The Simulation Nobody Believes Until the Machine Breaks
Here’s a conversation I’ve had too many times. The customer sees a bolt pattern on a weldment and says “that looks thin, can you run FEA?” I say sure, I’ll run it. They expect a colorful picture with a safety factor of 3.0 and a green bar that says “approved.” Instead, I come back with a plot showing the frame deflecting 1.2 mm under full load, a stress hotspot in a gusset nobody noticed, and a recommendation to add two more ribs.
The reason most FEA structural analysis for machine frames disappoints people is that they treat simulation like a verdict instead of a conversation. A finite element model is not a crystal ball. It’s a numerical experiment that tells you where to look — and if you set it up wrong, it will tell you exactly what you asked it to, which is almost never what you actually wanted to know.
This article is the checklist I run on every structural FEA job before I press “solve.” It’s not a textbook on finite element theory. It’s the list of mistakes I’ve made enough times that I now check for them before the solver starts.
Before You Mesh: The Decisions That Determine Whether Results Mean Anything
Most FEA errors don’t happen in the solver. They happen in the model setup, before the first element is created. If you get these wrong, the prettiest stress plot in the world is worth less than a hand calculation.
What Are You Actually Solving For?
“Analyze the frame” is not an objective. You need to answer a specific question:
- Does the frame deflect less than X mm under maximum load? (Stiffness question)
- Is the stress below the material yield divided by a safety factor? (Strength question)
- Does the natural frequency avoid the driving frequency? (Modal question)
Each question drives a different analysis type. A linear static analysis answers the first two. A frequency analysis answers the third. Trying to answer all three with one model is how you end up with a stress plot that doesn’t tell you anything useful.
Boundary Conditions: The Place Where Good Models Go to Die
How you constrain the model is as important as the geometry itself. Fixing every bolt hole face to “fixed” (zero translation, zero rotation) makes the model stiffer than the real frame. A weldment on levelling pads is not welded to the floor — it sits on four feet, and those feet can rotate slightly. A model that fixes all six degrees of freedom at every bolt hole will show lower stress and less deflection than the real machine, and you’ll find out the difference when the frame wobbles on the shop floor.
My default boundary conditions for a weldment:
| Condition | What It Represents | When to Use |
|---|---|---|
| Fixed on one foot (all DOF) | The foot that doesn’t slide | Always — prevents rigid body motion |
| Roller on other feet (normal direction only) | Feet that can slide/rotate on the floor | Standard for floor-standing machines |
| Symmetry restraints | If the geometry and load are symmetric | Saves mesh, but only if the load is truly symmetric |
Loads: Apply Force Where It Actually Enters the Structure
Point loads are a last resort. A 500 N force on a 1 mm point creates a stress singularity — the mesh refines to zero, the stress goes to infinity, and you learn nothing about the real structure. Apply the load over the actual contact area: the bolt pattern, the mounting face, the rail footprint.
For a machine frame, the loads that matter are usually:
- Station weight — the sub-assembly sitting on the frame, applied to its mounting holes.
- Process forces — press forces, clamping forces, reaction forces from actuators. These are often directional and may apply only during part of the cycle.
- Operator loads — someone leaning on the guarding, reaching into the station, stepping on the frame.
- Inertial loads — if the frame carries a moving mass (gantry, rotary indexer), the acceleration forces matter. Use g-loading equivalents, not just static weight.
Meshing: Good Enough Beats Perfect
A fine mesh everywhere is a waste of compute time. A coarse mesh everywhere is a waste of the analysis. The trick is to put the nodes where the gradient is steep.
Start Coarse, Refine Where It Matters
I always start with a default mesh, solve it, and look at the stress plot. Where the stress is high (red/yellow zones), I refine the mesh locally — usually to a size that puts at least 3–4 elements across the smallest section in that hotspot. Then I re-solve. If the stress changes by more than 10% between the coarse and refined mesh, I refine again. That’s mesh convergence in practice: refine until the answer stops changing.
Watch Out for Singularities
A sharp internal corner with a load on it will show a stress peak that goes to infinity as the mesh refines. That’s not a real stress — it’s a mathematical artifact. Real fillets, welds, and contact surfaces spread that peak over a small area. If your stress hotspot is at a sharp re-entrant corner, don’t trust the absolute number. Add a fillet in the model (even a small one) and re-run. If the stress drops by 40% with a 2 mm fillet, that’s the real answer.
