Home / Metal cutting / Milling
The finish is on the drawing. Start the calculation there.
Every other speeds and feeds calculator asks you for a feed and says nothing about the surface you end up with. That's backwards. Surface finish is specified before the process exists — so specify it here, and feed per tooth is back-solved from it instead of guessed at.
The finish grade is proposed from your operation conditions. Changing it picks the nearest ISO 1302 step and marks the value as a manual override, so a chosen grade is always visibly chosen rather than inherited.
Corner radius is fixed at 0.8 mm. Face mill inserts don't carry a true corner radius the way turning inserts do — the finish-generating geometry is a wiper chamfer. Until that chamfer is modelled properly this is a stated placeholder, not catalog geometry, so treat the finish figure as approximate.
Results
Starting values for good conditions, with a theoretical finish that is a floor rather than a prediction — real surfaces come out worse. Geometry inputs are the lead angle and a rake class — rake corrects force and power only, never speed or feed. Chipbreaker and edge preparation are yours to choose and are not modelled. Verify against your insert manufacturer's data and your machine before cutting.
Now find the speed that actually costs least
Cutting speed above comes from the material class. Whether it is the right speed depends on the Taylor exponent for your machine, your material and your grade — and that cannot be looked up. Two runs and some counting will give it to you.
RUN 1 — HIGHER SPEED
RUN 2 — LOWER SPEED
COSTS
Optimization
Both runs must use the same cutter and end at the same wear criterion. Speeds need at least 20% separation — two points close together will not fit a stable slope.
Why the driver variable matters
A driver variable is one you choose because a requirement demands it. The system then has to deliver it, and it belongs on the process specification — not in a note at the end of a cycle time report.
Finish is specified, not observed
Ra comes off the part drawing before any process exists. Treating it as whatever the cut happens to leave inverts the causality: you measure parts to find out what you built, instead of building to what was required.
Feed then stops being free
Once the finish grade is fixed, feed per tooth follows from it. The old rule still holds — take the largest feed that meets the finish — but it is a ceiling handed to you, not a parameter you tune for cycle time.
Which creates the real conflict
A fine grade forces a small feed, and light radial engagement thins the chip further. Push far enough and the edge is below the thickness it can shear, so it rubs. The calculation says so rather than staying quiet.
Where these numbers come from
Three things separate a usable starting point from a number that gets you a scrapped part.
Milling is not turning
The chip starts at zero on entry, peaks, and returns to zero on exit, so average thickness depends on how far the cutter is buried radially. Using feed times sine of the lead angle alone overstates it badly on light cuts, and that error propagates into every force and power figure.
Specific cutting force is not a constant
It climbs steeply as chips get thinner, because a real edge has a finite hone radius and the plowing penalty stays fixed while the chip shrinks around it. Thin chips cost disproportionately more energy per unit of material removed.
The rubbing floor belongs to the edge
Below roughly the edge hone radius the material deforms and passes under the edge instead of forming a chip. That floor is a property of the tool, not the workpiece — a sharp uncoated edge cuts thinner chips than a heavily honed coated insert.
Theoretical finish is a floor, not a prediction
The Ra this page reports is pure geometry: the cusp the corner leaves at a given feed. Nothing else is in it. Every real surface comes out worse, and the size of that gap is exactly what nobody publishes.
What the geometry misses
Built-up edge, runout, deflection under load, chatter, and the edge steadily wearing through the cut. All of them roughen the surface, none of them appear in a cusp calculation, and their combined size depends on the material and the setup.
Why the gap is guessed at
Shops that own a surface tester measure finished parts and keep the reading. It stays in that shop, attached to that job. Nobody is collecting the parameters alongside the measurement, so the correction between theory and reality has never been characterised.
What it would take
Not much. If you already measure, record what the calculator predicted, what the instrument read, and the conditions. One line per job. The ratio between the two numbers is the whole payload.
Send a measured Ra
Material and grade, nose radius or cutter, feed, speed and depth, the finish the tool predicted, and what your instrument actually read. Enough of those and the theoretical figure can carry a correction that reflects real machines instead of pure geometry — and the milling side can finally replace its assumed corner radius with something measured.
Contributors get the correction factors back as they emerge. Nothing is published with a shop name attached.
The same part, two setups, two answers
Catalog values assume a rigid machine, solid workholding and a clean continuous cut. Almost nothing on a real shop floor is all three at once, and the gap between the catalog and your setup is where inserts get destroyed.
Rigidity, in four places
The workpiece itself, how it is held, the tool and its overhang, and the machine. A thin-wall part in soft jaws on a worn machine with a long tool is four weak links, not one, and they compound rather than average out.
Interrupted cutting
An intermittent cut hammers the edge on every entry. Keyways, cross holes, flats, welded seams, anything out of round. It is a toughness problem, not a wear problem, and the answer is to back off speed rather than feed.
Skin and material condition
Cast skin, forge scale and flame-cut edges are abrasive and often carry embedded sand or oxide. The first pass through them is a different cut from every pass after it. Heat treatment and prior work hardening shift the picture again.
