Why a 9‑inch rail comes off the mill bent A stress simulator for barstock, built around the ARCA rail job. Every number on this page is computed live from the inputs on the left — change them and the whole page updates.

The short version: the bar was already full of locked‑in force before we touched it. We take off one face and off the other, which unbalances that force about as hard as it can be unbalanced, and the bar bends to rebalance itself. Nothing is wrong with the machine, the tool, the program, or the operator. The fix is the order and the symmetry of the cuts, plus what we write on the purchase order.

1The bar arrives with a tug of war inside it

Extruded, rolled, drawn and quenched stock does not come out of the mill stress‑free. When it was made, the outside cooled or got worked at a different rate from the inside. The layers near the faces end up squeezed and the layers in the middle end up stretched — or the other way round — and they have been pulling against each other on the rack ever since.

You cannot see it, feel it, or catch it with a micrometer, because the forces cancel. Add up all the pull across the cross‑section and you get zero. Add up all the bending and you get zero. That is why the bar is straight. A straight bar is not evidence that there is no stress in it. It is evidence that the stress is balanced.

The stock we are using, through its thickness

2We do not add stress. We break the tie.

The cutter does not put the bend in. It deletes one of the teams. Take off the top face and every pound of pull that layer was contributing simply leaves the part. What is left is unbalanced, and an unbalanced bar can only do two things:

• the leftover pull makes it a hair shorter or longer — about on our part, which nobody will ever notice;
• the leftover bending makes it bow — which we notice on every part.

It is the same effect as slitting cold‑rolled strip on the bandsaw and watching the cut open up, or a weldment pulling sideways when you saw a piece off it. This is old, settled metallurgy; the shop‑floor measurement that proves it is called a layer‑removal test, and it is exactly what this simulator runs, backwards.

Same stock, same total metal removed, three different splits

3Our split is the worst one available

Here is the part that matters for the process. For a given amount of metal, how much you remove barely matters. Where you remove it from is everything.

Taking off one face and off the other is, as far as the bar is concerned, nearly identical to taking the whole off one side. We pull of our stock out of one side of the neutral axis. That is close to the maximum imbalance the job allows.

Split the same metal evenly — a side — and the two released moments are equal and opposite. They cancel. The part still gets a touch shorter, and it still has its stress removed, but there is no net bending left to bend it.

Bow vs. how we split the same total removal drag the removal numbers on the left and this moves with them

4Why we only skim 0.010" first — and how to keep that fixture

The first op is a light skim for a good reason: the Pitbull clamps need stock at height to grip, and we need a flat to pull down against. I am not proposing we change the fixture. I am proposing we stop finishing the job in one direction.

Adding one op buys back the symmetry without touching the workholding. Every operation still has a flat to clamp, and the part never gets a one‑sided bite bigger than about half of the total.

Two numbers on those cards, because they are two different failures. Bow is how banana‑shaped the part is, and it depends only on the net split between the faces — which is why a balanced two‑op route and a balanced five‑pass route end up at the same bow. Out of flat on the machined face is what a straightedge laid on the finished surface reads, and that is what the extra passes buy: cut it last, light, and barely clamped, and the face comes out flat even if the part itself has moved. Section 6 is where that comes from.

The planner on the Process Plan tab has both of these loaded, step by step, and it will flag any operation that leaves the part below the your clamps need.

5The longer and thinner it gets, the worse it gets

Two scaling laws decide whether this is a nuisance or scrap:

• Bow grows with length squared. Our rail bows four times as far as a half‑length one from the identical cut. We are at the long end of the problem, and the rail length is set by the customer, so this one we just have to respect.
• Bow grows as one over thickness cubed. Same released moment, much less stiffness to resist it. Finishing at rather than costs us .

How bow scales with length and with finished thickness

6The clamp trap: Pitbulls hide bow, they do not remove it

This is the one that makes a part measure good in the fixture and bad on the bench, and it is why "it was flat when I checked it" and "it is bowed" can both be true.

Our clamps are stiff and they pull down. Suppose the bar has already moved after op 1. We drop it on the plate for op 2 and the clamps flatten it — elastically, like a leaf spring. We then machine a dead‑flat face on a bar that is being held bent. We unclamp, the bar springs back to the shape it wants, and it takes our beautiful flat face with it. The face is now curved by about the amount we clamped out, and the part is thinner in the middle than at the ends.

Clamp it flat, cut it flat, let it go ▶ Play

The simulator tracks this separately, because it is a different error from the bow itself. With the current route and pull‑down, it predicts of flatness error on the last face we cut and of taper between the two faces.

The fix is not stronger clamps — stronger clamps make it worse by hiding more. The fix is to have very little bow left to clamp out by the time the finishing pass happens:

• Do the clamping‑out on a roughing pass, never a finishing pass.
• Support the part along its length so the clamps only have to locate it, not bend it.
• For the last pass, side‑grip with the lightest pull‑down that will hold the cut. Slide the Clamp pull‑down control on the left to 0% and watch what happens to the face error.
• Leave enough stock to clean up the bow you expect. If the part will move , a finish allowance smaller than that will not clean up at all.

7Buy stock with less to give back

The other half of the fix is upstream of the machine, and it is the cheaper half. Bow is directly proportional to how much locked‑in stress arrives in the bar. Halve the stress, halve the bow, with the same program and the same fixture.

What to ask for, in order:

• 6061‑T6511 extruded bar. The 511 on the end means it was stretched after quench, which is a mechanical stress‑relief step. Usually the same price and lead time as plain T6 — you just have to ask for it by name.
• 6061‑T651 plate if the section suits us better than bar.
• Never plain 7075‑T6 for a long thin part. If we need the strength, 7075‑T7351 or 7050‑T7451 — that is what aerospace buys precisely because thin machined parts stay put.
• Avoid cold‑drawn or cold‑finished bar for anything that has to end up flat. It arrives the prettiest and behaves the worst: all that stress is concentrated in a skin a few thou deep, exactly where our first skim cut takes it.
• Put "stress relieved temper required — long thin machined part" on the PO and ask for the temper on the cert. A supplier substituting T6 for T6511 to fill an order is how this problem walks in the door.

The Stock & Materials tab has the whole list with typical stress levels, and clicking any row loads it into the simulator so you can see what the swap is worth on this exact part.

8Everything above is a model. Here is how we settle it.

This page runs real mechanics, but it runs them on typical stress numbers for each product form. The actual level in the lot on our rack is a number only a measurement can give us, and lot‑to‑lot spread of ±50% is normal. The good news is that the test takes about twenty minutes and the tooling is a bandsaw, a mill, and a surface plate.

Cut a drop to the same length, face a known amount off one side only, and measure how much it bowed. That single number pins the stress level for our lot, and from then on the simulator is predicting our material rather than a textbook average. Full procedure and the back‑calculation are on the Prove It tab.

Model: forward layer‑removal (Treuting–Read) analysis. The stock stress profile is built from a Legendre basis so it is self‑equilibrated by construction; each cut deletes layers, the surviving section's force and moment resultants are relaxed to zero, and the relaxed field is written back so the next cut sees it. Elastic, isotropic, uniform along the length, plane sections. It will not predict twist, it does not know about thermal gradients during the cut, and it treats clamping as elastic. The physics it does cover is the part that explains most of what we are seeing.