity and kills the osteoconductive surface. So the trade-off is real: biological fidelity versus clinical practicality.

Why Do S-N Curves Vary So Much Across Machines?

Three culprits, in order of sneakiness: load alignment, environmental control, and specimen clamping. A misaligned actuator by even 0.5 degrees can change local stress at the strut junctions by 15–20 percent—that’s not noise, that’s a different failure mode. Then there’s the bath. Bone lattices are usually tested in phosphate-buffered saline at 37°C, but if the pump recirculates too fast, you get localized drying or temperature gradients. I’ve watched identical specimens from the same build batch produce fatigue lives that differed by a factor of three—not because the material varied, but because one machine’s grips had a slight tilt.

Standardize your fixtures. Don’t trust the machine’s built-in alignment report; verify with a strain-gauged dummy specimen before every campaign. That’s boring, but it beats rerunning four weeks of tests.

What’s the Most Practical Criterion for a Small MedTech Firm?

Start with stress-life (S-N) if you have any historical data on your alloy and build process. It’s cheap, fast, and regulatory reviewers have seen it a thousand times. But that’s also its weakness—S-N curves treat every strut as if it were a polished coupon, ignoring surface roughness from powder-bed fusion or microporosity at nodes. For a small firm, I would pair S-N with a single defect-sensitive check: measure the largest pore near a high-stress strut via micro-CT on a sacrificial part, then apply a simple Kitagawa-Takahashi diagram. That catches the worst cases without forcing you into full crack-growth simulations.

The strain-life approach is tempting—it handles plasticity better—but it demands cyclic stress-strain curves for your exact build parameters, which you likely don’t have. Energy-based criteria sound elegant, and they collapse some data nicely, but they obscure the physical mechanism. When a reviewer asks “why did this fail here and not there?”, energy plots don’t answer well.

Is There an Industry Standard in the Works?

Fragmented, but moving. ASTM F3450 covers additive titanium alloys, and ISO 17850 is circling bone scaffolds, but neither mandates a fatigue criterion—they specify test methods and reporting. The awkward gap is that you can follow both standards, generate perfectly reproducible data, and still have zero idea how to predict failure in a new geometry. That’s not a standards failure; it’s a physics gap.

What I expect in the next three to five years is a consensus on *defect population characterization* as a prerequisite. Not a single criterion, but a rule: you must quantify your largest flaw and compare it against a stress-intensity threshold. That would shrink machine-to-machine scatter and make S-N curves portable across labs.

Until then, the practical move is to build your own internal checklist—pick one primary criterion, validate it on three geometries, and publish your setup openly. The firms that share their fixture drawings and load protocols are the ones whose data others actually trust. That reputation pays off faster than any refined model.