Three different movements, three different causes
People say a weld pulled and mean one of three things that have nothing in common except the cooling metal that caused them.
Transverse shrinkage is the two parts moving toward each other across the joint. The weld metal solidified at the melting point and then contracted several hundred degrees worth on the way to room temperature, and because it is bonded to both parts it takes them with it. The amount scales with how much weld metal there is and falls as the plate gets thicker, because a thicker plate is a bigger heat sink and a stiffer restraint. A wide root gap makes it worse out of proportion to the extra metal, because there is nothing to stop the gap closing until the root pass is in.
Longitudinal shrinkage is the whole member getting shorter along the weld. It is small in absolute terms, usually a few hundredths of an inch over a few feet, and it is the term people forget when a part comes out short and nobody can find the mistake in the cut list. It scales with the ratio of weld area to member area, so a small weld on a heavy section is negligible and a heavy weld on a light section is not.
Angular distortion is the one everybody has seen. It happens because the weld is not on the neutral axis of the thing it is shrinking. A fillet sits entirely on one side of the base plate, a single-V groove puts most of its metal in the top half of the joint, and in both cases the shrinkage acts at a distance from the mid-plane and therefore acts as a bending moment. That is the whole mechanism, and it is why a balanced double-sided weld distorts a fraction as much even though it contains more metal.
What the model actually does, and where it stops
This page takes the weld cross-section, works out where its centroid sits relative to the mid-plane of the member, and treats the product of the two as a bending moment resisted by the stiffness of the plate, which goes as thickness cubed. The transverse and longitudinal terms use the same two well-worn estimating relationships that fabricators have used on paper for decades: transverse movement proportional to weld area over thickness, longitudinal proportional to weld area over member area times length.
What the model does not contain is everything that makes distortion a craft rather than a calculation. Welding sequence changes the answer substantially: back-step welding, skip welding and alternating ends all reduce distortion without changing a single number this page uses. Interpass temperature changes it. The order the tacks went in changes it. Whether the plate was flat and stress-free before you started changes it, and hot rolled plate rarely is. Two identical joints welded by two people on the same day will not move the same amount.
So the honest description of the output is that it tells you the order of magnitude and which of the three movements will dominate. That is genuinely useful, because it tells you whether to pre-set, whether to balance the joint, or whether to stop worrying about it. It is not a prediction and it should not be treated as one.
The angular coefficient, and why it is an input
The default of 0.45 puts a quarter inch fillet welded one side on half inch plate at roughly two degrees, and a single-V groove in half inch plate at roughly the same, both unrestrained. Those are plausible figures for typical shop conditions in mild steel. They are not universal, because the coefficient is absorbing everything the model does not represent: material, sequence, heat input, cooling rate and how well the fit-up was.
The right way to use it is to calibrate. Weld a joint, measure what it actually did with a square and a feeler gauge, and adjust the coefficient until the page reproduces your result. From then on it predicts your shop rather than a generic one. That is the same discipline as measuring a K-factor on a test coupon before cutting a flat pattern, and it works for the same reason: an empirical coefficient is only worth what the measurement behind it is worth.
Pre-setting, and the mistake people make with it
Pre-setting means deliberately assembling the part out of position by the amount you expect it to move, so that it moves into position as it cools. For an angular pull of two degrees on a joint that should finish square, you set it up at 92 degrees. The page reports it both as an angle and as a rise measured at a stated distance from the joint, because a rise is what you can actually measure with a rule and a straightedge on a table.
The mistake is treating pre-set as a substitute for balancing. Pre-set corrects the shape and does nothing about the residual stress, which stays in the part and can come out later when it is machined, cut, or heated. On anything where dimensional stability over time matters, balancing the weld about the neutral axis is a better answer than compensating for an unbalanced one. Pre-set is the fix when balancing is not available, which on a fillet welded from one side because that is the only side you can reach, it usually is not.
