Development of a Milling Center for Large Steam Turbine Generator Coils
A generator coil is a long, curved stack of copper bars that has to be milled to a controlled profile after forming. This page explains the development of a milling center built for that job: how the four cutting heads work, why the control is an open CNC system, and where the process stops being economical. Written for engineers and buyers who specify large-part milling.
In this article
- 1
- 2
- 3
- 4
- 5
- 6
Why a generator coil needs its own milling center
A stator coil in a large steam turbine generator is not a small part. A single bar can run several meters in length, with a rectangular copper cross-section that has been bent into a Roebel or similar transposed form. After forming and insulation wrapping, the mating surfaces and the end profiles still need metal removed. The cut is shallow, often under 2 mm of stock, but it runs along a curved path and has to stay square to the bar axis over the full length.
That combination is what forces a dedicated machine. A general-purpose horizontal mill with a 1,000 mm table cannot reach the middle of a 4 m bar without repositioning, and every repositioning adds a setup error. The development of a milling center for this work therefore starts from the part envelope, not from a standard machine platform.
The tolerance target is modest compared with aerospace work. What matters is flatness along the length and repeatability from bar to bar, because the coils are stacked and then pressed. If bar 1 is 0.15 mm high at the midpoint and bar 20 is 0.15 mm low, the stack does not close evenly and the insulation is loaded unevenly.
Copper is also awkward to cut. It is soft, it conducts heat away from the shear zone quickly, and it smears instead of breaking a chip if the rake angle is wrong. Sharp, polished carbide with a high positive rake and a generous coolant flow is the usual answer.
Four milling heads on one workbench
The machine described here uses four milling heads, numbered 1# through 4#, mounted over a single long workbench. Each head carries two cutting blades, identified as disc A and disc C, with a stem B between them. The stem is driven by a servomotor that moves it horizontally for positioning. Splitting the cut across four heads is a cycle-time decision: the same pass removes stock at four points along the bar instead of one.
The workbench is the second half of the design. It has to hold a long copper bar without letting it sag, and copper sags under its own weight when the span is long. Support spacing, clamping force, and the flatness of the bed all feed directly into the profile tolerance. Clamps are usually soft jaws or low-pressure pads, because a copper bar marks easily.
Blade geometry matters more than spindle power here. The two discs cut on opposite sides of the stem, which balances the radial load and keeps the head from walking. If the discs are mismatched in diameter or wear unevenly, the head pulls to one side and the profile drifts over a long pass.
Heads 1# to 4# are positioned independently, so a program can use all four for a roughing pass and then a single head for a finishing pass. That is usually faster than trying to hold finish tolerance with all four engaged.
Open CNC control and what it buys you
The control system is an open architecture built around a PC and microprocessors rather than a closed proprietary controller. In practice this means the machine builder can write its own motion and profile routines, and the end user can change post-processing, add sensors, or log data without going back to the original vendor.
The cost argument is real. A closed controller with a long-axis, multi-head kinematic model is expensive to license and slow to modify. An open system lets the builder spread development cost and keep the software modular, so a change to one head's motion law does not force a rewrite of the whole program.
Access levels are part of the design. Operators get the run, jog, and offset screens. Process engineers get feed, speed, and compensation tables. Only the integrator or a trained maintenance lead gets the kinematic parameters and the servo tuning. That split prevents an accidental edit to a motion law that would scrap a batch of bars.
Open control is not free. The builder owns the safety logic, the look-ahead, and the interpolation quality. If the motion planner is weak, a long curved pass will show chord error at high feed, and no amount of servo tuning will hide it.
Keeping the machine running: the reliability measures
A coil line usually feeds an assembly station, so an unplanned stop on the mill backs up everything downstream. The reliability measures on this machine are aimed at that specific risk rather than at general uptime.
First, critical servo and spindle drives are duplicated or held as line-replaceable units on site. Second, the control logs axis following error and blade wear trend, so a drifting head is caught before it produces out-of-tolerance bars. Third, thermal behavior is monitored: copper conducts heat into the fixture, and a bed that grows 0.05 mm over a shift will show up in the profile.
In-process gauging is the practical backstop. Measuring the profile after a roughing pass, before the finishing pass, lets the control apply a compensation offset for the next bar instead of scrapping the current one.
