Using FEA on a Machine Base Without Fooling Yourself

Which load cases, boundary conditions and modal checks matter when analysing a machine base, and the modelling shortcuts that quietly produce confident wrong answers.

FEA on a machine base is straightforward to run and easy to get confidently wrong. The software will always produce a colourful picture and a number. Whether that number describes your machine depends on a handful of decisions made before the solve.

What you are actually asking

Three questions, and they want different setups.

1. Static deflection. How far does the tool move relative to the work under cutting load?
2. Modal behaviour. Where are the natural frequencies and what shape are the modes?
3. Local stress. Is anything overstressed at a detail?

For a machine base the first two dominate. Machine bases are rarely stress-limited; they are almost always stiffness- and dynamics-limited. If your analysis is mostly producing stress plots, you are probably answering the wrong question.

Boundary conditions are where studies go wrong

This is the single biggest source of confident nonsense in machine base FEA.

Do not fully fix the feet. A base that is encastré at every foot is far stiffer than anything that exists. It will pass any deflection target and tell you nothing.

Model the support the way the machine will actually sit:

• On levelling screws or wedge mounts, model the contact points with realistic stiffness, not infinite.
• On isolators, use the isolator’s published stiffness, which is deliberately low and will dominate the low-frequency modes.
• Grouted to a foundation, the support is genuinely stiff, and you can justify something close to fixed.

Three points versus six changes the answer dramatically, and it is a decision the installer often makes rather than the designer. If you do not know, run both and see how sensitive your design is. A base whose performance depends on which of two plausible support schemes is used is a fragile design regardless of what the best case says.

Load cases worth running

• Self weight, including everything bolted to the base. The machine is heavy and gravity is always on.
• Maximum cutting force at the worst tool position, not a convenient central one.
• Moving axis at each end of travel, if a carriage or gantry shifts mass across the structure. The worst case is usually at an extreme, not the middle.
• A torsional case, because twist is what limits a great many bases and a purely vertical load case will never show it.
• Lifting and transport, if the base will be craned by specific points. This one occasionally finds the highest stress in the whole model.

Measure the output that matters: relative displacement between tool point and work point, not absolute deflection of a node. The absolute number includes rigid body motion you do not care about.

Modal analysis

Modal analysis is cheap to run and it answers the question a static study cannot. Run it. It is cheap and it answers the question a static study cannot.

Look at the first few mode shapes, not just the frequencies. The shape tells you where to put material. A first mode that is the whole structure twisting says triangulate the ribbing. A first mode that is one large panel drumming says close up the rib spacing there and nowhere else.

Then check the frequencies against the machine’s actual excitations: spindle speed range, tooth passing frequency, servo bandwidth. Separation from those is the requirement; a round number is not.

FEA will not give you real damping. It can be assigned, but unless you have measured your structure you are assigning a guess. Modal FEA tells you where the modes are. How much they ring is a question for a tap test on the first unit.

Modelling choices that are fine, and ones that are not

Fine:

• Treating a welded structure as continuous for stiffness and modal work.
• Shell elements for plate structures. Faster and usually better behaved than solids.
• Removing small holes, chamfers and fillets away from areas of interest.
• Simplifying bolted-on masses to point masses at the right location.

Not fine:

• Fully fixed boundary conditions.
• One mesh density and no convergence check.
• Quoting a peak stress at a sharp re-entrant corner. That is a singularity; refine the mesh and it grows without limit.
• Analysing the base alone when the column and ways are what actually close the loop between tool and work.
• Believing the third decimal place of anything.

Convergence, briefly

Run the model at two or three mesh densities and watch deflection and first natural frequency settle. If halving the element size still moves the answer by more than a few percent, you are not converged and the number you were about to put in a report is not real.

This takes one extra afternoon and it is the difference between analysis and decoration.

Validate on the first unit

Whatever the model said, measure the real thing. A tap test with an instrumented hammer and an accelerometer on the finished base takes an hour and gives you the actual frequencies and the actual damping, including every joint, weld and support condition the model approximated.

Compare with the model. If they agree, you can trust the model for the next design. If they do not, you have learned something considerably more valuable than a passing analysis report.

Set your targets first using stiffness and deflection targets, and read natural frequency and chatter for what the modal results mean in the cut.

Common questions

← All guides