The connected pages
Weld cross-section drives everything on this page, and it is the same number the joint preparation calculator produces when you compare a single-V against a double-V. That comparison is where the distortion argument for a double-V gets made in numbers: it halves the fill and nearly cancels the angular term, at the price of turning the part over. For the money side of the same geometry, the weld cost calculator puts a price on the metal, gas and time.
Thermal movement while the part is still hot is a different problem from the permanent set left behind, and if you are fitting a hot assembly to a cold one, the thermal expansion calculator is the page for that.
What this page will not do
It will not tell you how big the weld should be. Weld size for anything that carries load, restrains load, or lifts anything at all is an engineering specification, produced by someone qualified to produce it, and executed by a welder qualified for that joint in that position. Inspection of the result is a third discipline again, with its own qualifications. Lifting points, trailer hitches, vehicle suspension and structural connections are all in that group, and a calculator that reported a weld capacity for them would be doing real harm.
So this page takes a weld size you have already been given, or that you have chosen for a non-critical job, and works out what follows from it. If nobody has given you a size and the failure of the joint would hurt somebody or cost a lot of money, that is the signal to stop calculating and get the specification.
The hazards of the work itself are named in full on the travel speed calculator, and they apply here without change.
Questions people ask
Can I just weld it and straighten it afterwards?
Sometimes, and it is common practice, but understand what you are choosing. Mechanical straightening in a press works on thin material and light sections and leaves the residual stress largely where it was. Flame straightening works by adding a controlled second distortion that opposes the first, and it is a genuine skill that takes practice to do without making things worse or damaging the material. Both are more work than pre-setting, and neither is available on an assembly that is already welded into something bigger. If you know the part will pull and by roughly how much, allowing for it during fit-up is nearly free and straightening it afterwards is not.
Why did my part move after I took it out of the fixture?
Because the fixture was holding the shape, not removing the cause. Weld metal contracts as it cools whether or not it is allowed to move. If it is allowed to move, you get distortion. If it is restrained, you get residual stress instead, locked into the part at a level that can approach the yield strength of the material. Releasing the clamps releases some of that, and the part springs to a new equilibrium. Machining it releases more, because cutting metal away removes the balance of stresses that was holding the shape. This is why heavily restrained weldments sometimes need a stress relief step before final machining, and why a part that measured perfectly in the jig can be out of tolerance on the bench.
Does a smaller weld really distort less?
Yes, and the relationship is stronger than people expect because the metal in a fillet goes as the square of the leg size. Stepping a quarter inch fillet up to five sixteenths adds 56 percent more weld metal, which is 56 percent more shrinkage force, 56 percent more time, and 56 percent more filler. Oversizing fillets is the single most common self-inflicted distortion problem in general fabrication, and it usually comes from a well-meant instinct that bigger is safer. On a weld whose size was specified by somebody qualified to specify it, bigger is not safer, it is different from what was designed. If nobody specified it, that is a signal to read the section below about what this page will not do.
Does welding sequence matter more than the numbers here?
Often, yes. Back-step welding, where you work in short runs each laid in the opposite direction to the overall progression, and skip welding, where you leave gaps and fill them later, both reduce distortion substantially without changing weld size or joint geometry at all. Alternating from end to end and from side to side does the same. None of that appears in this calculation, which is one of the reasons the output is an estimate rather than a prediction. If a part is distorting more than you can live with and the geometry is fixed, sequence is the first place to look, and it costs nothing but planning.
Do stainless and aluminium move more than steel?
In general yes, and for different reasons. Austenitic stainless has a coefficient of thermal expansion around half again that of carbon steel and much lower thermal conductivity, so the heat stays local and the expansion per degree is larger, and the distortion is noticeably worse for the same joint. Aluminium expands roughly twice as much as steel per degree but conducts heat away very quickly, so the heat spreads and the picture is more complicated. The coefficient on this page is calibrated for carbon steel, so for other materials the sensible approach is to calibrate it yourself on a test joint rather than to apply a correction factor from memory.