None of this replaces a maintenance schedule. Blade change intervals, way lubrication, and a check of the head alignment relative to the bed are the three tasks that most often get skipped, and each one shows up as a profile error rather than as a machine fault.
Boundary conditions: where this process stops working
The design has clear edges. It is built for long, shallow, curved cuts in soft, conductive metal. Push it outside that envelope and it stops being the right tool.
Thick stock removal is the first limit. Taking 5 mm or more in one pass loads the blade and the fixture beyond what the head was sized for. That work belongs on a heavier mill with a single large cutter, even if it means more setups.
Very tight form tolerance on a short part is the second limit. If the feature is 200 mm long and needs ±0.01 mm, a small 5-axis or 3-axis machining center with a Ø400 mm rotary table will hold it with less fuss. Long-travel machines trade stiffness for reach.
Hard or abrasive materials are the third. This process is aimed at copper and copper alloys. Milling Inconel or a hardened tool steel on the same platform would need different blades, different feeds, and a much stiffer loop between the head and the bed.
Finally, volume. Below roughly a few dozen bars, the fixture design and program proving cost more than the cycle-time saving. A shop with a large traveling-column mill and a good fixture can handle a one-off coil.
When a dedicated coil milling center makes sense
Compare the part and volume against a general-purpose large mill.
| Criterion | Dedicated coil milling center | General-purpose large mill |
|---|---|---|
| Part length | 3,000–4,000 mm in one setup | Repositioning above ~1,500 mm |
| Cut depth | Shallow, under 2 mm stock | Any depth within spindle power |
| Heads engaged | Up to 4 simultaneously | One spindle, one tool |
| Setup count per bar | One | Two or more |
| Bar-to-bar repeatability | High, same fixture every cycle | Depends on operator reset |
| Part volume | Batch production, hundreds of bars | One-offs and repair work |
| Control changes | Open, user-editable post | Vendor-locked controller |
| Best fit | Stator coil production line | Mixed large-part machine shop |
The verdict
If you are milling hundreds of long copper coils to a flatness profile, build or buy the dedicated four-head center. If you are cutting short features to tight form tolerance, or removing thick stock, use a standard machining center instead.
Questions engineers ask next
Why four milling heads instead of one large cutter?
Cycle time is the main reason. Four heads cut four points along the bar in the same pass, so the table does not have to travel the full length four times.
Balance is the second reason. Two discs per head cut on opposite sides of the stem, which cancels much of the radial load and keeps the head from walking on a long pass.
What tolerance can this kind of machine realistically hold?
For long copper bars, flatness and profile repeatability in the 0.05–0.15 mm range over the full length is realistic, with tighter local control at each head.
If your drawing calls for ±0.005 mm on a short, rigid feature, that is standard machining-center territory, not long-travel coil milling. Match the machine to the feature length.
How is the copper kept from moving during the cut?
Support spacing is set so the bar cannot sag between supports, and clamping uses soft jaws or low-pressure pads. Copper marks and deforms easily, so clamp force is kept just high enough to stop motion.
The fixture is the part of the design most often underestimated. A stiff bed with poor support still produces a bowed bar.
Is an open CNC control a maintenance problem?
It shifts responsibility to the integrator and the plant. You own the safety logic, the post-processor, and the tuning records.
The payoff is that you can add sensors, change motion parameters, and log process data without waiting on a vendor. Plants with in-house controls staff do well with this. Plants without that staff should negotiate a support agreement up front.
Can the same machine cut the end profiles as well as the bar sides?
Yes, if the head travel and the fixture can present the end face. Head positioning is independent, so a program can use one head for the end profile and park the others.
Check the head reach and the blade clearance against your actual coil radius before assuming it fits. Curved ends are where interference shows up first.
When is it cheaper to outsource the coil milling instead?
Below a few dozen bars, outsourcing usually wins. The fixture, the program, and the first-article proving are fixed costs that do not scale down.
Once the geometry repeats and the batch is large, in-house cycle time is what pays back the machine. Run the break-even on setup hours, not on spindle hours.
Send us the coil drawing
Upload a STEP file and we will return a quotation with a DFM analysis within 12 hours.
12-hour quote100% inspection±0.005 mmNo minimum order quantity
More from the shop floor
We publish setup notes, tooling trials and inspection data from the factory floor.