Practical rule: If the stress at a hotspot changes by less than 10% when you refine the mesh, the result is trustworthy. If it keeps climbing as the mesh gets finer, you’re looking at a singularity, not a real stress.
Material Properties: Don’t Default to Whatever the Drop-Down Says
Software defaults are convenient. They’re also wrong more often than you think.
For weldments, the default “Structural Steel” in most FEA packages uses a yield strength of 250 MPa (36 ksi) and a modulus of 200 GPa. That’s fine for A36 or S235. But if you’re using S355 or Q355, the yield is 355 MPa, and your safety factor calculation is off by 40%.
Two things people forget:
- Welds are not the same as the parent material. A fillet weld’s allowable stress is lower than the plate. If you’re analyzing a welded joint, use the weld’s allowable (typically 0.5–0.6 times the parent material’s shear strength), not the parent plate’s yield.
- Modulus doesn’t change much. E is 190–210 GPa for almost all steels. Don’t lose sleep over it. But yield strength varies with grade, thickness, and heat treatment. Check the actual material spec you’re buying.
Reading Results: What to Look At (and What to Ignore)
A stress plot with a single red dot at a bolt hole is not a design failure. It’s a stress concentration. Here’s how I triage results:
The Three Plots That Actually Matter
- Displacement plot. Where does the frame move the most? Is it in the direction that affects tooling accuracy? A 0.5 mm deflection at the table surface is fine for a general assembly station. It’s a problem for a vision station that needs 0.05 mm repeatability.
- Stress plot (von Mises). Where are the hotspots? Are they in regions with actual loads, or at singularities (sharp corners, point loads)? Compare the peak stress to the allowable, not to yield directly.
- Safety factor plot. The software calculates this as yield divided by stress. But remember: it uses the material yield you gave it and the stresses you calculated. If either input is wrong, the safety factor is a fiction.
Safety Factor: What’s Actually Safe?
The answer depends on what you’re doing. I use these starting points:
| Application | Minimum Safety Factor | Why |
|---|---|---|
| Static frame, no dynamic load | 2.0 | Weld quality uncertainty, material scatter |
| Frame with moving masses / fatigue loads | 3.0 | Fatigue life, vibration, repeated cycling |
| Part-contacting tooling (hardened) | 1.5–2.0 | Material is controlled, contact area known |
| Safety-critical load path (lifting, guarding) | 4.0+ | Code requirements, human safety |
A Pre-Analysis Checklist That Takes Five Minutes
Before you hit solve, run through this list. I keep it as a sticky note on my monitor.
- ☐ Objective stated in one sentence (deflection? stress? natural frequency?)
- ☐ Material properties match the actual spec (not the default)
- ☐ Boundary conditions represent real mounting (not “fixed everywhere”)
- ☐ Loads applied over contact areas, not at points
- ☐ Worst-case load identified (not nominal load — the maximum the machine will ever see)
- ☐ Mesh convergence checked at hotspots
- ☐ Singularities identified (sharp corners, point loads) and not used as the design number
- ☐ Result cross-checked against a hand calculation (rough deflection estimate, simple beam formula)
That last item is the one that saves you from embarrassing results. A quick hand calculation — even a rough cantilever beam estimate — will tell you if your FEA result is in the right ballpark. If the hand calculation says 0.3 mm deflection and the FEA says 12 mm, something is wrong with the model. Don’t trust the software until it agrees with the physics.
When Simulation Isn’t Enough
FEA tells you what the model does. The real machine does what the real machine does. The gap between the two is where weld residual stress, manufacturing tolerances, bolt torque, and floor flatness live.
For machines that matter — high-precision stations, load-bearing frames, anything that feeds a process that costs money when it stops — I always recommend a post-install deflection check. A dial indicator on the table surface while the heaviest station cycles once tells you more than a week of simulation. The FEA gets you in the neighborhood. The dial indicator tells you the address.
The Bottom Line
Finite element simulation best practices aren’t about learning the software. They’re about not fooling yourself. Set up the model to match reality, mesh with intent, check convergence, and cross-check against a rough hand calculation. The stress plot is a tool for finding weak spots — not a verdict that lets you stop thinking.
The next time someone asks you to “run FEA on that frame,” spend twenty minutes on setup before you press solve. The results will be more credible, the design will be better, and you won’t have to explain to the customer why the frame they built from your “approved” model deflects twice as much as the simulation said it would.