These are scored, not just noted. The rigidity factors combine into a figure that drives which surface finish grade is realistic to propose in the first place — a setup that cannot hold a fine finish should not be asked to. Interrupted cutting, skin and material condition then combine with it into a second figure that scales the cutting speed up or down from the catalog value. A favourable setup earns speed above the book figure; an unfavourable one has it taken away before you ever see a number.
Values you already have, in the form you already have them
The tool takes ISO designations wherever a standard exists, so nothing has to be converted, interpreted or looked up before you can start.
| ISO 513 | Workpiece and carbide grade classification | The P / M / K / N / S / H letter is already on the insert packaging and in every manufacturer catalog. Base cutting speed ranges are held per class, and grade application ranges use the same designations, so a grade you own maps directly onto a cut you're planning. |
| ISO 1302 | Surface roughness grade numbers | Finish targets are the preferred N4 to N12 steps, not arbitrary decimals. The drawing specifies a grade, you select that grade, and feed is back-solved from it — no step where someone types a number that isn't on any drawing. |
| ISO 1832 | Indexable insert designation | Nose radius is picked from the standard steps, addressed by the two-digit code stamped on the insert itself. Round insert diameters follow the same table, so what you read off the box is what you enter. |
| ISO 3685 ISO 8688 |
Tool life testing — referenced, not claimed | The two-speed test that back-solves your own Taylor n depends entirely on where you call the edge worn out. The figures below are offered as a starting point so the test is runnable today. For work that has to stand up — a published result, a supplier dispute, a dataset others will rely on — get the standard and follow it. ISO 3685 covers turning; ISO 8688 parts 1 and 2 cover face and end milling. |
Where a standard doesn't apply, the tool says so instead of pretending. ISO 513 classifies carbide, so it isn't stretched over HSS; plastics sit outside its scope entirely and are labelled that way rather than forced into the nearest letter. ISO 3685's own scope is turning steel and cast iron — beyond that, the method is being extended rather than followed.
The desktop version
The browser calculator covers milling. The full application is where the work happens.
| Operations | Turning, milling, drilling |
|---|---|
| Finish driven | Feed back-solved from target Ra in turning and milling, with ISO 1302 preferred grade steps and a real nose radius on the turning side |
| Insert grades | Sandvik reference with cross-references to Iscar, Kennametal, Walter, Seco, Tungaloy, Ceratizit, Mitsubishi, Sumitomo |
| Milling detail | Lead angles 65 / 42 / 25 / 19 / 10°, round insert diameters, high feed milling grade tables |
| Setup scoring | Piece, mount and tool robustness, machine rigidity, interrupted cutting, skin condition and material condition — scored and carried into both the finish grade and the cutting speed |
| Machine limits | Power and torque checked against the machine, with depth of cut solved back from available power |
| Built-in calculators | Insert set cost is built from insert price, usable edges, how many are replaced at once, body cost over its life, off-machine labour and spares — with the money-versus-time line held: presetting and gauging cost money but not spindle time, so they belong in Ct and never in tc. Machine cost per minute is built the same way from ownership, floor space, maintenance, power and the operator rate. |
| Units | Metric and Imperial throughout — m/min or SFM, mm/rev or IPR, mm or inch. Switching converts every entered value and every result rather than relabelling them, so a job set up in one system can be read in the other without retyping anything. |
| Runs on | Windows, standalone, no internet connection required |
Get the desktop application
Free download. Tell me where to send it and I'll also send the build notes when the physics model or the grade tables change.
The tool life study
Catalog speeds come from lab conditions. Your machine, your fixturing, your coolant and your bar stock are not those conditions — which is why the tool back-solves the Taylor exponent from your own cuts instead of assuming one.
Two speeds, far apart
Run the same job at two cutting speeds with everything else held constant, and record how long each edge lasts. The speeds need at least 20% separation — two points close together won't fit a stable slope, and the whole result rests on that slope.
One definition of worn out
Both runs have to end at the same criterion or the exponent is meaningless. As a starting point: average flank wear of 0.3 mm, or 0.6 mm where the wear is irregular. Crater depth is the alternative measure where cratering dominates.
Then the economics follow
From two points you get n and C, and from those the minimum-cost tool life, the maximum-production tool life, and the speeds that correspond to each. Both are computed. Which one you want depends on whether the machine is the bottleneck.
Those wear figures are a starting point to make the test runnable, not a substitute for the standard. If the result has to stand up to scrutiny, ISO 3685 defines the workpiece, the tooling, the cutting fluid, the equipment and the wear assessment properly, and it is the reference to work from. It is a paid document — that cost sits with whoever needs the rigour, not with the tool.
Build the dataset that doesn't exist yet
Every shop that runs this test learns something about one material and grade combination, and then keeps it. Pooled, those results would be the only tool life dataset measured on production machines rather than in a laboratory — and no tooling vendor is ever going to publish it. If you run the test, send the numbers. Contributors get the pooled result back.
This is an open call, not a live database. Register and you'll hear when there's enough data to be worth returning.
Built in a machine shop, not a marketing department
- Every build is stamped. The build identifier sits in the disclaimer tooltip, so you always know which version produced a given number.
- It warns rather than clamps. When a parameter set is outside a sensible range the tool says so and leaves the value as you set it. The decision stays with you.
- User results improve it. The finish model is theoretical, and the milling corner radius is a stated placeholder. If your results differ from the calculation, that's worth sending.